# Azure AI Foundry Documentation - Full Content ## What is Microsoft Foundry (new)? ### What is Microsoft Foundry (new)? URL: https://learn.microsoft.com/en-us/azure/ai-foundry/what-is-foundry?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . What is Microsoft Foundry? Feedback Summarize this article for me In this article Note This document refers to the Microsoft Foundry (classic) portal. 🔄 Switch to the Microsoft Foundry (new) documentation if you're using the new portal. Note This document refers to the Microsoft Foundry (new) portal. Microsoft Foundry is a unified Azure platform-as-a-service offering for enterprise AI operations, model builders, and application development. This foundation combines production-grade infrastructure with friendly interfaces, enabling developers to focus on building applications rather than managing infrastructure. Microsoft Foundry unifies agents, models, and tools under a single management grouping with built-in enterprise-readiness capabilities including tracing, monitoring, evaluations, and customizable enterprise setup configurations. The platform provides streamlined management through unified Role-based access control (RBAC), networking, and policies under one Azure resource provider namespace. Tip Azure AI Foundry is now Microsoft Foundry. Screenshots appearing throughout this documentation are in the process of being updated. Tip Azure AI Foundry is now Microsoft Foundry. Screenshots appearing throughout this documentation are in the process of being updated. Microsoft Foundry portals There are two different portals for you to use to interact with Microsoft Foundry. A toggle in the portal banner allows you to switch between the two versions. Portal Banner display When to use Microsoft Foundry (classic) Choose this portal when working with multiple resource types: Azure OpenAI, Foundry resources, hub-based projects, or Found --- ## Quickstart: Create resources ### Quickstart: Create resources URL: https://learn.microsoft.com/en-us/azure/ai-foundry/tutorials/quickstart-create-foundry-resources?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Quickstart: Set up Microsoft Foundry resources Feedback Summarize this article for me In this article In this quickstart, you create a Microsoft Foundry project and deploy a model. If you're managing a team, you also grant access to team members. After you complete these steps, you or your team can start building AI applications using the deployed model. Tip This quickstart shows you how to create resources to build an agent with a basic setup. For more advanced scenarios that use your own resources, see Set up your environment for agent development . Prerequisites An Azure account with an active subscription. If you don't have one, create a free Azure account, which includes a free trial subscription . If you're creating the project for yourself: Access to a role that allows you to create a Foundry resource, such as Azure Account AI Owner or Azure AI Owner on the subscription or resource group. For more information about permissions, see Role-based access control for Microsoft Foundry . If you're creating the project for a team: Access to a role that allows you to complete role assignments, such as Owner . For more information about permissions, see Role-based access control for Microsoft Foundry . A list of user email addresses or Microsoft Entra security group IDs for team members who need access. Select your preferred method by using the following tabs: Azure CLI Foundry portal Install the Azure CLI . Sign in to Azure: az login Access to the Microsoft Foundry portal . Create a project Create a Foundry project to organize your work. The project contains models, agents, and other resources your team uses. Azure CLI Foundry por --- ## Quickstart: Chat with an agent ### Quickstart: Chat with an agent URL: https://learn.microsoft.com/en-us/azure/ai-foundry/quickstarts/get-started-code?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Microsoft Foundry quickstart Feedback Summarize this article for me In this article Note This document refers to the Microsoft Foundry (classic) portal. 🔄 Switch to the Microsoft Foundry (new) documentation if you're using the new portal. Note This document refers to the Microsoft Foundry (new) portal. In this quickstart, you use Microsoft Foundry to interact with a Foundry model, create, and chat with an agent. Prerequisites A model deployed in Microsoft Foundry. If you don't have a model, first complete Quickstart: Set up Microsoft Foundry resources . The required language runtimes, global tools, and Visual Studio Code extensions as described in Prepare your development environment . Set environment variables and get the code Store your project endpoint as an environment variable. Also set these values for use in your scripts. PROJECT_ENDPOINT= AGENT_NAME="MyAgent" MODEL_DEPLOYMENT_NAME="gpt-4.1-mini" Python C# TypeScript Java REST API Foundry portal Follow along below or get the code: Get the code Sign in using the CLI az login command to authenticate before running your Python scripts. Follow along below or get the code: Get the code Sign in using the CLI az login command to authenticate before running your C# scripts. Follow along below or get the code: Get the code Sign in using the CLI az login command to authenticate before running your TypeScript scripts. Follow along below or get the code: Get the code Sign in using the CLI az login command to authenticate before running your Java scripts. Follow along below or get the code: Get the code . Sign in using the CLI az login command to a --- ## Tutorial: Idea to prototype ### Tutorial: Idea to prototype URL: https://learn.microsoft.com/en-us/azure/ai-foundry/tutorials/developer-journey-idea-to-prototype?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Tutorial: Idea to prototype - Build and evaluate an enterprise agent Feedback Summarize this article for me In this article This tutorial covers the first stage of the Microsoft Foundry developer journey: from an initial idea to a working prototype. You build a modern workplace assistant that combines internal company knowledge with external technical guidance by using the Microsoft Foundry SDK. Business scenario : Create an AI assistant that helps employees by combining: Company policies (from SharePoint documents) Technical implementation guidance (from Microsoft Learn via MCP) Complete solutions (combining both sources for business implementation) Batch evaluation to validate agent performance on realistic business scenarios Tutorial outcome : By the end you have a running Modern Workplace Assistant that can answer policy, technical, and combined implementation questions; a repeatable batch evaluation script; and clear extension points (other tools, multi‑agent patterns, richer evaluation). You will: Build a Modern Workplace Assistant with SharePoint and MCP integration. Demonstrate real business scenarios combining internal and external knowledge. Implement robust error handling and graceful degradation. Create evaluation framework for business-focused testing. Prepare foundation for governance and production deployment. This minimal sample demonstrates enterprise-ready patterns with realistic business scenarios. Important Code in this article uses packages that are currently in preview. This preview is provided without a service-level agreement, and we don't recommend it for production workloads. Certain features might not --- ## Ask AI ### Ask AI URL: https://learn.microsoft.com/en-us/azure/ai-foundry/concepts/ask-ai?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Ask AI for help (preview) Feedback Summarize this article for me In this article Note This document refers to the Microsoft Foundry (classic) portal. 🔄 Switch to the Microsoft Foundry (new) documentation if you're using the new portal. Note This document refers to the Microsoft Foundry (new) portal. You can ask AI to assist you in Microsoft Foundry. To start using AI to ask questions, select the AI icon in the upper right of the Microsoft Foundry portal. A chat window opens where you can type your questions and receive answers in real-time. Important Items marked (preview) in this article are currently in public preview. This preview is provided without a service-level agreement, and we don't recommend it for production workloads. Certain features might not be supported or might have constrained capabilities. For more information, see Supplemental Terms of Use for Microsoft Azure Previews . Prerequisites Access to the Foundry portal . Capabilities What This AI Can Do - The Ask AI experience is designed to provide assistance by answering questions based on: Documentation : This documentation includes details about Foundry such as Quickstarts, How-tos, or reference documentation of the Microsoft Foundry SDK. The agent can help you navigate the documentation or find answers for you. Model Catalog : Provide information about specific models in the Foundry model catalog, including their capabilities and features. Troubleshooting : Help diagnose and resolve common Foundry problems by searching the troubleshooting knowledge base and providing step-by-step solutions. What This AI Cannot Do - While the agent is a powerful tool, it has so --- ## Agent development overview ### Agent development overview URL: https://learn.microsoft.com/en-us/azure/ai-foundry/agents/overview?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . What is Foundry Agent Service? Feedback Summarize this article for me In this article Note This document refers to the Microsoft Foundry (classic) portal. 🔄 Switch to the Microsoft Foundry (new) documentation if you're using the new portal. Note This document refers to the Microsoft Foundry (new) portal. Most businesses don't want just chatbots. They want automation that's faster and has fewer errors. That desire might mean summarizing documents, processing invoices, managing support tickets, or publishing blog posts. In all cases, the goal is the same: freeing people and resources to focus on higher-value work by offloading repetitive and predictable tasks. Large language models (LLMs) introduce a new type of automation with systems that can understand unstructured data, make decisions, and generate content. In practice, businesses can have difficulty moving beyond demos and into production. LLMs can drift, be incorrect, and lack accountability. Without visibility, policy enforcement, and orchestration, these models are hard to trust in real business workflows. Microsoft Foundry is designed to change that. It's a platform that combines models, tools, frameworks, and governance into a unified system for building intelligent agents. At the center of this system is Foundry Agent Service , which enables the operation of agents across development, deployment, and production. Agent Service connects the core pieces of Foundry, such as models, tools, and frameworks, into a single runtime. It manages conversations, orchestrates tool calls, enforces content safety, and integrates with identity, networking, and observability systems. Thes --- ## FAQ ### FAQ URL: https://learn.microsoft.com/en-us/azure/ai-foundry/agents/faq?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Foundry Agent Service frequently asked questions Feedback Summarize this article for me In this article Find answers to common questions about Foundry Agent Service. If you can't find answers to your questions in this article and you still need help, see Foundry Tools support and help options . Foundry Agent Service is part of Foundry Tools. For getting started, see: What is Foundry Agent Service? Set up your environment Quotas and limits Setup and access What's the difference between basic setup and standard setup? Basic setup stores agent state in Microsoft-managed resources. Standard setup stores agent data (threads, files, and vector stores) in your Azure resources that you connect through capability hosts. To compare setup options and choose the right one, see Set up your environment and Capability hosts . What permissions (RBAC roles) do I need? Role requirements depend on what you're doing. Common roles include: Azure AI Account Owner : Create accounts and projects Azure AI User : Create and edit agents Role Based Access Control Administrator (or Owner ): Required for standard setup to assign roles to connected resources For standard setup, you also need the Microsoft.Authorization/roleAssignments/write permission. For complete role requirements, see Required permissions and Role-based access control (RBAC) in Microsoft Foundry . Where should I start if I'm new to Foundry Agent Service? Start with Set up your environment , then create your first agent. Classic experience: Quickstart: Create a new agent New Foundry experience: Quickstart: Get started with agents in code General Do you store any data used in the Foundry Age --- ### FAQ URL: https://learn.microsoft.com/en-us/azure/ai-foundry/agents/concepts/foundry-iq-faq?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Foundry IQ frequently asked questions Feedback Summarize this article for me In this article Find answers to common questions about Foundry IQ . General What is Foundry IQ? Foundry IQ enables agents to access, process, and act on knowledge from anywhere. With Foundry IQ, you create a knowledge base that connects to one or more knowledge sources . The agentic retrieval engine processes queries, and an optional large language model (LLM) from Azure OpenAI in Foundry Models adds query planning and reasoning. Agents built in Foundry Agent Service call the knowledge base to retrieve relevant content. What is the difference between Foundry IQ and agentic retrieval? Foundry IQ consists of knowledge bases, knowledge sources, and native integrations with Azure OpenAI in Foundry Models and Foundry Agent Service. The Microsoft Foundry (new) portal offers a streamlined, end-to-end setup experience, but you can also create the Foundry IQ components programmatically. Agentic retrieval is the multi-query retrieval engine that powers Foundry IQ knowledge bases. For custom solutions, you can use agentic retrieval directly via the Azure AI Search APIs. How is Foundry IQ different from existing RAG patterns or Azure OpenAI On Your Data? One Foundry IQ knowledge base provides access to multiple sources, removing the need to connect each agent to each source individually. The agentic retrieval engine plans which sources to query and performs iterative search if initial results don't meet relevance standards. Indexing and data synchronization are triggered automatically. Iterative search depends on specifying a medium retrieval reasoning effort in th --- ## Quotas, limits, models and region support ### Quotas, limits, models and region support URL: https://learn.microsoft.com/en-us/azure/ai-foundry/agents/concepts/limits-quotas-regions?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Quotas, limits, models, and regional support Feedback Summarize this article for me In this article This article describes the quotas, limits, and regional availability for Foundry Agent Service. Supported regions Foundry Agent Service is available in the following Azure regions: Australia East Brazil South Canada East East US East US 2 France Central Germany West Central Italy North Japan East Norway East South Africa North South Central US South India Sweden Central Switzerland North UK South West Europe West US West US 3 Azure OpenAI model support Foundry Agent Service is compatible with current Azure OpenAI models. For a complete list of supported models and their availability by region, see Foundry Models sold directly by Azure . Other model collections The following Foundry models are available for your agents to use. Models sold directly by Azure: MAI-DS-R1 : Deterministic, precision-focused reasoning. grok-4 : Frontier-scale reasoning for complex, multiple-step problem solving. grok-4-fast-reasoning : Accelerated agentic reasoning optimized for workflow automation. grok-4-fast-non-reasoning : High-throughput, low-latency generation and system routing. grok-3 : Strong reasoning for complex, system-level workflows. grok-3-mini : Lightweight model optimized for interactive, high-volume use cases. Llama-3.3-70B-Instruct : Versatile model for enterprise Q&A, decision support, and system orchestration. Llama-4-Maverick-17B-128E-Instruct-FP8 : FP8-optimized model that delivers fast, cost-efficient inference. DeepSeek-V3-0324 : Multimodal understanding across text and images. DeepSeek-V3.1 : Enhanced multimodal reasoning and gro --- ## Development lifecycle ### Development lifecycle URL: https://learn.microsoft.com/en-us/azure/ai-foundry/agents/concepts/development-lifecycle?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Agent development lifecycle Feedback Summarize this article for me In this article The agent development lifecycle in Microsoft Foundry spans from initial creation through production monitoring. Following this lifecycle helps you build reliable agents, catch issues early, and ship with confidence. Use the Foundry portal or code to build, customize, and test your agent's behavior. Then iterate with tracing, evaluation, and monitoring to improve quality and reliability. When you're ready, publish your agent as an agent application to share it and integrate it into your apps. This article is for developers who want to build, test, and ship production-ready agents. Prerequisites A Microsoft Foundry project Familiarity with the Agents playground For code development: Familiarity with the development environment setup Lifecycle at a glance Use this lifecycle as a practical checklist while you build and ship an agent. Choose an agent type : Start with a prompt-based agent, a workflow, or a hosted agent. Create your agent and start testing : Iterate in the playground or in code. Add tools and data : Attach tools for retrieval and actions, and validate the configuration before you save. Save changes as versions : Capture meaningful milestones and compare versions. Debug with tracing : Use tracing to confirm tool calls, latency, and end-to-end behavior. For details, see Agent tracing overview . Evaluate quality and safety : Run repeatable evaluations to catch regressions before publishing. For conceptual guidance, see Agent evaluators . Publish and integrate : Publish a stable endpoint and integrate it into your application. For steps, se --- ## Agent runtime components ### Agent runtime components URL: https://learn.microsoft.com/en-us/azure/ai-foundry/agents/concepts/runtime-components?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Agent runtime components Feedback Summarize this article for me In this article Agent runtime components are the core objects—agents, conversations, and responses—that power stateful, multi-turn interactions in Microsoft Foundry Agent Service. Together, these components let you generate outputs, persist state across turns, and build conversational applications. This article explains the roles of an agent , conversation , and response , and how they work together during response generation. Prerequisites A Microsoft Foundry project . Familiarity with the agent development lifecycle (optional). How runtime components work together When you work with an agent, you follow a consistent pattern: Create an agent : Define an agent to start sending messages and receiving responses. Create a conversation (optional) : Use a conversation to maintain history across turns. If you don't use a conversation, carry forward context by using the output from a previous response. Generate a response : The agent processes input items in the conversation and any instructions provided in the request. The agent might append items to the conversation. Check response status : Monitor the response until it finishes (especially in streaming or background mode). Retrieve the response : Display the generated response to the user. The following diagram illustrates how these components interact in a typical agent loop. You provide user input (and optionally conversation history), the service generates a response (including tool calls when configured), and the resulting items can be reused as context for the next turn. What is an agent? An agent is a persisted or --- ## Agent identity ### Agent identity URL: https://learn.microsoft.com/en-us/azure/ai-foundry/agents/concepts/agent-identity?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Agent identity concepts in Microsoft Foundry Feedback Summarize this article for me In this article An agent identity is a specialized identity type in Microsoft Entra ID that's designed specifically for AI agents. It provides a standardized framework for governing, authenticating, and authorizing AI agents across Microsoft services. This framework enables agents to securely access resources, interact with users, and communicate with other systems. Microsoft Foundry automatically provisions and manages agent identities throughout the agent lifecycle. This integration simplifies permission management while maintaining security and auditability as agents move from development to production. This article explains how agent identities relate to Microsoft Entra ID objects, how Foundry uses them when an agent calls tools, and how to apply least-privilege access with Azure role-based access control (RBAC). Prerequisites Understanding of Microsoft Entra ID and OAuth authentication Familiarity with Azure role-based access control (RBAC) Basic knowledge of AI agents and their runtime requirements For Foundry-specific RBAC roles and scopes, see Azure role-based access control in Foundry . How agent identities work in Foundry Foundry uses Microsoft Entra ID agent identities to support two related needs: Management and governance : Give administrators a consistent way to inventory agents, apply policies, and audit activity. Tool authentication : Let an agent authenticate to downstream systems (for example, Azure Storage) without embedding secrets in prompts, code, or connection strings. At a high level, Foundry does the following: Provisions --- ## Hosted agents ### Hosted agents URL: https://learn.microsoft.com/en-us/azure/ai-foundry/agents/concepts/hosted-agents?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . What are hosted agents? Feedback Summarize this article for me In this article When you build agentic applications by using open-source frameworks, you typically manage containerization, web server setup, security integration, memory persistence, infrastructure scaling, data transmission, instrumentation, and version rollbacks. These tasks become even more challenging in heterogeneous cloud environments. Important Hosted agents are currently in public preview . See Limits, pricing, and availability (preview) for current constraints. Hosted agents in Foundry Agent Service solve these challenges for Microsoft Foundry users. By using this managed platform, you can deploy and operate AI agents securely and at scale. You can use your custom agent code or a preferred agent framework with streamlined deployment and management. Prerequisites A Microsoft Foundry project Basic understanding of containerization and Docker Familiarity with Azure Container Registry Knowledge of your preferred agent framework (LangGraph, Microsoft Agent Framework, or custom code) At a glance Hosted agents let you bring your own agent code and run it as a managed containerized service. Use this article to: Understand what hosted agents are and when to use them. Package and test your agent locally before deployment. Create, manage, publish, and monitor hosted agents. If you want to jump to a task, see: Package code and test locally Create a hosted agent Manage hosted agents Publish hosted agents to channels Troubleshoot hosted agent endpoints Limits, pricing, and availability (preview) Hosted agents are currently in preview. Private networking support : You can --- ## Capability hosts ### Capability hosts URL: https://learn.microsoft.com/en-us/azure/ai-foundry/agents/concepts/capability-hosts?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Capability hosts Feedback Summarize this article for me In this article Note This document refers to the Microsoft Foundry (classic) portal. 🔄 Switch to the Microsoft Foundry (new) documentation if you're using the new portal. Note This document refers to the Microsoft Foundry (new) portal. Note Updating capability hosts is not supported. To modify a capability host, you must delete the existing one and recreate it with the new configuration. Capability hosts are sub-resources that you configure at both the Microsoft Foundry account and Foundry project scopes. They tell Foundry Agent Service where to store and process agent data, including: Conversation history (threads) File uploads Vector stores Prerequisites A Microsoft Foundry project If you use your own resources for agent data (standard agent setup), create the required Azure resources and connections: Use your own resources Add a new connection to your project Required permissions: Contributor role on the Foundry account to create capability hosts User Access Administrator or Owner role to assign access to Azure resources (for standard agent setup) For details, see Required permissions and Role-based access control (RBAC) in Microsoft Foundry . Why use capability hosts? Capability hosts let you bring your own Azure resources instead of using the default Microsoft-managed platform resources. This gives you: Data sovereignty - Keep all agent data within your Azure subscription. Security control - Use your own storage accounts, databases, and search services. Compliance - Meet specific regulatory or organizational requirements. How do capability hosts work? Creating capabili --- ## Create a workflow ### Create a workflow URL: https://learn.microsoft.com/en-us/azure/ai-foundry/agents/concepts/workflow?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Build a workflow in Microsoft Foundry Feedback Summarize this article for me In this article Workflows are UI-based tools in Microsoft Foundry. Use them to create declarative, predefined sequences of actions that orchestrate agents and business logic in a visual builder. Workflows enable you to build intelligent automation systems that seamlessly blend AI agents with business processes in a visual manner. Traditional single-agent systems are limited in their ability to handle complex, multifaceted tasks. By orchestrating multiple agents, each with specialized skills or roles, you can create systems that are more robust, adaptive, and capable of solving real-world problems collaboratively. Prerequisites An Azure account with an active subscription. If you don't have one, create a free Azure account, which includes a free trial subscription . A project in Microsoft Foundry. For more information, see Create projects . Access to create and run workflows in your Foundry project. For more information, see Azure role-based access control (RBAC) in Foundry . Decide when to use workflows Workflows are ideal for scenarios where you need to: Orchestrate multiple agents in a repeatable process. Add branching logic (for example, if/else) and variable handling without writing code. Create human-in-the-loop steps (for example, approvals or clarifying questions). If you want to edit workflow YAML in Visual Studio Code or run workflows in a local playground, see: Work with Declarative (Low-code) Agent workflows in Visual Studio Code Work with Hosted (Pro-code) Agent workflows in Visual Studio Code Understand workflow patterns Foundry provides te --- ## Quickstart - Deploy your first hosted agent ### Quickstart - Deploy your first hosted agent URL: https://learn.microsoft.com/en-us/azure/ai-foundry/agents/quickstarts/quickstart-hosted-agent?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Quickstart: Deploy your first hosted agent using Azure Developer CLI Feedback Summarize this article for me In this article In this quickstart, you deploy a containerized AI agent with Foundry tools to Foundry Agent Service. The sample agent uses web search and optionally MCP tools to answer questions. By the end, you have a running hosted agent that you can interact with through the Foundry playground. In this quickstart, you: Set up an agent sample project with Foundry tools Test the agent locally Deploy to Foundry Agent Service Interact with your agent in the playground Clean up resources Prerequisites Before you begin, you need: An Azure subscription - Create one for free A Microsoft Foundry project with: An Azure OpenAI model deployment (for example gpt-5 ) This example uses gpt-5 , you may need to use another model (such as gpt-4.1 ) depending on your quotas and limits . (Optional) An MCP tool , if you have one you want to use. An Azure OpenAI resource Azure Developer CLI version 1.23.0 or later (Optional) Azure CLI version 2.80 or later Docker Desktop installed and running Python 3.10 or later Note Hosted agents are currently in preview. Step 1: Set up the sample project Initialize a new project with the Foundry starter template and configure it with the agent-with-foundry-tools sample. Initialize the starter template: azd init -t https://github.com/Azure-Samples/azd-ai-starter-basic This interactive command prompts you for an environment name (for example, my-hosted-agent ). The environment name determines your resource group name ( rg-my-hosted-agent ). Note If a resource group with the same name already exists, azd pro --- ## Deploy a hosted agent ### Deploy a hosted agent URL: https://learn.microsoft.com/en-us/azure/ai-foundry/agents/how-to/deploy-hosted-agent?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Deploy a hosted agent Feedback Summarize this article for me In this article This article shows you how to deploy a containerized agent to Foundry Agent Service. Use hosted agents when you need to run custom agent code built with frameworks like LangGraph, Microsoft Agent Framework, or your own implementation. Prerequisites A Microsoft Foundry project Python 3.10 or later for SDK-based development Azure CLI version 2.80 or later Docker Desktop installed for local container development Familiarity with Azure Container Registry Agent code using a supported framework Required permissions You need one of the following role combinations depending on your deployment scenario: Scenario Required roles Create new Foundry project Azure AI Owner on Foundry resource Deploy to existing project with new resources Azure AI Owner on Foundry + Contributor on subscription Deploy to fully configured project Reader on account + Azure AI User on project For more information, see Authentication and authorization . Package and test your agent locally Before deploying to Foundry, validate your agent works locally using the hosting adapter. Run your agent locally The hosting adapter starts a local web server that exposes your agent as a REST API: @baseUrl = http://localhost:8088 POST {{baseUrl}}/responses Content-Type: application/json. { "input": { "messages": [ { "role": "user", "content": "Where is Seattle?" } ] } } A successful response: { "id": "resp_abc123", "object": "response", "output": [ { "type": "message", "role": "assistant", "content": "Seattle is a major city in the Pacific Northwest region of the United States..." } ], "status": "complet --- ## Manage hosted agent lifecycle ### Manage hosted agent lifecycle URL: https://learn.microsoft.com/en-us/azure/ai-foundry/agents/how-to/manage-hosted-agent?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Manage hosted agent lifecycle Feedback Summarize this article for me In this article This article shows you how to manage hosted agent deployments in Foundry Agent Service. After you deploy a hosted agent, you can start, stop, update, and delete it as your needs change. Prerequisites A deployed hosted agent Azure CLI version 2.80 or later Azure Cognitive Services CLI extension: az extension add --name cognitiveservices --upgrade Start an agent deployment Start a hosted agent to make it available for requests. Use this command to start a new deployment or restart a stopped agent. az cognitiveservices agent start \ --account-name myAccount \ --project-name myProject \ --name myAgent \ --agent-version 1 \ --min-replicas 1 \ --max-replicas 2 Argument Required Description --account-name -a Yes Microsoft Foundry account name --project-name Yes AI project name --name -n Yes Hosted agent name --agent-version Yes Agent version to start --min-replicas No Minimum replicas (default: 1) --max-replicas No Maximum replicas (default: 1) State transitions when starting: Stopped → Starting → Started (success) or Failed (error) Stop an agent deployment Stop a running agent to pause processing and reduce costs. The agent version remains available for restarting later. az cognitiveservices agent stop \ --account-name myAccount \ --project-name myProject \ --name myAgent \ --agent-version 1 Argument Required Description --account-name -a Yes Microsoft Foundry account name --project-name Yes AI project name --name -n Yes Hosted agent name --agent-version Yes Agent version to stop State transitions when stopping: Running → Stopping → Stopped (success) --- ## System message design ### System message design URL: https://learn.microsoft.com/en-us/azure/ai-foundry/openai/concepts/advanced-prompt-engineering?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . System message design Feedback Summarize this article for me In this article System messages help you steer an Azure OpenAI chat model toward the behavior, tone, and output format you want. This article explains what system messages are, how they affect responses, and how to design them for consistency and safety. What this article covers This article focuses on the system message (sometimes called a system prompt or metaprompt ) for chat-based experiences. If you want broader prompt guidance (few-shot examples, ordering, and token efficiency), see Prompt engineering techniques . Prerequisites To use system messages, you need access to an Azure OpenAI resource with a chat completion model deployment. For setup instructions, see Create and deploy an Azure OpenAI resource . What is a system message? A system message is a set of instructions and context you provide to the model to guide its responses. You typically use it to: Define the assistant’s role and boundaries. Set tone and communication style. Specify output formats (for example, JSON). Add safety and quality constraints for your scenario. A system message can be one short sentence: You are a helpful AI assistant. Or it can be multiple lines with structured rules and formatting requirements. Important A system message influences the model, but it doesn’t guarantee compliance. You still need to test and iterate, and you should layer system messages with other mitigations (for example, filtering and evaluation). How system messages work In chat-based APIs, you send a set of messages that include roles such as system , user , and assistant . The system message typically appea --- ## Safety system messages ### Safety system messages URL: https://learn.microsoft.com/en-us/azure/ai-foundry/openai/concepts/system-message?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Safety system messages Feedback Summarize this article for me In this article Safety system messages help you guide an Azure OpenAI model’s behavior, improve response quality, and reduce the likelihood of harmful outputs. They work best as one layer in a broader safety strategy. Note This article uses "system message" interchangeably with "metaprompt" and "system prompt." Here, we use "system message" to align with common terminology. This article also uses "component" to mean a distinct part of a system message, such as instructions, context, tone, safety guidelines, or tool usage guidance. What is a system message? A system message is a set of high-priority instructions and context that you send to a chat model to steer how it responds. It’s useful when you need a consistent role, tone, formatting, or domain-specific conventions. What is a safety system message? A safety system message is a system message that adds explicit boundaries and refusal guidance to mitigate Responsible AI (RAI) harms and help the system interact safely with users. Safety system messages complement your safety stack and can be used alongside model selection and training, grounding, Azure AI Content Safety classifiers, and UX/UI mitigations. Learn more about Responsible AI practices for Azure OpenAI models . Key components of a system message Most system messages combine multiple components: Role and task : What the assistant is and what it’s responsible for. Audience and tone : Who the response is for, and the expected voice. Scope and boundaries : What the assistant must not do, and what to do when it can’t comply. Safety guidelines : Rules that redu --- ### Safety system messages URL: https://learn.microsoft.com/en-us/azure/ai-foundry/openai/concepts/safety-system-message-templates?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Safety system message templates Feedback Summarize this article for me In this article Note This document refers to the Microsoft Foundry (classic) portal. 🔄 Switch to the Microsoft Foundry (new) documentation if you're using the new portal. Note This document refers to the Microsoft Foundry (new) portal. This article contains recommended safety system messages for your generative AI systems to help reduce the propensity of harm in various concern areas. Before you begin evaluating and integrating your safety system messages, visit the Safety system message conceptual guide to get started. Note Using a safety system message is one of many techniques you can use to mitigate risks in AI systems. It’s different from the Azure AI Content Safety service. How to use these templates Use these templates as a starting point. They’re intentionally generic so you can adapt them for your scenario. Start small and iterate. Add one component at a time, then test. Replace bracketed placeholders. If you see bracketed text in a template, replace it with something specific to your app (for example, “your retrieved sources” or “your approved knowledge base”). Avoid conflicting instructions. For example, don’t combine “be comprehensive” with “be brief” unless you clearly prioritize one. Tell the model what to do when it can’t comply. Clear refusal and fallback behavior helps reduce unsafe completions. Where to put the text In Foundry portal : Paste these components into your Safety system message field (or your System message field), then test in the playground. In your app : Put the combined text into the highest-priority instruction you send to t --- ## Safety system message templates ### Safety system message templates URL: https://learn.microsoft.com/en-us/azure/ai-foundry/openai/concepts/safety-system-message-templates?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Safety system message templates Feedback Summarize this article for me In this article Note This document refers to the Microsoft Foundry (classic) portal. 🔄 Switch to the Microsoft Foundry (new) documentation if you're using the new portal. Note This document refers to the Microsoft Foundry (new) portal. This article contains recommended safety system messages for your generative AI systems to help reduce the propensity of harm in various concern areas. Before you begin evaluating and integrating your safety system messages, visit the Safety system message conceptual guide to get started. Note Using a safety system message is one of many techniques you can use to mitigate risks in AI systems. It’s different from the Azure AI Content Safety service. How to use these templates Use these templates as a starting point. They’re intentionally generic so you can adapt them for your scenario. Start small and iterate. Add one component at a time, then test. Replace bracketed placeholders. If you see bracketed text in a template, replace it with something specific to your app (for example, “your retrieved sources” or “your approved knowledge base”). Avoid conflicting instructions. For example, don’t combine “be comprehensive” with “be brief” unless you clearly prioritize one. Tell the model what to do when it can’t comply. Clear refusal and fallback behavior helps reduce unsafe completions. Where to put the text In Foundry portal : Paste these components into your Safety system message field (or your System message field), then test in the playground. In your app : Put the combined text into the highest-priority instruction you send to t --- ## Publish and share agents ### Publish and share agents URL: https://learn.microsoft.com/en-us/azure/ai-foundry/agents/how-to/publish-agent?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Publish and share agents in Microsoft Foundry Feedback Summarize this article for me In this article Publishing promotes an agent from a development asset into a managed Azure resource with a dedicated endpoint, independent identity, and governance capabilities. This article shows you how to publish an agent, configure its authentication and permissions, update published versions, and consume the agent through its stable endpoint. When you publish an agent, Microsoft Foundry creates an Agent Application resource with a dedicated invocation URL and its own Microsoft Entra agent identity blueprint and agent identity. A deployment is created under the application that references your agent version and registers it in the Entra Agent Registry for discovery and governance. Publishing enables you to share agents with teammates, your organization, or customers without granting access to your Foundry project or source code. The stable endpoint remains consistent as you iterate and deploy new agent versions. Prerequisites A Foundry project with at least one agent version created Azure AI Project Manager role on the Foundry project scope to publish agents Azure AI User role on the Agent Application scope to chat with a published agent Familiarity with Azure role-based access control (RBAC) for permission configuration Familiarity with Agent identity concepts in Foundry Install the required language runtimes, global tools, and VS Code extensions as described in Prepare your development environment Important Code in this article uses packages that are currently in preview. This preview is provided without a service-level agreement, and we d --- ## Publish to Microsoft 365 Copilot and Teams ### Publish to Microsoft 365 Copilot and Teams URL: https://learn.microsoft.com/en-us/azure/ai-foundry/agents/how-to/publish-copilot?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Publish agents to Microsoft 365 Copilot and Microsoft Teams Feedback Summarize this article for me In this article Use this article to publish a Microsoft Foundry agent so people can use it in Microsoft 365 Copilot and Microsoft Teams. Publishing creates an agent application with a stable endpoint and then prepares a Microsoft 365 publishing package for testing and distribution. Prerequisites Access to the Microsoft Foundry portal A Foundry project with an agent version you tested and want to publish The following role assignments: Azure AI Project Manager role on the Foundry project scope to publish agents Azure AI User role on the Agent Application scope to invoke or chat with published agents For details, see Role-based access control in the Foundry portal . An Azure subscription where you can create Azure Bot Service resources and Microsoft Entra ID app registrations Permissions to register applications in Microsoft Entra ID (for the automatic app registration) Before you begin Test your agent thoroughly in the Foundry portal before publishing. Confirm it responds correctly and any tools work as expected. If your agent uses tools that access Azure resources, plan to reassign any required permissions after publishing. A published agent application uses its own agent identity separate from your project identity. For details, see Agent identity concepts in Microsoft Foundry and Publish and share agents in Microsoft Foundry . Decide whether you want Shared scope or Organization scope for distribution: Shared scope : Best for personal or team-level testing. No admin approval required. Organization scope : Best for organization-wi --- ## Agent 365 ### Agent 365 URL: https://learn.microsoft.com/en-us/azure/ai-foundry/agents/how-to/agent-365?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Publish an agent to Agent 365 Feedback Summarize this article for me In this article Use this article to publish a Microsoft Foundry hosted agent to Microsoft Agent 365 (Agent 365) by running the FoundryA365 sample. The sample uses the Azure Developer CLI to create the required Azure resources, publish an agent application, and then guides you through admin approval and (optionally) Microsoft Teams configuration. Prerequisites Enrollment in the Frontier preview program . An Azure subscription where you can create resources. The required permissions: Owner role on the Azure subscription Azure AI User or Cognitive Services User role at subscription or resource group scope A tenant admin role for organization-wide configuration Azure CLI Azure Developer CLI Docker .NET 9.0 SDK Git Before you begin Hosted agents are only available in the North Central US region. Create all resources for this sample in that region. Start Docker before you deploy. Treat your deployment outputs as sensitive. The azd env get-values output can include IDs and endpoints you don't want to publish. Run the sample Use the FoundryA365 sample on GitHub: https://go.microsoft.com/fwlink/?linkid=2343518 Clone the sample repository and switch to the sample folder. git clone https://github.com/microsoft-foundry/foundry-samples.git cd foundry-samples\samples\csharp\FoundryA365 Authenticate to Azure and Azure Developer CLI. # Azure CLI az login az login --scope https://ai.azure.com/.default az login --scope https://graph.microsoft.com//.default az login --scope https://management.azure.com/.default # Azure Developer CLI azd auth login Note Depending on your tenant se --- ## Service monitoring ### Service monitoring URL: https://learn.microsoft.com/en-us/azure/ai-foundry/agents/how-to/metrics?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Monitor Foundry Agent Service with Azure Monitor Feedback Summarize this article for me In this article Note This document refers to the Microsoft Foundry (classic) portal. 🔄 Switch to the Microsoft Foundry (new) documentation if you're using the new portal. Note This document refers to the Microsoft Foundry (new) portal. This article describes: The types of monitoring data you can collect for this service. Ways to analyze that data. Note If you're already familiar with this service and/or Azure Monitor and just want to know how to analyze monitoring data, see the Analyze section near the end of this article. When you have critical applications and business processes that rely on Azure resources, you need to monitor and get alerts for your system. The Azure Monitor service collects and aggregates metrics and logs from every component of your system. Azure Monitor provides you with a view of availability, performance, and resilience, and notifies you of issues. You can use the Azure portal, PowerShell, Azure CLI, REST API, or client libraries to set up and view monitoring data. For more information on Azure Monitor, see the Azure Monitor overview . For more information on how to monitor Azure resources in general, see Monitor Azure resources with Azure Monitor . Monitoring is available for agents in a standard agent setup . Note This feature is currently in public preview. This preview is provided without a service-level agreement, and we don't recommend it for production workloads. Certain features might not be supported or might have constrained capabilities. For more information, see Supplemental Terms of Use for Microsoft Azu --- ## Monitor agents in the dashboard ### Monitor agents in the dashboard URL: https://learn.microsoft.com/en-us/azure/ai-foundry/observability/how-to/how-to-monitor-agents-dashboard?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Monitor agents with the Agent Monitoring Dashboard (preview) Feedback Summarize this article for me In this article Important Items marked (preview) in this article are currently in public preview. This preview is provided without a service-level agreement, and we don't recommend it for production workloads. Certain features might not be supported or might have constrained capabilities. For more information, see Supplemental Terms of Use for Microsoft Azure Previews . Use the Agent Monitoring Dashboard in Microsoft Foundry to track operational metrics and evaluation results for your agents. This dashboard helps you understand token usage, latency, success rates, and evaluation outcomes for production traffic. This article covers two approaches: viewing metrics in the Foundry portal and setting up continuous evaluation programmatically with the Python SDK. Prerequisites A Foundry project with at least one agent . An Application Insights resource connected to your project. Azure role-based access control (RBAC) access to the Application Insights resource. For log-based views, you also need access to the associated Log Analytics workspace. To verify access, open the Application Insights resource in the Azure portal, select Access control (IAM) , and confirm your account has an appropriate role. For log access, assign the Log Analytics Reader role . Connect Application Insights The Agent Monitoring Dashboard reads telemetry from the Application Insights resource connected to your Foundry project. If you haven't connected Application Insights yet, follow the tracing setup steps and then return to this article. How to set up tracing i --- ### Monitor agents in the dashboard URL: https://learn.microsoft.com/en-us/azure/ai-foundry/observability/how-to/how-to-monitor-agents-dashboard?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Monitor agents with the Agent Monitoring Dashboard (preview) Feedback Summarize this article for me In this article Important Items marked (preview) in this article are currently in public preview. This preview is provided without a service-level agreement, and we don't recommend it for production workloads. Certain features might not be supported or might have constrained capabilities. For more information, see Supplemental Terms of Use for Microsoft Azure Previews . Use the Agent Monitoring Dashboard in Microsoft Foundry to track operational metrics and evaluation results for your agents. This dashboard helps you understand token usage, latency, success rates, and evaluation outcomes for production traffic. This article covers two approaches: viewing metrics in the Foundry portal and setting up continuous evaluation programmatically with the Python SDK. Prerequisites A Foundry project with at least one agent . An Application Insights resource connected to your project. Azure role-based access control (RBAC) access to the Application Insights resource. For log-based views, you also need access to the associated Log Analytics workspace. To verify access, open the Application Insights resource in the Azure portal, select Access control (IAM) , and confirm your account has an appropriate role. For log access, assign the Log Analytics Reader role . Connect Application Insights The Agent Monitoring Dashboard reads telemetry from the Application Insights resource connected to your Foundry project. If you haven't connected Application Insights yet, follow the tracing setup steps and then return to this article. How to set up tracing i --- ## Trace agent overview ### Trace agent overview URL: https://learn.microsoft.com/en-us/azure/ai-foundry/observability/concepts/trace-agent-concept?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Agent tracing overview (preview) Feedback Summarize this article for me In this article Important Items marked (preview) in this article are currently in public preview. This preview is provided without a service-level agreement, and we don't recommend it for production workloads. Certain features might not be supported or might have constrained capabilities. For more information, see Supplemental Terms of Use for Microsoft Azure Previews . Microsoft Foundry provides an observability platform for monitoring and tracing AI agents. It captures key details during an agent run, such as inputs, outputs, tool usage, retries, latencies, and costs. Understanding the reasoning behind your agent's executions is important for troubleshooting and debugging. However, understanding complex agents presents challenges for several reasons: There could be a high number of steps involved in generating a response, making it hard to keep track of all of them. The sequence of steps might vary based on user input. The inputs/outputs at each stage might be long and deserve more detailed inspection. Each step of an agent's runtime might also involve nesting. For example, an agent might invoke a tool, which uses another process, which then invokes another tool. If you notice strange or incorrect output from a top-level agent run, it might be difficult to determine exactly where in the execution the issue was introduced. Trace results solve this by allowing you to view the inputs and outputs of each primitive involved in a particular agent run, displayed in the order they were invoked, making it easy to understand and debug your AI agent's behavior. Prere --- ## Set up tracing ### Set up tracing URL: https://learn.microsoft.com/en-us/azure/ai-foundry/observability/how-to/trace-agent-setup?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Set up tracing in Microsoft Foundry (preview) Feedback Summarize this article for me In this article Important Items marked (preview) in this article are currently in public preview. This preview is provided without a service-level agreement, and we don't recommend it for production workloads. Certain features might not be supported or might have constrained capabilities. For more information, see Supplemental Terms of Use for Microsoft Azure Previews . Use tracing (preview) to debug your AI agents and monitor their behavior in production. Tracing captures detailed telemetry—including latency, exceptions, prompt content, and retrieval operations—so you can identify and fix issues faster. Prerequisites A Foundry project . An Azure Monitor Application Insights resource to store traces (create a new one or connect an existing one). Access to the Application Insights resource connected to your project. Connect Application Insights to your Foundry project Foundry stores traces in Azure Application Insights by using OpenTelemetry semantic conventions . Sign in to Microsoft Foundry . Make sure the New Foundry toggle is on. These steps refer to Foundry (new) . Open your Foundry project. In the left navigation, select Tracing . Create or connect an Application Insights resource: To connect an existing resource, select the resource and then select Connect . To create a new resource, select Create new and complete the wizard. A confirmation message appears when the connection succeeds. After you connect the resource, your project is ready to use tracing. Important Make sure you have the permissions you need to query telemetry. For log-base --- ### Set up tracing URL: https://learn.microsoft.com/en-us/azure/ai-foundry/observability/how-to/trace-agent-setup?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Set up tracing in Microsoft Foundry (preview) Feedback Summarize this article for me In this article Important Items marked (preview) in this article are currently in public preview. This preview is provided without a service-level agreement, and we don't recommend it for production workloads. Certain features might not be supported or might have constrained capabilities. For more information, see Supplemental Terms of Use for Microsoft Azure Previews . Use tracing (preview) to debug your AI agents and monitor their behavior in production. Tracing captures detailed telemetry—including latency, exceptions, prompt content, and retrieval operations—so you can identify and fix issues faster. Prerequisites A Foundry project . An Azure Monitor Application Insights resource to store traces (create a new one or connect an existing one). Access to the Application Insights resource connected to your project. Connect Application Insights to your Foundry project Foundry stores traces in Azure Application Insights by using OpenTelemetry semantic conventions . Sign in to Microsoft Foundry . Make sure the New Foundry toggle is on. These steps refer to Foundry (new) . Open your Foundry project. In the left navigation, select Tracing . Create or connect an Application Insights resource: To connect an existing resource, select the resource and then select Connect . To create a new resource, select Create new and complete the wizard. A confirmation message appears when the connection succeeds. After you connect the resource, your project is ready to use tracing. Important Make sure you have the permissions you need to query telemetry. For log-base --- ## Tracing integrations ### Tracing integrations URL: https://learn.microsoft.com/en-us/azure/ai-foundry/observability/how-to/trace-agent-framework?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Configure tracing for AI agent frameworks (preview) Feedback Summarize this article for me In this article Important Items marked (preview) in this article are currently in public preview. This preview is provided without a service-level agreement, and we don't recommend it for production workloads. Certain features might not be supported or might have constrained capabilities. For more information, see Supplemental Terms of Use for Microsoft Azure Previews . When AI agents behave unexpectedly in production, tracing gives you the visibility to quickly identify the root cause. Tracing captures detailed telemetry—including LLM calls, tool invocations, and agent decision flows—so you can debug issues, monitor latency, and understand agent behavior across requests. Microsoft Foundry provides tracing integrations for popular agent frameworks that require minimal code changes. In this article, you learn how to: Configure automatic tracing for Microsoft Agent Framework and Semantic Kernel Set up the langchain-azure-ai tracer for LangChain and LangGraph Instrument the OpenAI Agents SDK with OpenTelemetry Verify that traces appear in the Foundry portal Troubleshoot common tracing issues Prerequisites A Foundry project with tracing connected to Application Insights. Contributor or higher role on the Application Insights resource for trace ingestion. Access to the connected Application Insights resource for viewing traces. For log-based queries, you might also need access to the associated Log Analytics workspace. Python 3.10 or later (required for all code samples in this article). The langchain-azure-ai package version 0.1.0 or later (re --- ### Tracing integrations URL: https://learn.microsoft.com/en-us/azure/ai-foundry/observability/how-to/trace-agent-framework?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Configure tracing for AI agent frameworks (preview) Feedback Summarize this article for me In this article Important Items marked (preview) in this article are currently in public preview. This preview is provided without a service-level agreement, and we don't recommend it for production workloads. Certain features might not be supported or might have constrained capabilities. For more information, see Supplemental Terms of Use for Microsoft Azure Previews . When AI agents behave unexpectedly in production, tracing gives you the visibility to quickly identify the root cause. Tracing captures detailed telemetry—including LLM calls, tool invocations, and agent decision flows—so you can debug issues, monitor latency, and understand agent behavior across requests. Microsoft Foundry provides tracing integrations for popular agent frameworks that require minimal code changes. In this article, you learn how to: Configure automatic tracing for Microsoft Agent Framework and Semantic Kernel Set up the langchain-azure-ai tracer for LangChain and LangGraph Instrument the OpenAI Agents SDK with OpenTelemetry Verify that traces appear in the Foundry portal Troubleshoot common tracing issues Prerequisites A Foundry project with tracing connected to Application Insights. Contributor or higher role on the Application Insights resource for trace ingestion. Access to the connected Application Insights resource for viewing traces. For log-based queries, you might also need access to the associated Log Analytics workspace. Python 3.10 or later (required for all code samples in this article). The langchain-azure-ai package version 0.1.0 or later (re --- ## Tool catalog (preview) ### Tool catalog (preview) URL: https://learn.microsoft.com/en-us/azure/ai-foundry/agents/concepts/tool-catalog?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Discover and manage tools in the Foundry tool catalog (preview) Feedback Summarize this article for me In this article Foundry Tools is the place to discover and manage tools you use with agents and workflows in Microsoft Foundry. Note This feature is currently in public preview. This preview is provided without a service-level agreement, and we don't recommend it for production workloads. Certain features might not be supported or might have constrained capabilities. For more information, see Supplemental Terms of Use for Microsoft Azure Previews . You can use Foundry Tools to: Discover tools such as Model Context Protocol (MCP) servers and built-in tools. Configure tools once, then add them to agents or workflows. Filter, search, and sort tools. Prerequisites To use Foundry Tools, you need: Access to a Foundry project in the Foundry portal. Permission to view and manage tools in that project. Where to find Foundry Tools In the Foundry portal, go to your project and then select Build > Tools . Key concepts Use these definitions to keep the terminology consistent: Term Meaning Foundry Tools The portal experience where you discover, configure, and manage tools for agents and workflows. Tool catalog The browsable list of available tools (public and organizational). Private tool catalog An organization-scoped catalog for tools that only users in your organization can discover and configure. MCP server A server that exposes tools using the Model Context Protocol (MCP). Remote MCP server An MCP server hosted by the publisher. You configure it by providing the required settings (for example, an endpoint and authentication details). Lo --- ## Create a private tool catalog (preview) ### Create a private tool catalog (preview) URL: https://learn.microsoft.com/en-us/azure/ai-foundry/agents/how-to/private-tool-catalog?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Create a private tool catalog (preview) Feedback Summarize this article for me In this article Note This feature is currently in public preview. This preview is provided without a service-level agreement, and we don't recommend it for production workloads. Certain features might not be supported or might have constrained capabilities. For more information, see Supplemental Terms of Use for Microsoft Azure Previews . Create a private tool catalog so developers in your organization can discover, configure, and use MCP server tools through Foundry Tools. A private tool catalog uses Azure API Center to register organization-scoped tools that only your developers can access. Prerequisites A Foundry project. For setup guidance, see Create projects in Microsoft Foundry . Permissions to discover and configure tools in your Foundry project. For more information, see Role-based access control in Microsoft Foundry . An Azure API Center . Note The API Center name is the name that developers use to find the catalog in Foundry Tools. Use a descriptive name. One or more remote MCP servers that you want to share with your organization. Register them with API Center by following Configure environments and deployments in Azure API Center . Plan administrator and developer access Before you create the catalog, decide who manages it and who consumes it. Goal Who Where What to do Create and manage the tool catalog Catalog admins Azure API Center Create the API Center resource, register MCP servers, and (optionally) configure authorization settings. Discover tools from the private catalog Developers Azure API Center (RBAC) Assign access so developers --- ## Tool best practices ### Tool best practices URL: https://learn.microsoft.com/en-us/azure/ai-foundry/agents/concepts/tool-best-practice?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Tool best practices for Microsoft Foundry Agent Service Feedback Summarize this article for me In this article When you build agents in Microsoft Foundry Agent Service, tools extend what your agent can do—retrieving information, calling APIs, and connecting to external services. This article helps you configure tools effectively, control when the model calls them, and keep your data secure. Tip In your agent instructions, describe what each tool is for and when to use it. For example: When you need information from my indexed documents, use File Search. When you need to call an API, use the OpenAPI tool. When a tool call fails or returns no results, explain what happened and ask a follow-up question. Prerequisites Access to a Foundry project in the Foundry portal with the Azure AI Developer role or equivalent permissions. A model deployed in the same project. Any required connections configured for the tools you plan to use (for example, Azure AI Search, SharePoint, or Bing grounding). Configure and validate tool usage Configure tools and connections in the Foundry tool catalog. See Discover and manage tools in the Foundry tool catalog (preview) . Review run traces to confirm when your agent calls tools and to inspect tool inputs and outputs. For end-to-end tracing setup, see Trace your application . Improve tool-calling reliability Control tool calling with tool_choice Use tool_choice for the most deterministic control over tool calling. auto : The model decides whether to call tools. required : The model must call one or more tools. none : The model doesn't call tools. For details, see tool_choice in Foundry project REST (prev --- ## Tool governance ### Tool governance URL: https://learn.microsoft.com/en-us/azure/ai-foundry/agents/how-to/tools/governance?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Govern MCP tools with AI Gateway (preview) Feedback Summarize this article for me In this article Control how your agents access external tools by routing MCP traffic through AI Gateway in Microsoft Foundry. AI Gateway provides a single, governed entry point where you can enforce authentication, rate limits, IP restrictions, and audit logging—without modifying your MCP servers or agent code. Note Only new MCP tools created in the Microsoft Foundry portal that don't use managed OAuth are routed through AI Gateway. Prerequisites To enable governance for tools using AI Gateway in Microsoft Foundry: AI Gateway must be connected to the Microsoft Foundry resource Governance is activated at the Microsoft Foundry resource level. All governance functionality depends on this connection. Permissions to manage API Management policies: API Management Service Contributor or Owner role on the connected APIM instance. For more information, see Azure RBAC for API Management . The MCP server must support one of the following authentication methods: Managed identity (Microsoft Entra) Key-based (API key or token) Custom OAuth identity passthrough Unauthenticated (if applicable) Key benefits Secure routing for all new MCP tools using a gateway endpoint Consistent access control and authentication enforcement Centralized observability for gateway traffic (such as logs and metrics) Unified policies for throttling, IP restrictions, and routing Seamless reuse of tools through public and private catalogs Enable AI Gateway for your Foundry resource If AI Gateway isn't already connected to your Foundry resource, enable it first. Follow the steps in Configu --- ## Code interpreter ### Code interpreter URL: https://learn.microsoft.com/en-us/azure/ai-foundry/agents/how-to/tools/code-interpreter?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Code Interpreter tool for Microsoft Foundry agents Feedback Summarize this article for me In this article Code Interpreter enables a Microsoft Foundry agent to run Python code in a sandboxed execution environment. Use this tool for data analysis, chart generation, and iterative problem-solving tasks that benefit from code execution. In this article, you create an agent that uses Code Interpreter, upload a CSV file for analysis, and download a generated chart. When enabled, your agent can write and run Python code iteratively to solve data analysis and math tasks, and to generate charts. Important Code Interpreter has additional charges beyond the token-based fees for Azure OpenAI usage. If your agent calls Code Interpreter simultaneously in two different conversations, two Code Interpreter sessions are created. Each session is active by default for one hour with an idle timeout of 30 minutes. Usage support Microsoft Foundry support Python SDK C# SDK JavaScript SDK Java SDK REST API Basic agent setup Standard agent setup ✔️ ✔️ ✔️ - - - ✔️ ✔️ ✔️ indicates the feature is supported. - indicates the feature isn't currently available for that SDK or API. Prerequisites Basic or standard agent environment. See agent environment setup for details. Latest prerelease SDK package installed ( azure-ai-projects>=2.0.0b1 for Python). See the quickstart for installation steps. Azure AI model deployment configured in your project. For file operations: CSV or other supported files to upload for analysis. Note Code Interpreter isn't available in all regions. See Check regional and model availability . Create an agent with Code Interpreter The foll --- ## Custom code interpreter (preview) ### Custom code interpreter (preview) URL: https://learn.microsoft.com/en-us/azure/ai-foundry/agents/how-to/tools/custom-code-interpreter?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Custom code interpreter tool for agents (preview) Feedback Summarize this article for me In this article Important Items marked (preview) in this article are currently in public preview. This preview is provided without a service-level agreement, and we don't recommend it for production workloads. Certain features might not be supported or might have constrained capabilities. For more information, see Supplemental Terms of Use for Microsoft Azure Previews . A custom code interpreter gives you full control over the runtime environment for agent-generated Python code. You can configure custom Python packages, compute resources, and Azure Container Apps environment settings. The code interpreter container exposes a Model Context Protocol (MCP) server. Use a custom code interpreter when the built-in Code Interpreter tool for agents doesn't meet your requirements—for example, when you need specific Python packages, custom container images, or dedicated compute resources. For more information about MCP and how agents connect to MCP tools, see Connect to Model Context Protocol servers (preview) . Usage support This article uses the Azure CLI and a runnable sample project. Microsoft Foundry support Python SDK C# SDK JavaScript SDK Java SDK REST API Basic agent setup Standard agent setup ✔️ ✔️ - - - ✔️ - ✔️ For the latest SDK and API support for agents tools, see Best practices for using tools in Microsoft Foundry Agent Service . Prerequisites Azure CLI version 2.60.0 or later. (Optional) uv for faster Python package management. An Azure subscription and resource group with the following role assignments: Azure AI Owner Container Apps Ma --- ## Browser automation (preview) ### Browser automation (preview) URL: https://learn.microsoft.com/en-us/azure/ai-foundry/agents/how-to/tools/browser-automation?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Automate browser tasks with the Browser Automation tool (preview) Feedback Summarize this article for me In this article Important Items marked (preview) in this article are currently in public preview. This preview is provided without a service-level agreement, and we don't recommend it for production workloads. Certain features might not be supported or might have constrained capabilities. For more information, see Supplemental Terms of Use for Microsoft Azure Previews . This article explains how to configure and use the Browser Automation tool with Foundry agents to automate web browsing workflows. Warning The Browser Automation tool comes with significant security risks. Both errors in judgment by the AI and the presence of malicious or confusing instructions on web pages that the AI encounters can cause it to execute commands you or others don't intend. These actions can compromise the security of your or other users' browsers, computers, and any accounts to which the browser or AI has access, including personal, financial, or enterprise systems. By using the Browser Automation tool, you acknowledge that you bear responsibility and liability for any use of it and of any resulting agents you create with it. This responsibility extends to any other users to whom you make Browser Automation tool functionality available, including through resulting agents. Use the Browser Automation tool on low-privilege virtual machines with no access to sensitive data or critical resources. For guidance on optimizing tool usage, see Best practices for using tools in Microsoft Foundry Agent Service . In Microsoft Foundry, the Browser Automatio --- ## Computer Use (preview) ### Computer Use (preview) URL: https://learn.microsoft.com/en-us/azure/ai-foundry/agents/how-to/tools/computer-use?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Use the computer use tool for agents (Preview) Feedback Summarize this article for me In this article Important Items marked (preview) in this article are currently in public preview. This preview is provided without a service-level agreement, and we don't recommend it for production workloads. Certain features might not be supported or might have constrained capabilities. For more information, see Supplemental Terms of Use for Microsoft Azure Previews . Warning The computer use tool comes with significant security and privacy risks, including prompt injection attacks. For more information about intended uses, capabilities, limitations, risks, and considerations when choosing a use case, see the Azure OpenAI transparency note . Create agents that interpret screenshots and automate UI interactions like clicking, typing, and scrolling. The computer use tool uses the computer-use-preview model to propose actions based on visual content, enabling agents to interact with desktop and browser applications through their user interfaces. This guide shows how to integrate the computer use tool into an application loop (screenshot → action → screenshot) by using the latest SDKs. Usage support Microsoft Foundry support Python SDK C# SDK JavaScript SDK Java SDK REST API Basic agent setup Standard agent setup ✔️ ✔️ ✔️ ✔️ - - ✔️ ✔️ Prerequisites An Azure subscription. Create one for free . A basic or standard agent environment . The latest prerelease SDK package: Python : azure-ai-projects>=2.0.0b1 , azure-identity , python-dotenv C#/.NET : Azure.AI.Agents.Persistent (prerelease) TypeScript : @azure/ai-projects v2-beta, @azure/identity Access --- ## Image generation (preview) ### Image generation (preview) URL: https://learn.microsoft.com/en-us/azure/ai-foundry/agents/how-to/tools/image-generation?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Use the image generation tool (preview) Feedback Summarize this article for me In this article Important Items marked (preview) in this article are currently in public preview. This preview is provided without a service-level agreement, and we don't recommend it for production workloads. Certain features might not be supported or might have constrained capabilities. For more information, see Supplemental Terms of Use for Microsoft Azure Previews . Important The image generation tool requires the gpt-image-1 model. See the Azure OpenAI transparency note for limitations and responsible AI considerations. You also need a compatible orchestrator model ( gpt-4o , gpt-4o-mini , gpt-4.1 , gpt-4.1-mini , gpt-4.1-nano , o3 , or gpt-5 series) deployed in the same Foundry project. The image generation tool in Microsoft Foundry Agent Service generates images from text prompts in conversations and multistep workflows. Use it to create AI-generated visuals and return base64-encoded output that you can save to a file. Usage support Microsoft Foundry support Python SDK C# SDK JavaScript SDK Java SDK REST API Basic agent setup Standard agent setup ✔️ ✔️ ✔️ ✔️ - ✔️ ✔️ ✔️ Prerequisites An Azure account with an active subscription. A Foundry project. A basic or standard agent environment. See agent environment setup . Permissions to create and manage agent versions in the project. Two model deployments in the same Foundry project: A compatible Azure OpenAI model deployment for the agent (for example, gpt-4o ). An image generation model deployment ( gpt-image-1 ). Set these environment variables for the samples: FOUNDRY_PROJECT_ENDPOINT FOUNDRY_MODE --- ## What is memory? ### What is memory? URL: https://learn.microsoft.com/en-us/azure/ai-foundry/agents/concepts/what-is-memory?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Memory in Microsoft Foundry Agent Service (preview) Feedback Summarize this article for me In this article Important Memory (preview) in Foundry Agent Service and the Memory Store API (preview) are licensed to you as part of your Azure subscription and are subject to terms applicable to "Previews" in the Microsoft Product Terms and the Microsoft Products and Services Data Protection Addendum , as well as the Microsoft Generative AI Services Previews terms in the Supplemental Terms of Use for Microsoft Azure Previews . Memory in Microsoft Foundry Agent Service is a managed, long-term memory solution. It enables agent continuity across sessions, devices, and workflows. By creating and managing memory stores, you can build agents that retain user preferences, maintain conversation history, and deliver personalized experiences. This article provides an overview of agent memory, including its concepts, use cases, and limitations. For usage instructions, see Create and use memory in Foundry Agent Service . What is memory? Memory is persistent knowledge retained by an agent across sessions. Generally, agent memory falls into two categories: Short-term memory tracks the current session's conversation and maintains immediate context for ongoing interactions. Agent orchestration frameworks typically manage this memory as part of the session context. Long-term memory retains distilled knowledge across sessions. The model can recall and build on previous user interactions over time. Long-term memory requires a persistent system that extracts, consolidates, and manages knowledge. Memory in Foundry Agent Service is designed for long-term memo --- ## Create and use memory ### Create and use memory URL: https://learn.microsoft.com/en-us/azure/ai-foundry/agents/how-to/memory-usage?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Create and use memory in Foundry Agent Service (preview) Feedback Summarize this article for me In this article Important Memory (preview) in Foundry Agent Service and the Memory Store API (preview) are licensed to you as part of your Azure subscription and are subject to terms applicable to "Previews" in the Microsoft Product Terms and the Microsoft Products and Services Data Protection Addendum , as well as the Microsoft Generative AI Services Previews terms in the Supplemental Terms of Use for Microsoft Azure Previews . Memory in Foundry Agent Service is a managed, long-term memory solution. It enables agent continuity across sessions, devices, and workflows. By creating and managing memory stores, you can build agents that retain user preferences, maintain conversation history, and deliver personalized experiences. Memory stores act as persistent storage, defining which types of information are relevant to each agent. You control access using the scope parameter, which segments memory across users to ensure secure and isolated experiences. This article explains how to create, manage, and use memory stores. For conceptual information, see Memory in Foundry Agent Service . Usage support Capability Python SDK REST API Create, update, list, and delete memory stores ✔️ ✔️ Update and search memories ✔️ ✔️ Attach memory to a prompt agent ✔️ ✔️ Prerequisites An Azure subscription. Create one for free . A Microsoft Foundry project with authorization and permissions configured. Chat model deployment (for example, gpt-4.1 ) in your project. Embedding model deployment (for example, text-embedding-3-small ) in your project. For Python ex --- ## Retrieval Augmented Generation (RAG) overview ### Retrieval Augmented Generation (RAG) overview URL: https://learn.microsoft.com/en-us/azure/ai-foundry/concepts/retrieval-augmented-generation?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Retrieval augmented generation (RAG) and indexes Feedback Summarize this article for me In this article Retrieval augmented generation (RAG) is a pattern that combines search with large language models (LLMs) so responses are grounded in your data. This article explains how RAG works in Microsoft Foundry, what role indexes play, and how agentic retrieval changes classic RAG patterns. LLMs are trained on public data available at training time. If you need answers based on your private data, or on frequently changing information, RAG helps you: Retrieve relevant information from your data (often through an index). Provide that information to the model as grounding data. Generate a response that can include citations back to source content. What is RAG? Large language models (LLMs) like ChatGPT are trained on public internet data that was available when the model was trained. The public data might not be sufficient for your needs. For example, you might want answers based on private documents, or you might need up-to-date information. RAG addresses this by retrieving relevant content from your data and including it in the model input. The model can then generate responses grounded in the retrieved content. Key concepts for RAG: Grounding data : Retrieved content you provide to the model to reduce guessing. Index : A data structure optimized for retrieval (keyword, semantic, vector, or hybrid search). Embeddings : Numeric representations of content used for vector similarity search. See Understand embeddings . System message and prompts : Instructions that guide how the model uses retrieved content. See Prompt engineering and Safety --- ## Azure AI Search ### Azure AI Search URL: https://learn.microsoft.com/en-us/azure/ai-foundry/agents/how-to/tools/ai-search?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Connect an Azure AI Search index to Foundry agents Feedback Summarize this article for me In this article Tip For a managed knowledge base experience, see Foundry IQ . For tool optimization, see best practices . Ground your Foundry agent's responses in your proprietary content by connecting it to an Azure AI Search index. The Azure AI Search tool retrieves indexed documents and generates answers with inline citations, enabling accurate, source-backed responses. Usage support Microsoft Foundry support Python SDK C# SDK JavaScript SDK Java SDK REST API Basic agent setup Standard agent setup ✔️ ✔️ ✔️ ✔️ - ✔️ ✔️ ✔️ Java SDK samples are coming soon. Prerequisites Estimated setup time: 15-30 minutes if you have an existing search index A basic or standard agent environment . The latest prerelease package. See the quickstart for details. Python : pip install azure-ai-projects --pre C# : Install the Azure.AI.Projects NuGet package (prerelease) JavaScript/TypeScript : npm install @azure/ai-projects An Azure subscription and Microsoft Foundry project with: Project endpoint Model deployment name Authentication configured (for example, DefaultAzureCredential ) An Azure AI Search index configured for vector search with: One or more Edm.String fields that are searchable and retrievable One or more Collection(Edm.Single) vector fields that are searchable At least one retrievable text field that contains the content you want the agent to cite A retrievable field that contains a source URL (and optionally a title) so citations can include a link A connection between your Foundry project and your Azure AI Search service (see Setup ). For keyless --- ## File search ### File search URL: https://learn.microsoft.com/en-us/azure/ai-foundry/agents/how-to/tools/file-search?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . File search tool for agents Feedback Summarize this article for me In this article Use the file search tool to enable Microsoft Foundry agents to search through your documents and retrieve relevant information. File search augments agents with knowledge from outside their model, such as proprietary product information or user-provided documents. In this article, you learn how to: Upload files and create a vector store Configure an agent with file search enabled Query your documents through the agent Note By using the standard agent setup, the improved file search tool ensures your files remain in your own storage. Your Azure AI Search resource ingests the files, so you maintain complete control over your data. Important File search has additional charges beyond the token-based fees for model usage. Usage support Microsoft Foundry support Python SDK C# SDK JavaScript SDK Java SDK REST API Basic agent setup Standard agent setup ✔️ ✔️ ✔️ ✔️ - ✔️ ✔️ ✔️ Prerequisites A basic or standard agent environment The latest prerelease SDK package: Python : pip install azure-ai-projects azure-identity python-dotenv --pre C# : dotnet add package Azure.AI.Projects.OpenAI --prerelease TypeScript : npm install @azure/ai-projects @azure/identity dotenv Storage Blob Data Contributor role on your project's storage account (required for uploading files to your project's storage) Azure AI Owner role on your Foundry resource (required for creating agent resources) Environment variables configured: FOUNDRY_PROJECT_ENDPOINT , MODEL_DEPLOYMENT_NAME Create an agent with file search Create an agent with the file search tool The following code sample shows ho --- ## Vector stores for file search ### Vector stores for file search URL: https://learn.microsoft.com/en-us/azure/ai-foundry/agents/concepts/vector-stores?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Vector stores for file search Feedback Summarize this article for me In this article Vector store objects give the file search tool the ability to search your files. When you add a file to a vector store, the service parses, chunks, embeds, and indexes it so the tool can run both keyword and semantic search. Vector stores can be attached to both agents and conversations. Currently, you can attach at most one vector store to an agent and at most one vector store to a conversation. For a conceptual overview of conversations, see Agent runtime components . In the current agents developer experience, response generation uses responses and conversations . Some SDKs and older samples use the term run . If you see both terms, treat run as response generation. For migration guidance, see How to migrate to the new agent service . For a list of limits for vector search (such as maximum allowable file sizes), see the quotas and limits article. Prerequisites A Microsoft Foundry project . An agent or conversation that uses the file search tool. If you use standard agent setup, connect Azure Blob Storage and Azure AI Search during setup so your files remain in your storage. See Agent environment setup . Roles and permissions vary by task (for example, creating projects, assigning roles for standard setup, or creating and editing agents). See the required permissions table in Agent environment setup . Feature availability can vary by region. For current coverage, see Microsoft Foundry feature availability across cloud regions . Key limits and defaults Vector stores are often the first place retrieval workflows fail in production, so it helps t --- ## SharePoint (preview) ### SharePoint (preview) URL: https://learn.microsoft.com/en-us/azure/ai-foundry/agents/how-to/tools/sharepoint?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Use SharePoint tool with the agent API (preview) Feedback Summarize this article for me In this article Important Items marked (preview) in this article are currently in public preview. This preview is provided without a service-level agreement, and we don't recommend it for production workloads. Certain features might not be supported or might have constrained capabilities. For more information, see Supplemental Terms of Use for Microsoft Azure Previews . Note This article describes the Microsoft SharePoint tool for Foundry Agent Service. For information on using and deploying SharePoint sites, see the SharePoint documentation . See best practices for information on optimizing tool usage. Use the SharePoint tool (preview) for SharePoint grounding in Microsoft Foundry Agent Service by retrieving content from a SharePoint site or folder (for example, contoso.sharepoint.com/sites/policies ). When a user asks a question, the agent can invoke the SharePoint tool to retrieve relevant text from documents the user can access. The agent then generates a response based on that retrieved content. Note SharePoint tool access requires either a Microsoft 365 Copilot license or an enabled pay-as-you-go model. See Prerequisites for details. This integration uses identity passthrough (On-Behalf-Of) so SharePoint permissions continue to apply to every request. For details on the underlying Microsoft 365 Copilot Retrieval API integration, see How it works . Usage support Microsoft Foundry support Python SDK C# SDK JavaScript SDK Java SDK REST API Basic agent setup Standard agent setup ✔️ ✔️ ✔️ ✔️ - ✔️ ✔️ ✔️ Prerequisites Eligible license or pay-a --- ## Web search overview ### Web search overview URL: https://learn.microsoft.com/en-us/azure/ai-foundry/agents/how-to/tools/web-overview?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Web grounding tools overview Feedback Summarize this article for me In this article Large language models work with a knowledge cutoff. They can't access new information beyond a fixed point in time. By connecting with web grounding tools, your agents can incorporate real-time public web data when generating responses. For example, you can ask questions such as "what is the top AI news today" and receive current, cited answers. How web grounding works The grounding process involves several key steps: Query formulation : The agent identifies information gaps and constructs search queries based on the user's input. Search execution : The grounding tool submits queries to Bing and retrieves results. Information synthesis : The agent processes search results and integrates findings into responses. Source attribution : The agent provides transparency by citing search sources with URLs. Prerequisites Before using any web grounding tool, ensure you have: A basic or standard agent environment . The latest prerelease SDK package. See the quickstart for installation steps. An Azure OpenAI model deployment in your Foundry project. Important Web Search (preview) uses Grounding with Bing Search and Grounding with Bing Custom Search are First Party Consumption Services with terms for online services . They're governed by the Grounding with Bing terms of use and the Microsoft Privacy Statement . The Microsoft Data Protection Addendum doesn't apply to data sent to Grounding with Bing Search or Grounding with Bing Custom Search. When you use these services, your data flows outside the Azure compliance and Geo boundary. This also means use of the --- ## Web search tool (preview) ### Web search tool (preview) URL: https://learn.microsoft.com/en-us/azure/ai-foundry/agents/how-to/tools/web-search?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Web search tool (preview) Feedback Summarize this article for me In this article Important Items marked (preview) in this article are currently in public preview. This preview is provided without a service-level agreement, and we don't recommend it for production workloads. Certain features might not be supported or might have constrained capabilities. For more information, see Supplemental Terms of Use for Microsoft Azure Previews . The web search tool in Foundry Agent Service enables models to retrieve and ground responses with real-time information from the public web before generating output. When enabled, the model can return up-to-date answers with inline citations, helping you build agents that provide current, factual information to users. Important Web Search(preview) uses Grounding with Bing Search and Grounding with Bing Custom Search, which are First Party Consumption Services governed by these Grounding with Bing terms of use and the Microsoft Privacy Statement . The Microsoft Data Protection Addendum doesn't apply to data sent to Grounding with Bing Search and Grounding with Bing Custom Search. When you use Grounding with Bing Search and Grounding with Bing Custom Search, data transfers occur outside compliance and geographic boundaries. Use of Grounding with Bing Search and Grounding with Bing Custom Search incurs costs. See pricing for details. See the management section for information about how Azure admins can manage access to use of web search. Usage support Microsoft Foundry support Python SDK C# SDK JavaScript SDK Java SDK REST API Basic agent setup Standard agent setup ✔️ ✔️ ✔️ ✔️ - ✔️ ✔️ ✔️ Prerequisites --- ## Grounding with Bing tools ### Grounding with Bing tools URL: https://learn.microsoft.com/en-us/azure/ai-foundry/agents/how-to/tools/bing-tools?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Grounding agents with Bing Search tools Feedback Summarize this article for me In this article Traditional language models work with a knowledge cutoff. They can't access new information beyond a fixed point in time. By using Grounding with Bing Search and Grounding with Bing Custom Search (preview), your agents can incorporate real-time public web data when generating responses. By using these tools, you can ask questions such as "what is the top AI news today". The grounding process involves several key steps: Query formulation : The agent identifies information gaps and constructs search queries. Search execution : The grounding tool submits queries to search engines and retrieves results. Information synthesis : The agent processes search results and integrates findings into responses. Source attribution : The agent provides transparency by citing search sources. Important Grounding with Bing Search and Grounding with Bing Custom Search are First Party Consumption Services with terms for online services . They're governed by the Grounding with Bing terms of use and the Microsoft Privacy Statement . The Microsoft Data Protection Addendum doesn't apply to data sent to Grounding with Bing Search or Grounding with Bing Custom Search. When you use these services, your data flows outside the Azure compliance and Geo boundary. This also means use of these services waives all elevated Government Community Cloud security and compliance commitments, including data sovereignty and screened/citizenship-based support, as applicable. Use of Grounding with Bing Search and Grounding with Bing Custom Search incurs costs. See pricing for deta --- ## Fabric data agent (preview) ### Fabric data agent (preview) URL: https://learn.microsoft.com/en-us/azure/ai-foundry/agents/how-to/tools/fabric?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Use the Microsoft Fabric data agent (preview) Feedback Summarize this article for me In this article Important Items marked (preview) in this article are currently in public preview. This preview is provided without a service-level agreement, and we don't recommend it for production workloads. Certain features might not be supported or might have constrained capabilities. For more information, see Supplemental Terms of Use for Microsoft Azure Previews . Note See best practices for information on optimizing tool usage. Use the Microsoft Fabric data agent with Foundry Agent Service to analyze enterprise data in chat. The Fabric data agent turns enterprise data into a conversational question and answer experience. First, build and publish a Fabric data agent. Then, connect your Fabric data agent with the published endpoint. When a user sends a query, the agent determines if it should use the Fabric data agent. If so, it uses the end user's identity to generate queries over data they have access to. Lastly, the agent generates responses based on queries returned from the Fabric data agent. By using identity passthrough (On-Behalf-Of) authorization, this integration simplifies access to enterprise data in Fabric while maintaining robust security, ensuring proper access control and enterprise-grade protection. Usage support Microsoft Foundry support Python SDK C# SDK JavaScript SDK Java SDK REST API Basic agent setup Standard agent setup ✔️ ✔️ ✔️ ✔️ - ✔️ ✔️ ✔️ Prerequisites Note The model you select during agent setup is only used for orchestration and response generation. It doesn't affect which model the Fabric data agent uses for N --- ## What is Foundry IQ? ### What is Foundry IQ? URL: https://learn.microsoft.com/en-us/azure/ai-foundry/agents/concepts/what-is-foundry-iq?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Foundry IQ (preview) Feedback Summarize this article for me In this article Important Items marked (preview) in this article are currently in public preview. This preview is provided without a service-level agreement, and we don't recommend it for production workloads. Certain features might not be supported or might have constrained capabilities. For more information, see Supplemental Terms of Use for Microsoft Azure Previews . Agents need context from scattered enterprise content to accurately answer questions. With Foundry IQ, you can create a configurable, multi-source knowledge base that provides agents with permission-aware responses based on your organization's data. A knowledge base consists of knowledge sources (connections to internal and external data stores) and parameters that control retrieval behavior. Multiple agents can share the same knowledge base. When an agent queries the knowledge base, Foundry IQ uses agentic retrieval to process the query, retrieve relevant information, enforce user permissions, and return grounded answers with citations. Capabilities Connect one knowledge base to multiple agents. Supported knowledge sources include internal data stores (such as Azure Blob Storage, SharePoint, and OneLake) and public web data. Automate document chunking, vector embedding generation, and metadata extraction for indexed knowledge sources. Schedule recurring indexer runs for incremental data refresh. Issue keyword, vector, or hybrid queries across indexed and remote knowledge sources. Use the agentic retrieval engine with a large language model (LLM) to plan queries, select sources, run parallel searches, an --- ## Connect knowledge base to agents ### Connect knowledge base to agents URL: https://learn.microsoft.com/en-us/azure/ai-foundry/agents/how-to/foundry-iq-connect?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Connect a Foundry IQ knowledge base to Foundry Agent Service Feedback Summarize this article for me In this article Important Items marked (preview) in this article are currently in public preview. This preview is provided without a service-level agreement, and we don't recommend it for production workloads. Certain features might not be supported or might have constrained capabilities. For more information, see Supplemental Terms of Use for Microsoft Azure Previews . In this article, you learn how to connect a knowledge base in Foundry IQ to an agent in Foundry Agent Service. The connection uses the Model Context Protocol (MCP) to facilitate tool calls. When invoked by the agent, the knowledge base orchestrates the following operations: Plans and decomposes a user query into subqueries. Processes the subqueries simultaneously using keyword, vector, or hybrid techniques. Applies semantic reranking to identify the most relevant results. Synthesizes the results into a unified response with source references. The agent uses the response to ground its answers in enterprise data or web sources, ensuring factual accuracy and transparency through source attribution. For an end-to-end example of integrating Azure AI Search and Foundry Agent Service for knowledge retrieval, see the agentic-retrieval-pipeline-example Python sample on GitHub. Usage support Microsoft Foundry support Python SDK C# SDK JavaScript SDK Java SDK REST API Basic agent setup Standard agent setup ✔️ ✔️ - - - ✔️ ✔️ ✔️ Prerequisites An Azure AI Search service with a knowledge base containing one or more knowledge sources . A Microsoft Foundry project with an LLM deplo --- ## Azure Speech in Foundry Tools ### Azure Speech in Foundry Tools URL: https://learn.microsoft.com/en-us/azure/ai-foundry/agents/how-to/tools/azure-ai-speech?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Connect Azure Speech in Foundry Tools to an agent Feedback Summarize this article for me In this article Azure Speech in Foundry Tools lets your agent convert speech to text and generate speech audio from text. You connect the tool by adding a remote Model Context Protocol (MCP) server to your agent in Foundry Agent Service. Important The Speech MCP tool doesn't support Network-secured Microsoft Foundry . For more information, see Connect to Model Context Protocol servers . Prerequisites An Azure subscription. Create one for free . A Microsoft Foundry resource created in a supported region . Your Foundry resource includes speech capabilities and is used by the Speech MCP server. The Speech MCP Server is available in all regions where Foundry Agent Service supports MCP tools . Usage support This article shows how to connect the tool in Foundry portal. If you want to work with code, see Connect to Model Context Protocol servers for SDK examples in Python, C#, and JavaScript. Security and privacy Treat your Speech resource key and storage SAS URLs as secrets: Don’t paste keys or SAS URLs into agent prompts, chat transcripts, screenshots, or source control. Use the shortest practical SAS expiry time. Scope SAS URLs to the minimum required resource (for example, a single container). Rotate keys periodically as a security best practice, or immediately if you suspect they're exposed. Set up storage You need an Azure Storage account to store input audio files for speech-to-text processing and receive output audio files from text-to-speech processing. Create an Azure Storage account . Ensure your user account has the Storage Blob Data Co --- ## Azure Language tools and agents ### Azure Language tools and agents URL: https://learn.microsoft.com/azure/ai-services/language-service/concepts/foundry-tools-agents?context=/azure/ai-foundry/context/context?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Azure Language tools and agents Feedback Summarize this article for me In this article Azure Language integrates with Foundry Tools to provide agents and endpoints for building conversational applications. These tools combine Azure Language natural language processing capabilities with Microsoft Foundry agent experiences. This article introduces the Azure Language integrations that are available in Foundry Tools: An Azure Language Model Context Protocol (MCP) server endpoint. An intent routing agent that combines Conversational Language Understanding (CLU) and Custom Question Answering (CQA). An exact question answering agent that uses CQA to return curated, deterministic responses. Key concepts Agent : An AI experience that can interpret user input and choose actions or tools to complete tasks. Tool : A capability an agent can call to retrieve information or perform actions. Model Context Protocol (MCP) : An open protocol for exposing tools and contextual data to agents and large language models. Connected resources and connections : Configuration in Microsoft Foundry that lets an agent access external services (including credentials). Azure Language MCP server 🆕 The Azure Language MCP server in the Microsoft Foundry portal connects agents to Azure Language services through the Model Context Protocol (MCP). This integration helps you build conversational applications that can call Azure Language capabilities as tools. The MCP server exposes Azure Language features through an agent-friendly endpoint that supports real-time workflows. Core capabilities Language processing : Access to Azure Language natural language processing (NL --- ## Try CLU multi-turn conversations ### Try CLU multi-turn conversations URL: https://learn.microsoft.com/azure/ai-services/language-service/conversational-language-understanding/how-to/quickstart-multi-turn-conversations?context=/azure/ai-foundry/context/context?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Quickstart: Multi-turn CLU models with entity slot filling Feedback Summarize this article for me In this article In this article, get started building a CLU model that uses entity slot filling to enable multi-turn conversations. This approach allows your model to collect information progressively across multiple conversation turns, rather than requiring users to provide all details in a single interaction to complete tasks naturally and efficiently. Note Multi-turn entity slot filling is available only in Microsoft Foundry (classic). This quickstart uses the classic portal at https://ai.azure.com/ . For more information about the portals, see What is Microsoft Foundry? . Important Deploying and using models can incur costs in your Azure subscription. Prerequisites Azure subscription - If you don't have one, you can create one for free . Required permissions - Ensure that the person establishing the account and project has the Azure AI Account Owner role at the subscription level. Alternatively, the Contributor or Cognitive Services Contributor role at the subscription scope also meets this requirement. For more information, see Language role-based access control and Assign Azure roles . Azure Language in Foundry Tools resource - Create a Language resource in the Azure portal. Note You need the owner role assigned on the resource group to create a Language resource. Microsoft Foundry project - Create a project in Foundry. For more information, see Create a Foundry project . Deployed OpenAI model - Deploy an OpenAI model in Foundry as described in the Deploy an OpenAI model section. Configure roles, permissions, and settings Begi --- ## Detect Personally Identifiable Information (PII) ### Detect Personally Identifiable Information (PII) URL: https://learn.microsoft.com/azure/ai-services/language-service/personally-identifiable-information/quickstart?context=/azure/ai-foundry/context/context?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Quickstart: Detect personally identifiable information (PII) Feedback Summarize this article for me In this article Note This quickstart guides you through the process of locating personally identifiable information (PII) in documents. To detect PII in conversations, see How to detect and redact PII in conversations . To detect PII in text, see How to detect and redact PII in text . Reference documentation | More samples | Package (NuGet) | Library source code Use this quickstart to create a Personally Identifiable Information (PII) detection application with the client library for .NET. In the following example, you create a C# application that can identify recognized sensitive information in text. Tip You can try the Microsoft Foundry platform to perform Azure Language tasks without the need to write code. Prerequisites Azure subscription - Create one for free Once you have your Azure subscription, create a Foundry resource . The Visual Studio IDE Setting up Create an Azure resource To use the code sample below, you need to deploy an Azure resource. This resource will contain a key and endpoint you use to authenticate the API calls you send to Azure Language. Use the following link to create a language resource using the Azure portal. You need to sign in using your Azure subscription. On the Select additional features screen that appears, select Continue to create your resource . In the Create language screen, provide the following information: Detail Description Subscription The subscription account that your resource will be associated with. Select your Azure subscription from the drop-down menu. Resource group A resource gr --- ## Try Azure Language detection ### Try Azure Language detection URL: https://learn.microsoft.com/azure/ai-services/language-service/language-detection/quickstart?context=/azure/ai-foundry/context/context?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Quickstart: Use the Azure Language Detection client library and REST API Feedback Summarize this article for me In this article Reference documentation | More samples | Package (NuGet) | Library source code Use this quickstart to create a language detection application with the client library for .NET. In the following example, you create a C# application that can identify the language a text sample was written in. Prerequisites Azure subscription - Create one for free The Visual Studio IDE Setting up Create an Azure resource To use the code sample below, you need to deploy an Azure resource. This resource will contain a key and endpoint you use to authenticate the API calls you send to Azure Language. Use the following link to create a language resource using the Azure portal. You need to sign in using your Azure subscription. On the Select additional features screen that appears, select Continue to create your resource . In the Create language screen, provide the following information: Detail Description Subscription The subscription account that your resource will be associated with. Select your Azure subscription from the drop-down menu. Resource group A resource group is a container that stores the resources you create. Select Create new to create a new resource group. Region The location of your Language resource. Different regions may introduce latency depending on your physical location, but have no impact on the runtime availability of your resource. For this quickstart, either select an available region near you, or choose East US . Name The name for your Language resource. This name will also be used to create an endp --- ## Azure text translation ### Azure text translation URL: https://learn.microsoft.com/azure/ai-services/translator/text-translation/overview?context=/azure/ai-foundry/context/context?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . What is Azure Text translation in Foundry Tools? Feedback Summarize this article for me In this article Azure Translator in Foundry Tools is a cloud-based REST API feature of Translator that uses neural machine translation technology to enable quick and accurate source-to-target text translation in real time across all supported languages . In this overview, you learn how the Text translation REST APIs enable you to build intelligent solutions for your applications and workflows. Text translation documentation contains the following article types: Quickstarts . Getting-started instructions to guide you through making requests to the service. How-to guides . Instructions for accessing and using the service in more specific or customized ways. Reference articles . REST API documentation and programming language-based content. Text translation features Use the following tabs to compare the latest preview version and the latest GA version. Latest preview version Latest GA version With the latest preview release, you can optionally select either the standard neural machine translation (NMT) or a Large Language Model (LLM) deployment (GPT-4o-mini or GPT-4o). However, using an LLM model requires you to have a Microsoft Foundry resource. For more information, see configure Azure resources . For a quick overview of Microsoft Foundry, see What is Microsoft Foundry? Languages . Returns a list of languages supported by the Translate and Transliterate APIs. This request doesn't require authentication; just copy and paste the following GET request into your preferred REST API tool or browser: https://api.cognitive.microsofttranslator.com/lang --- ## Azure document translation ### Azure document translation URL: https://learn.microsoft.com/azure/ai-services/translator/document-translation/overview?context=/azure/ai-foundry/context/context?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . What is Azure Translator document translation in Foundry Tools? Feedback Summarize this article for me In this article Document translation is a cloud-based machine translation feature of Azure Translator . You can translate multiple and complex documents across all supported languages and dialects while preserving original document structure and data format. The Document translation API supports two translation processes: Asynchronous batch translation supports the processing of multiple documents and large files. The batch translation process requires an Azure Blob storage account with storage containers for your source and translated documents. Synchronous single file supports the processing of single file translations. The file translation process doesn't require an Azure Blob storage account. The final response contains the translated document and is returned directly to the calling client. Prerequisites Asynchronous batch translation prerequisites Before you start, you need: An active Azure subscription . A Translator resource. For resource creation and endpoint/key retrieval steps, see Use Document translation APIs programmatically . An Azure Blob Storage account with source and target containers. For setup guidance, see Create Azure Blob Storage containers . A way to authorize access to your storage URLs: Shared access signature (SAS) tokens , or Managed identities for Document translation . Synchronous translation prerequisites Before you start, you need: An active Azure subscription . A Translator resource with a custom domain endpoint. For setup and endpoint/key retrieval, see Use Document translation APIs programmati --- ## Connect to MCP server ### Connect to MCP server URL: https://learn.microsoft.com/en-us/azure/ai-foundry/agents/how-to/tools/model-context-protocol?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Connect agents to Model Context Protocol servers (preview) Feedback Summarize this article for me In this article Important Items marked (preview) in this article are currently in public preview. This preview is provided without a service-level agreement, and we don't recommend it for production workloads. Certain features might not be supported or might have constrained capabilities. For more information, see Supplemental Terms of Use for Microsoft Azure Previews . Note When you use a Network Secured Microsoft Foundry , you can't use private MCP servers deployed in the same virtual network. You can only use publicly accessible MCP servers. Connect your Foundry agents to Model Context Protocol (MCP) servers by using the MCP tool. This extends agent capabilities with external tools and data sources. By connecting to remote MCP server endpoints, your agents can access tools hosted by developers and organizations that MCP-compatible clients like Foundry Agent Service can use. MCP is an open standard that defines how applications provide tools and contextual data to large language models (LLMs). It enables consistent, scalable integration of external tools into model workflows. In this article, you learn how to: Add a remote MCP server as a tool. Authenticate to an MCP server by using a project connection. Review and approve MCP tool calls. Troubleshoot common MCP integration issues. For conceptual details about how MCP integration works, see How it works . Usage support The following table shows SDK and setup support for MCP connections. Microsoft Foundry support Python SDK C# SDK JavaScript SDK Java SDK REST API Basic agent setup --- ## MCP authentication ### MCP authentication URL: https://learn.microsoft.com/en-us/azure/ai-foundry/agents/how-to/mcp-authentication?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Set up authentication for Model Context Protocol (MCP) tools (preview) Feedback Summarize this article for me In this article Important Items marked (preview) in this article are currently in public preview. This preview is provided without a service-level agreement, and we don't recommend it for production workloads. Certain features might not be supported or might have constrained capabilities. For more information, see Supplemental Terms of Use for Microsoft Azure Previews . Most Model Context Protocol (MCP) servers require authentication to access the server and its underlying service. Proper authentication ensures your agents can securely connect to MCP servers, invoke their tools, and access protected resources while maintaining appropriate access controls. In this article, you: Choose an authentication method based on your security requirements Configure key-based, Microsoft Entra, or OAuth authentication Set up and validate your MCP server connection Note If you don't already have an account with the MCP server publisher, create one through the publisher's website. Prerequisites Before you begin, you need: Access to the Foundry portal and a project. If you don't have one, see Create projects in Foundry . Permissions to create project connections and configure agents. For details, see Role-based access control in the Foundry portal . The remote MCP server endpoint URL you want to connect to. Credentials for your selected authentication method: Key-based authentication: an API key, personal access token (PAT), or other token. Microsoft Entra authentication: role assignments for the agent identity or project managed identit --- ## Build your own MCP server ### Build your own MCP server URL: https://learn.microsoft.com/en-us/azure/ai-foundry/mcp/build-your-own-mcp-server?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Build and register a Model Context Protocol (MCP) server Feedback Summarize this article for me In this article The Model Context Protocol (MCP) provides a standard interface for AI agents to interact with APIs and external services. When you need to integrate private or internal enterprise systems that don't have existing MCP server implementations, you can build your own custom server. This article shows you how to create a remote MCP server using Azure Functions, register it in a private organizational tool catalog using Azure API Center, and connect it to Foundry Agent Service. This approach enables you to securely integrate internal APIs and services into the Microsoft Foundry ecosystem, allowing agents to call your enterprise-specific tools through a standardized MCP interface. Prerequisites A Foundry project with Agent Service enabled. For setup instructions, see Quickstart: Get started with Agent Service . An Azure subscription and permissions to create resources. At minimum, you typically need the Contributor role on the target resource group. Python version 3.11 or higher installed on your local development machine. Azure Functions Core Tools version 4.0.7030 or higher. Azure Developer CLI installed for deployment automation. For local development and debugging: Visual Studio Code Azure Functions extension for Visual Studio Code An Azure API Center resource (optional, required only for organizational tool catalog registration). Note Agent Service connects only to publicly accessible MCP server endpoints. Understand the request flow The high-level flow looks like this: You deploy an MCP server (this article uses Azure F --- ## Agent2Agent (A2A) tool (preview) ### Agent2Agent (A2A) tool (preview) URL: https://learn.microsoft.com/en-us/azure/ai-foundry/agents/how-to/tools/agent-to-agent?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Add an A2A agent endpoint to Foundry Agent Service (preview) Feedback Summarize this article for me In this article Important Items marked (preview) in this article are currently in public preview. This preview is provided without a service-level agreement, and we don't recommend it for production workloads. Certain features might not be supported or might have constrained capabilities. For more information, see Supplemental Terms of Use for Microsoft Azure Previews . Note For information on optimizing tool usage, see best practices . You can extend the capabilities of your Microsoft Foundry agent by adding an Agent2Agent (A2A) agent endpoint that supports the A2A protocol . The A2A tool enables agent-to-agent communication, making it easier to share context between Foundry agents and external agent endpoints through a standardized protocol. This guide shows you how to configure and use the A2A tool in your Foundry Agent Service. Connecting agents via the A2A tool versus a multi-agent workflow: Using the A2A tool : When Agent A calls Agent B through the A2A tool, Agent B's answer goes back to Agent A. Agent A then summarizes the answer and generates a response for the user. Agent A keeps control and continues to handle future user input. Using a multi-agent workflow : When Agent A calls Agent B through a workflow or other multi-agent orchestration, Agent B takes full responsibility for answering the user. Agent A is out of the loop. Agent B handles all subsequent user input. For more information, see Build a workflow in Microsoft Foundry . Usage support The following table shows SDK and setup support. A checkmark (✔️) indicates --- ## Agent2Agent (A2A) authentication (preview) ### Agent2Agent (A2A) authentication (preview) URL: https://learn.microsoft.com/en-us/azure/ai-foundry/agents/concepts/agent-to-agent-authentication?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Agent2Agent (A2A) authentication Feedback Summarize this article for me In this article The Agent2Agent (A2A) protocol enables your agents to invoke other agents. Most A2A endpoints require authentication to access the endpoint and its underlying service. Configuring authentication ensures that only authorized users can invoke your A2A tools in Foundry Agent Service. This article explains the authentication methods available for A2A connections and helps you choose the right approach for your scenario. Authentication scenarios In general, there are two authentication scenarios: Shared authentication : Every user of your agent uses the same identity to authenticate to the A2A endpoint. Individual user context doesn't persist. This approach is ideal when all users should have the same level of access. For example, if you build a chat agent to retrieve information from Azure Cosmos DB for your organization, you might want every user to access the same shared container without requiring individual sign-in. Individual authentication : Each user of your agent authenticates with their own account, so their user context persists across interactions. This approach is essential when actions should be scoped to the user's permissions. For example, if you build a coding agent that retrieves commits and pull requests from GitHub, you want each developer to sign in with their own GitHub account so they only see repositories they have access to. Prerequisites Before you choose an authentication method, you need: Access to the Foundry portal and a project. If you don't have one, see Create projects in Foundry . The Azure AI User role or higher --- ## OpenAPI tool ### OpenAPI tool URL: https://learn.microsoft.com/en-us/azure/ai-foundry/agents/how-to/tools/openapi?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Connect agents to OpenAPI tools Feedback Summarize this article for me In this article Connect your Microsoft Foundry agents to external APIs using OpenAPI 3.0 specifications. Agents that connect to OpenAPI tools can call external services, retrieve real-time data, and extend their capabilities beyond built-in functions. OpenAPI specifications define a standard way to describe HTTP APIs so you can integrate existing services with your agents. Microsoft Foundry supports three authentication methods: anonymous , API key , and managed identity . For help choosing an authentication method, see Choose an authentication method . Usage support Microsoft Foundry support Python SDK C# SDK JavaScript SDK Java SDK REST API Basic agent setup Standard agent setup ✔️ ✔️ ✔️ ✔️ - ✔️ ✔️ ✔️ Prerequisites Before you begin, make sure you have: An Azure subscription with the right permissions. Azure RBAC role: Contributor or Owner on the Foundry project. A Foundry project created with an endpoint configured. An AI model deployed in your project. A basic or standard agent environment . SDK installed for your preferred language: Python: azure-ai-projects (latest prerelease version) C#: Azure.AI.Projects.OpenAI TypeScript/JavaScript: @azure/ai-projects Environment variables Variable Description AZURE_AI_PROJECT_ENDPOINT Your Foundry project endpoint URL (not the external OpenAPI service endpoint). AZURE_AI_MODEL_DEPLOYMENT_NAME Your deployed model name. OPENAPI_PROJECT_CONNECTION_NAME (For API key auth) Your project connection name for the OpenAPI service. OpenAPI 3.0 specification file that meets these requirements: Each function must have an operatio --- ## Function calling ### Function calling URL: https://learn.microsoft.com/en-us/azure/ai-foundry/agents/how-to/tools/function-calling?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Use function calling with Microsoft Foundry agents Feedback Summarize this article for me In this article Microsoft Foundry agents support function calling, which lets you extend agents with custom capabilities. Define a function with its name, parameters, and description, and the agent can request your app to call it. Your app executes the function and returns the output. The agent then uses the result to continue the conversation with accurate, real-time data from your systems. Important Runs expire 10 minutes after creation. Submit your tool outputs before they expire. You can run agents with function tools in the Microsoft Foundry portal. However, the portal doesn't support adding, removing, or updating function definitions on an agent. Use the SDK or REST API to configure function tools. Usage support Microsoft Foundry support Python SDK C# SDK JavaScript SDK Java SDK REST API Basic agent setup Standard agent setup ✔️ ✔️ ✔️ - - ✔️ ✔️ ✔️ Prerequisites Before you start, make sure you have: A basic or standard agent environment . A Foundry project and a deployed model. The latest prerelease SDK package for your language ( azure-ai-projects>=2.0.0b1 for Python, Azure.AI.Projects.OpenAI prerelease for .NET). For installation and authentication steps, see the quickstart . Environment variables Each language uses different environment variable names. Use one set consistently. Language Project endpoint Model deployment name Python AZURE_AI_PROJECT_ENDPOINT AZURE_AI_MODEL_DEPLOYMENT_NAME C# FOUNDRY_PROJECT_ENDPOINT MODEL_DEPLOYMENT_NAME REST API AZURE_AI_FOUNDRY_PROJECT_ENDPOINT (use the request body field) Tip If you use DefaultAzu --- ### Function calling URL: https://learn.microsoft.com/en-us/azure/ai-foundry/openai/how-to/function-calling?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . How to use function calling with Azure OpenAI in Microsoft Foundry Models Feedback Summarize this article for me In this article If one or more functions are included in your request, the model determines if any of the functions should be called based on the context of the prompt. When the model determines that a function should be called, it responds with a JSON object including the arguments for the function. The models formulate API calls and structure data outputs, all based on the functions you specify. It's important to note that while the models can generate these calls, it's up to you to execute them, ensuring you remain in control. At a high level you can break down working with functions into three steps: Call the chat completions API with your functions and the user’s input Use the model’s response to call your API or function Call the chat completions API again, including the response from your function to get a final response Single tool/function calling example First demonstrate a toy function call that can check the time in three hardcoded locations with a single tool/function defined. We have added print statements to help make the code execution easier to follow: Microsoft Entra ID API Key import os import json from openai import OpenAI from datetime import datetime from zoneinfo import ZoneInfo from azure.identity import DefaultAzureCredential, get_bearer_token_provider token_provider = get_bearer_token_provider( DefaultAzureCredential(), "https://cognitiveservices.azure.com/.default" ) client = OpenAI( base_url = "https://YOUR-RESOURCE-NAME.openai.azure.com/openai/v1/", api_key=token_provider, ) # Define the d --- ## Foundry Models sold directly by Azure ### Foundry Models sold directly by Azure URL: https://learn.microsoft.com/en-us/azure/ai-foundry/foundry-models/concepts/models-sold-directly-by-azure?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Foundry Models sold directly by Azure Feedback Summarize this article for me In this article Note This document refers to the Microsoft Foundry (classic) portal. 🔄 Switch to the Microsoft Foundry (new) documentation if you're using the new portal. Note This document refers to the Microsoft Foundry (new) portal. This article lists a selection of Microsoft Foundry Models sold directly by Azure along with their capabilities, deployment types, and regions of availability , excluding deprecated and legacy models . To see a list of Azure OpenAI models that are supported by the Foundry Agent Service, see Models supported by Agent Service . Models sold directly by Azure include all Azure OpenAI models and specific, selected models from top providers. Depending on the kind of project you use in Microsoft Foundry, you see a different selection of models. Specifically, if you use a Foundry project built on a Foundry resource, you see the models that are available for standard deployment to a Foundry resource. Alternatively, if you use a hub-based project hosted by a Foundry hub, you see models that are available for deployment to managed compute and serverless APIs. These model selections often overlap because many models support multiple deployment options . Foundry Models are available for standard deployment to a Foundry resource. To learn more about attributes of Foundry Models sold directly by Azure, see Explore Foundry Models . Note Foundry Models sold directly by Azure also include select models from top model providers, such as: Black Forest Labs: FLUX.2-pro , FLUX.1-Kontext-pro , FLUX-1.1-pro Cohere: Cohere-command-a , embed-v-4-0 --- ## Foundry Models from partners and community ### Foundry Models from partners and community URL: https://learn.microsoft.com/en-us/azure/ai-foundry/foundry-models/concepts/models-from-partners?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Foundry Models from partners and community Feedback Summarize this article for me In this article Note This document refers to the Microsoft Foundry (classic) portal. 🔄 Switch to the Microsoft Foundry (new) documentation if you're using the new portal. Note This document refers to the Microsoft Foundry (new) portal. This article lists capabilities for a selection of Microsoft Foundry Models from partners and community. Most Foundry Model providers are trusted third-party organizations, partners, research labs, and community contributors. The selection of models that you see in Foundry depends on the kind of project you use. To learn more about attributes of Foundry Models from partners and community, see Explore Foundry Models . Note For a list of models sold directly by Azure, see Foundry Models sold directly by Azure . For a list of Azure OpenAI models that are supported by the Foundry Agent Service, see Models supported by Agent Service . Anthropic Anthropic's flagship product is Claude, a frontier AI model trusted by leading enterprises and millions of users worldwide for complex tasks including coding, agents, financial analysis, research, and office tasks. Claude delivers exceptional performance while maintaining high safety standards. To work with Claude models in Foundry, see Deploy and use Claude models in Microsoft Foundry . Important To use Claude models in Microsoft Foundry, you need a paid Azure subscription with a billing account in a country or region where Anthropic offers the models for purchase. The following paid subscription types are currently restricted: Cloud Solution Providers (CSP), sponsored accounts wi --- ## Model versions ### Model versions URL: https://learn.microsoft.com/en-us/azure/ai-foundry/foundry-models/concepts/model-versions?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Model versions in Microsoft Foundry Models Feedback Summarize this article for me In this article Microsoft Foundry Models are committed to providing the best generative AI models for customers. As part of this commitment, Foundry Models regularly releases new model versions to incorporate the latest features and improvements from key model providers in the industry. How model versions work We want to make it easy for customers to stay up to date as models improve. Customers can choose to start with a particular version and stay on it or to automatically update as new versions are released. We distinguish two different versions when working with models: The version of the model itself. The version of the API used to consume a model deployment. The version of a model is decided when you deploy it. You can choose an update policy, which can include the following options: Deployments set with a specific version or without offering an upgrade policy require a manual upgrade if a new version is released. When the model is retired, those deployments stop working. Deployments set to Auto-update to default automatically update to use the new default version. Deployments set to Upgrade when expired automatically update when its current version is retired. Note Update policies are configured per deployment and vary by model and provider. The API version indicates the contract that you use to interface with the model in code. When using REST APIs, you indicate the API version using the query parameter api-version . Azure SDKs versions are usually paired with specific APIs versions but you can indicate the API version you want to use. A giv --- ## Marketplace configuration for partner models ### Marketplace configuration for partner models URL: https://learn.microsoft.com/en-us/azure/ai-foundry/foundry-models/how-to/configure-marketplace?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Azure Marketplace requirements for Foundry Models from partners Feedback Summarize this article for me In this article Note This document refers to the Microsoft Foundry (classic) portal. 🔄 Switch to the Microsoft Foundry (new) documentation if you're using the new portal. Note This document refers to the Microsoft Foundry (new) portal. Certain Microsoft Foundry Models are offered directly by the model provider through the Azure Marketplace. This article explains the requirements to use Azure Marketplace if you plan to use such models in your workloads. Models sold directly by Azure, like DeepSeek, Black Forest Labs, or Azure OpenAI in Foundry Models, don't have this requirement. Permissions required to subscribe to Models from Partners and Community Foundry Models from partners and community available for deployment (for example, Cohere models) require Azure Marketplace. Model providers define the license terms and set the price for use of their models using Azure Marketplace. When deploying third-party models, ensure you have the following permissions in your account: On the Azure subscription: Microsoft.MarketplaceOrdering/agreements/offers/plans/read Microsoft.MarketplaceOrdering/agreements/offers/plans/sign/action Microsoft.MarketplaceOrdering/offerTypes/publishers/offers/plans/agreements/read Microsoft.Marketplace/offerTypes/publishers/offers/plans/agreements/read Microsoft.SaaS/register/action On the resource group—to create and use the SaaS resource: Microsoft.SaaS/resources/read Microsoft.SaaS/resources/write Country/region availability Users can access models from partners and community with pay-as-you-go billing only --- ## GPT-5 vs GPT-4.1 ### GPT-5 vs GPT-4.1 URL: https://learn.microsoft.com/en-us/azure/ai-foundry/foundry-models/how-to/model-choice-guide?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . GPT-5 vs GPT-4.1: choosing the right model for your use case Feedback Summarize this article for me In this article GPT-5 is the first model from OpenAI that introduces four adjustable levels of thinking, controlling the amount of time and tokens the model uses when responding to a prompt. When selecting which model to use, or whether to use a reasoning model at all, it is important to consider your application’s priorities. Scenarios like researching and producing a report involve the collection, processing, and generation of large amounts of data. Customers in these scenarios are typically willing to wait many minutes for a high-quality report to be generated. A reasoning model like GPT-5 with medium or high thinking is great for this use case. Another example is a coding assistant, where you want to vary the amount of thinking based on the complexity of the coding task. Here, you want your customers to have control over the amount of time and level of effort the model exerts before providing a response. GPT-5 or GPT-5 mini with controllable thinking levels are a great solution. In contrast, a customer service assistant that is answering customer questions live, retrieving information from a highly efficient search index, and providing human-like responses needs to be fast, friendly, and efficient. For these scenarios, OpenAI’s GPT-4.1 is a far better option. Choosing the right model for your use case can be a challenging endeavor, so we’ve created this simple guide to help you pick between the two latest flagship models from OpenAI – GPT-5 and GPT-4.1. Microsoft Foundry offers multiple variants of generative AI models to meet --- ## Automatic model updates ### Automatic model updates URL: https://learn.microsoft.com/en-us/azure/ai-foundry/openai/how-to/working-with-models?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Working with models Feedback Summarize this article for me In this article Azure OpenAI in Microsoft Foundry Models is powered by a diverse set of models with different capabilities and price points. Model availability varies by region . You can get a list of models that are available for both inference and fine-tuning by your Azure OpenAI resource by using the Models List API . Model updates Azure OpenAI supports automatic updates for select model deployments. On models where automatic update support is available, a model version upgrade policy drop-down is available. You can learn more about Azure OpenAI model versions and how they work in the Azure OpenAI model versions article. Note Automatic model updates are only supported for Standard deployment types. For more information on how to manage model updates and migrations on provisioned deployment types, refer to the section on managing models on provisioned deployment types Auto update to default When you set your deployment to Auto-update to default , your model deployment is automatically updated within two weeks of a change in the default version. For a preview version, it updates automatically when a new preview version is available starting two weeks after the new preview version is released. If you're still in the early testing phases for inference models, we recommend deploying models with auto-update to default set whenever it's available. Specific model version As your use of Azure OpenAI evolves, and you start to build and integrate with applications you might want to manually control model updates. You can first test and validate that your application behavior is --- ## Legacy models ### Legacy models URL: https://learn.microsoft.com/en-us/azure/ai-foundry/openai/concepts/legacy-models?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Azure OpenAI in Microsoft Foundry Models retired models Feedback Summarize this article for me In this article Azure OpenAI offers a variety of models for different use cases. The following models are no longer available for deployment. Retired models These models are no longer available for new deployments. Model Deprecation date Retirement date Suggested replacement o1-preview July 28, 2025 o1 gpt-4.5-preview July 14, 2025 gpt-4.1 version: 2025-04-14 gpt-4o-realtime-preview - 2024-10-01 February 25, 2025 March 26, 2025 gpt-4o-realtime-preview (version 2024-12-17) or gpt-4o-mini-realtime-preview (version 2024-12-17) gpt-35-turbo - 0301 February 13, 2025 gpt-35-turbo (0125) gpt-4o-mini gpt-35-turbo - 0613 February 13, 2025 gpt-35-turbo (0125) gpt-4o-mini gpt-4 gpt-4-32k - 0314 June 6, 2025 gpt-4o version: 2024-11-20 gpt-4 gpt-4-32k - 0613 June 6, 2025 gpt-4o version: 2024-11-20 gpt-35-turbo-16k - 0613 April 30, 2025 gpt-4.1-mini version: 2025-04-14 babbage-002 January 27, 2025 davinci-002 January 27, 2025 dall-e-2 January 27, 2025 dalle-3 ada July 6, 2023 June 14, 2024 babbage July 6, 2023 June 14, 2024 curie July 6, 2023 June 14, 2024 davinci July 6, 2023 June 14, 2024 text-ada-001 July 6, 2023 June 14, 2024 gpt-35-turbo-instruct text-babbage-001 July 6, 2023 June 14, 2024 gpt-35-turbo-instruct text-curie-001 July 6, 2023 June 14, 2024 gpt-35-turbo-instruct text-davinci-002 July 6, 2023 June 14, 2024 gpt-35-turbo-instruct text-davinci-003 July 6, 2023 June 14, 2024 gpt-35-turbo-instruct code-cushman-001 July 6, 2023 June 14, 2024 gpt-35-turbo-instruct code-davinci-002 July 6, 2023 June 14, 2024 gpt-35-turbo-instruct text-simila --- ## Overview ### Overview URL: https://learn.microsoft.com/en-us/azure/ai-foundry/openai/concepts/model-router?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Model router for Microsoft Foundry Feedback Summarize this article for me In this article Note This document refers to the Microsoft Foundry (classic) portal. 🔄 Switch to the Microsoft Foundry (new) documentation if you're using the new portal. Note This document refers to the Microsoft Foundry (new) portal. Model router is a trained language model that intelligently routes your prompts in real time to the most suitable large language model (LLM). You deploy model router like any other Foundry model. Thus, it delivers high performance while saving on costs, reducing latencies, and increasing responsiveness, while maintaining comparable quality, all packaged as a single model deployment. Note You do not need to separately deploy the supported LLMs for use with model router, with the exception of the Claude models. To use model router with your Claude models, first deploy them from the model catalog. The deployments are invoked by model router if they're selected for routing. To try model router quickly, follow How to use model router . After you deploy model router, send a request to the deployment. Model router selects an underlying model for each request based on your routing settings. Tip The Microsoft Foundry (new) portal offers enhanced configuration options for model router. Switch to the Microsoft Foundry (new) documentation to see the latest features. How model router works As a trained language model, model router analyzes your prompts in real time based on complexity, reasoning, task type, and other attributes. It does not store your prompts. It routes only to eligible models based on your access and deployment types, h --- ### Overview URL: https://learn.microsoft.com/en-us/azure/ai-foundry/guardrails/guardrails-overview?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Guardrails and controls overview in Microsoft Foundry Feedback Summarize this article for me In this article Microsoft Foundry offers safety and security guardrails that can be applied to core models, including image generation models, and agents. Agent guardrails are in preview. Guardrails consist of a set of controls. The controls define a risk to be detected, intervention points to scan for the risk, and the response action to take in the model or agent when the risk is detected. For example, a risk detection could be the annotation of the risk or blocking the model or agent from further output. Risks are flagged via a set of classification models designed to detect and prevent the output of undesirable behavior and/or harmful content. Four intervention points are currently supported: user input, tool call (Preview), tool response (Preview), and output. Tool call and tool responses intervention points are applicable to agents only and scan the tool call made as well as content sent to the tool, and the output back from the tool, respectively. Variations in API configurations and application design might affect completions and thus filtering behavior. Important The guardrail system applies to all Models sold directly by Azure, except for prompts and completions processed by the audio models such as Whisper. For more information, see Audio models in Azure OpenAI . The guardrail system currently applies only to agents developed in the Foundry Agent Service, not to other agents registered in the Foundry Control Plane. Guardrails for agents vs models An individual Foundry guardrail can be applied to one or many models and one or m --- ### Overview URL: https://learn.microsoft.com/en-us/azure/ai-foundry/responsible-ai/openai/overview?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Overview of responsible AI practices for Azure OpenAI models Feedback Summarize this article for me In this article Important Non-English translations are provided for convenience only. Please consult the EN-US version of this document for the definitive version. Many of the Azure OpenAI models are generative AI models that demonstrate improvements in advanced capabilities such as content and code generation, summarization, and search. With many of these improvements also come increased responsible AI challenges related to harmful content, manipulation, human-like behavior, privacy, and more. For more information about the capabilities, limitations, and appropriate use cases for these models, review the Transparency Note . In addition to the Transparency Note, we provide technical recommendations and resources to help customers design, develop, deploy, and use AI systems that implement the Azure OpenAI models responsibly. Our recommendations are grounded in the Microsoft Responsible AI Standard , which sets policy requirements that our own engineering teams follow. Much of the content of the Standard follows a pattern, asking teams to identify, measure, and mitigate potential harms, and plan for how to operate the AI system. In alignment with those practices, these recommendations are organized into four stages: Identify : Identify and prioritize potential harms that could result from your AI system through iterative red-teaming, stress-testing, and analysis. Measure : Measure the frequency and severity of those harms by establishing clear metrics, creating measurement test sets, and completing iterative, systematic testing (bot --- ## What's new ### What's new URL: https://learn.microsoft.com/en-us/azure/ai-foundry/foundry-models/whats-new-model-router?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . What's new in model router in Microsoft Foundry Models? Feedback Summarize this article for me In this article This article provides a summary of the latest releases and major documentation updates for Azure model router. Note This document refers to the Microsoft Foundry (classic) portal. 🔄 Switch to the Microsoft Foundry (new) documentation if you're using the new portal. Note This document refers to the Microsoft Foundry (new) portal. November 2025 Anthropic models added Version 2025-11-18 of model router adds support for three Anthropic models: claude-haiku-4-5 , claude-opus-4-1 , and claude-sonnet-4-5 . To include these in your model router deployment, you need to first deploy them yourself to your Foundry resource (see Deploy and use Claude models ). Then enable them with model subset configuration in your model router deployment. Model router GA version A new model router model is now available. Version 2025-11-18 includes support for all underlying models in previous versions, as well as 10 new language models. It also includes new features that make it more versatile and effective. Routing profiles let you skew model router's choices to optimize for quality or cost while maintaining a baseline level of performance. Model router supports custom subsets : you can specify which underlying models to include in routing decisions. This gives you more control over cost, compliance, and performance characteristics. For more information on model router and its capabilities, see the Model router concepts guide . August 2025 New version of model router (preview) Model router now supports GPT-5 series models. Model router for Micro --- ## Get started with model router ### Get started with model router URL: https://learn.microsoft.com/en-us/azure/ai-foundry/openai/how-to/model-router?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Use model router for Microsoft Foundry Feedback Summarize this article for me In this article Note This document refers to the Microsoft Foundry (classic) portal. 🔄 Switch to the Microsoft Foundry (new) documentation if you're using the new portal. Note This document refers to the Microsoft Foundry (new) portal. Model router for Microsoft Foundry is a deployable AI chat model that selects the best large language model (LLM) to respond to a prompt in real time. It uses different preexisting models to deliver high performance and save on compute costs, all in one model deployment. To learn more about how model router works, its advantages, and limitations, see the Model router concepts guide . Use model router through the Chat Completions API like you'd use a single base model such as GPT-4. Follow the same steps as in the Chat completions guide . Tip The Microsoft Foundry (new) portal offers enhanced configuration options for model router. Switch to the Microsoft Foundry (new) documentation to see the latest features. Supported underlying models With the 2025-11-18 version, Model Router adds nine new models including Anthropic's Claude, DeepSeek, Llama, Grok models to support a total of 18 models available for routing your prompts. Note You don't need to separately deploy the supported LLMs for use with model router, with the exception of the Claude models. To use model router with your Claude models, first deploy them from the model catalog. The deployments will get invoked by Model router if they're selected for routing. Model router version Underlying models Underlying model version 2025-11-18 gpt-4.1 gpt-4.1-mini gpt-4.1-nano --- ## Responses API with Foundry Models ### Responses API with Foundry Models URL: https://learn.microsoft.com/en-us/azure/ai-foundry/foundry-models/how-to/generate-responses?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . How to generate text responses with Microsoft Foundry Models Feedback Summarize this article for me In this article Note This document refers to the Microsoft Foundry (classic) portal. 🔄 Switch to the Microsoft Foundry (new) documentation if you're using the new portal. Note This document refers to the Microsoft Foundry (new) portal. This article explains how to generate text responses for Foundry Models, such as Microsoft AI, DeepSeek, and Grok models, by using the Responses API. For a full list of the Foundry Models that support use of the Responses API, see Supported Foundry Models . Prerequisites To use the Responses API with deployed models in your application, you need: An Azure subscription. If you're using GitHub Models , you can upgrade your experience and create an Azure subscription in the process. Read Upgrade from GitHub Models to Microsoft Foundry Models if that's your case. A Foundry project. This kind of project is managed under a Foundry resource. If you don't have a Foundry project, see Create a project for Microsoft Foundry . Your Foundry project's endpoint URL, which is of the form https://YOUR-RESOURCE-NAME.services.ai.azure.com/api/projects/YOUR_PROJECT_NAME . A deployment of a Foundry Model, such as the MAI-DS-R1 model used in this article. If you don't have a deployment already, see Add and configure Foundry Models to a model deployment to your resource. Use the Responses API to generate text Use the code in this section to make Responses API calls for Foundry Models. In the code samples, you create the client to consume the model and then send it a basic request. Note Use keyless authentication with Micr --- ## Use blocklists ### Use blocklists URL: https://learn.microsoft.com/en-us/azure/ai-foundry/openai/how-to/use-blocklists?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . How to use block lists in Microsoft Foundry models Feedback Summarize this article for me In this article Note This document refers to the Microsoft Foundry (classic) portal. 🔄 Switch to the Microsoft Foundry (new) documentation if you're using the new portal. Note This document refers to the Microsoft Foundry (new) portal. The configurable Guardrails and controls available in Microsoft Foundry are sufficient for most content moderation needs. However, you might need to filter terms specific to your use case. For this, you can use custom block lists. The configurable content filters available in Azure OpenAI are sufficient for most content moderation needs. However, you might need to filter terms specific to your use case. For this, you can use custom block lists. Prerequisites An Azure subscription. Create one for free . Once you have your Azure subscription, create an Azure OpenAI resource in the Azure portal to get your token, key, and endpoint. Enter a unique name for your resource, select the subscription you entered on the application form, select a resource group, supported region, and supported pricing tier. Then select Create . The resource takes a few minutes to deploy. After it finishes, select go to resource . In the left pane, under Resource Management , select Subscription Key and Endpoint . The endpoint and either of the keys are used to call APIs. Azure CLI installed cURL installed Use block lists Azure OpenAI API Foundry You can create block lists with the Azure OpenAI API. The following steps help you get started. Get your token First, you need to get a token for accessing the APIs for creating, editing, and de --- ## Image generation ### Image generation URL: https://learn.microsoft.com/en-us/azure/ai-foundry/openai/how-to/dall-e?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . How to use Azure OpenAI image generation models Feedback Summarize this article for me In this article Note This document refers to the Microsoft Foundry (classic) portal. 🔄 Switch to the Microsoft Foundry (new) documentation if you're using the new portal. Note This document refers to the Microsoft Foundry (new) portal. OpenAI's image generation models create images from user-provided text prompts and optional images. This article explains how to use these models, configure options, and benefit from advanced image generation capabilities in Azure. Prerequisites An Azure subscription. You can create one for free . An Azure OpenAI resource created in a supported region. See Region availability . Deploy a dall-e-3 or gpt-image-1 -series model with your Azure OpenAI resource. For more information on deployments, see Create a resource and deploy a model with Azure OpenAI . GPT-image-1 series models are newer and feature a number of improvements over DALL-E 3. They are available in limited access: Apply for GPT-image-1 access ; Apply for GPT-image-1.5 access . Overview Use image generation via image generation API or responses API Experiment with image generation in the image playground Learn about image generation tokens Aspect GPT-Image-1.5 GPT-Image-1 GPT-Image-1-Mini DALL·E 3 Input / Output Modalities & Format Accepts text + image inputs; outputs images only in base64 (no URL option). Accepts text + image inputs; outputs images only in base64 (no URL option). Accepts text + image inputs; outputs images only in base64 (no URL option). Accepts text (primary) input; limited image editing inputs (with mask). Outputs as URL or base64 --- ## Image generation quickstart ### Image generation quickstart URL: https://learn.microsoft.com/en-us/azure/ai-foundry/openai/dall-e-quickstart?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Quickstart: Generate images with Azure OpenAI in Microsoft Foundry Models Feedback Summarize this article for me In this article Note This document refers to the Microsoft Foundry (classic) portal. 🔄 Switch to the Microsoft Foundry (new) documentation if you're using the new portal. Note This document refers to the Microsoft Foundry (new) portal. Use this guide to get started calling the Azure OpenAI in Microsoft Foundry Models image generation REST APIs by using Python. Prerequisites An Azure subscription. Create one for free . Python 3.8 or later version . The following Python libraries installed: os , requests , json . An Azure OpenAI resource created in a supported region. See Region availability . Then, you need to deploy a gpt-image-1 -series or dalle3 model with your Azure resource. For more information, see Create a resource and deploy a model with Azure OpenAI . Setup Retrieve key and endpoint To successfully call the Azure OpenAI APIs, you need the following information about your Azure OpenAI resource: Variable Name Value Endpoint api_base The endpoint value is located under Keys and Endpoint for your resource in the Azure portal. You can also find the endpoint via the Deployments page in Foundry portal. An example endpoint is: https://docs-test-001.openai.azure.com/ . Key api_key The key value is also located under Keys and Endpoint for your resource in the Azure portal. Azure generates two keys for your resource. You can use either value. Go to your resource in the Azure portal. On the navigation pane, select Keys and Endpoint under Resource Management . Copy the Endpoint value and an access key value. You can use e --- ## Video generation ### Video generation URL: https://learn.microsoft.com/en-us/azure/ai-foundry/openai/video-generation-quickstart?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Quickstart: Generate a video with Sora (preview) Feedback Summarize this article for me In this article Note This document refers to the Microsoft Foundry (classic) portal. 🔄 Switch to the Microsoft Foundry (new) documentation if you're using the new portal. Note This document refers to the Microsoft Foundry (new) portal. In this quickstart, you generate video clips using the Azure OpenAI service. The example uses the Sora model, which is a video generation model that creates realistic and imaginative video scenes from text instructions and/or image or video inputs. This guide shows you how to create a video generation job, poll for its status, and retrieve the generated video. For more information on video generation, see Video generation concepts . Prerequisites An Azure subscription. Create one for free . Python 3.8 or later version . We recommend using Python 3.10 or later, but having at least Python 3.8 is required. If you don't have a suitable version of Python installed, you can follow the instructions in the VS Code Python Tutorial for the easiest way of installing Python on your operating system. An Azure OpenAI resource created in one of the supported regions. For more information about region availability, see the models and versions documentation . Then, you need to deploy a sora model with your Azure OpenAI resource. For more information, see Create a resource and deploy a model with Azure OpenAI . Microsoft Entra ID prerequisites For the recommended keyless authentication with Microsoft Entra ID, you need to: Install the Azure CLI used for keyless authentication with Microsoft Entra ID. Assign the Cognitive Service --- ## Vision-enabled chats ### Vision-enabled chats URL: https://learn.microsoft.com/en-us/azure/ai-foundry/openai/how-to/gpt-with-vision?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Use vision-enabled chat models Feedback Summarize this article for me In this article Note This document refers to the Microsoft Foundry (classic) portal. 🔄 Switch to the Microsoft Foundry (new) documentation if you're using the new portal. Note This document refers to the Microsoft Foundry (new) portal. Vision-enabled chat models are large multimodal models (LMM) developed by OpenAI that can analyze images and provide textual responses to questions about them. They incorporate both natural language processing and visual understanding. The current vision-enabled models are the o-series reasoning models , GPT-5 series, GPT-4.1 series, GPT-4.5, GPT-4o series. The vision-enabled models can answer general questions about what's present in the images you upload. Tip To use vision-enabled models, you call the Chat Completion API on a supported model that you have deployed. If you're not familiar with the Chat Completion API, see the Vision-enabled chat how-to guide . Call the Chat Completion APIs The following command shows the most basic way to use a vision-enabled chat model with code. If this is your first time using these models programmatically, we recommend starting with our Chat with images quickstart . REST Python Send a POST request to https://{RESOURCE_NAME}.openai.azure.com/openai/v1/chat/completions where RESOURCE_NAME is the name of your Azure OpenAI resource Required headers : Content-Type : application/json api-key : {API_KEY} Body : The following is a sample request body. The format is the same as the chat completions API for GPT-4o, except that the message content can be an array containing text and images (either a v --- ### Vision-enabled chats URL: https://learn.microsoft.com/en-us/azure/ai-foundry/openai/gpt-v-quickstart?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Quickstart: Use images in your AI chats Feedback Summarize this article for me In this article Get started using images in your chats with Azure OpenAI in Microsoft Foundry Models. Note This document refers to the Microsoft Foundry (classic) portal. 🔄 Switch to the Microsoft Foundry (new) documentation if you're using the new portal. Note This document refers to the Microsoft Foundry (new) portal. Important Extra usage fees might apply when using chat completion models with vision functionality. Use this article to get started using Microsoft Foundry to deploy and test a chat completion model with image understanding. Prerequisites An Azure subscription. Create one for free . Once you have your Azure subscription, create an Azure OpenAI resource . For more information about resource creation, see the resource deployment guide . A Foundry project with your Azure OpenAI resource added as a connection. Prepare your media You need an image to complete this quickstart. You can use this sample image or any other image you have available. Go to Foundry Browse to Foundry and sign in with the credentials associated with your Azure OpenAI resource. During or after the sign-in workflow, select the appropriate directory, Azure subscription, and Azure OpenAI resource. Select the project you'd like to work in. On the left nav menu, select Models + endpoints and select + Deploy model . Choose an image-capable deployment by selecting model name: gpt-4o or gpt-4o-mini . In the window that appears, select a name and deployment type. Make sure your Azure OpenAI resource is connected. For more information about model deployment, see the resource de --- ## Image prompt engineering techniques ### Image prompt engineering techniques URL: https://learn.microsoft.com/en-us/azure/ai-foundry/openai/concepts/gpt-4-v-prompt-engineering?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Image prompt engineering techniques Feedback Summarize this article for me In this article Note This document refers to the Microsoft Foundry (classic) portal. 🔄 Switch to the Microsoft Foundry (new) documentation if you're using the new portal. Note This document refers to the Microsoft Foundry (new) portal. To unlock the full potential of vision-enabled chat models, it's essential to tailor the prompts to your specific needs. Here are some guidelines to enhance the accuracy and efficiency of your prompts. Fundamentals of writing an image prompt Contextual specificity: Adding context to the scenario at hand gives the model a better understanding of an appropriate output. This level of specificity aids in focusing on relevant aspects and avoiding extraneous details. Task-oriented prompts: Focusing on a specific task helps the model to develop the output while taking that perspective into consideration. Handle refusals: When the model indicates an inability to perform a task, refining the prompt can be an effective solution. More specific prompts can guide the model towards a clearer understanding and better execution of the task. Keep these tips in mind: Request explanations for generated responses to enhance transparency in the model's output If using a single-image prompt, place the image before the text Ask the model to describe the image in details first and complete your specific task from the description Add examples: Add examples that represent the type of responses you're looking for Break down requests: Try breaking down complex requests step-by-step to create manageable sub-goals Define output format: Clearly mention t --- ### Image prompt engineering techniques URL: https://learn.microsoft.com/en-us/azure/ai-foundry/openai/concepts/gpt-4-v-prompt-engineering?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Image prompt engineering techniques Feedback Summarize this article for me In this article Note This document refers to the Microsoft Foundry (classic) portal. 🔄 Switch to the Microsoft Foundry (new) documentation if you're using the new portal. Note This document refers to the Microsoft Foundry (new) portal. To unlock the full potential of vision-enabled chat models, it's essential to tailor the prompts to your specific needs. Here are some guidelines to enhance the accuracy and efficiency of your prompts. Fundamentals of writing an image prompt Contextual specificity: Adding context to the scenario at hand gives the model a better understanding of an appropriate output. This level of specificity aids in focusing on relevant aspects and avoiding extraneous details. Task-oriented prompts: Focusing on a specific task helps the model to develop the output while taking that perspective into consideration. Handle refusals: When the model indicates an inability to perform a task, refining the prompt can be an effective solution. More specific prompts can guide the model towards a clearer understanding and better execution of the task. Keep these tips in mind: Request explanations for generated responses to enhance transparency in the model's output If using a single-image prompt, place the image before the text Ask the model to describe the image in details first and complete your specific task from the description Add examples: Add examples that represent the type of responses you're looking for Break down requests: Try breaking down complex requests step-by-step to create manageable sub-goals Define output format: Clearly mention t --- ## Realtime API for speech and audio ### Realtime API for speech and audio URL: https://learn.microsoft.com/en-us/azure/ai-foundry/openai/how-to/realtime-audio?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Use the GPT Realtime API for speech and audio Feedback Summarize this article for me In this article Note This document refers to the Microsoft Foundry (classic) portal. 🔄 Switch to the Microsoft Foundry (new) documentation if you're using the new portal. Note This document refers to the Microsoft Foundry (new) portal. Azure OpenAI GPT Realtime API for speech and audio is part of the GPT-4o model family that supports low-latency, "speech in, speech out" conversational interactions. The GPT Realtime API is designed to handle real-time, low-latency conversational interactions. It's a great fit for use cases involving live interactions between a user and a model, such as customer support agents, voice assistants, and real-time translators. Most users of the Realtime API, including applications that use WebRTC or a telephony system, need to deliver and receive audio from an end-user in real time. The Realtime API isn't designed to connect directly to end user devices. It relies on client integrations to terminate end user audio streams. You can use the Realtime API via WebRTC, SIP, or WebSocket to send audio input to the model and receive audio responses in real time. In most cases, we recommend using the WebRTC API for low-latency real-time audio streaming. For more information, see: Realtime API via WebRTC Realtime API via SIP Realtime API via WebSockets Supported models The GPT real-time models are available for global deployments in East US 2 and Sweden Central regions . gpt-4o-mini-realtime-preview ( 2024-12-17 ) gpt-4o-realtime-preview ( 2024-12-17 ) gpt-realtime ( 2025-08-28 ) gpt-realtime-mini ( 2025-10-06 ) gpt-realtime-min --- ## Realtime API for speech and audio quickstart ### Realtime API for speech and audio quickstart URL: https://learn.microsoft.com/en-us/azure/ai-foundry/openai/realtime-audio-quickstart?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . GPT Realtime API for speech and audio Feedback Summarize this article for me In this article Note This document refers to the Microsoft Foundry (classic) portal. 🔄 Switch to the Microsoft Foundry (new) documentation if you're using the new portal. Note This document refers to the Microsoft Foundry (new) portal. Azure OpenAI GPT Realtime API for speech and audio is part of the GPT-4o model family that supports low-latency, "speech in, speech out" conversational interactions. You can use the Realtime API via WebRTC or WebSocket to send audio input to the model and receive audio responses in real time. Follow the instructions in this article to get started with the Realtime API via WebSockets. Use the Realtime API via WebSockets in server-to-server scenarios where low latency isn't a requirement. Tip In most cases, use the Realtime API via WebRTC for real-time audio streaming in client-side applications such as a web application or mobile app. WebRTC is designed for low-latency, real-time audio streaming and is the best choice for most scenarios. Supported models The GPT real-time models are available for global deployments. gpt-4o-realtime-preview (version 2024-12-17 ) gpt-4o-mini-realtime-preview (version 2024-12-17 ) gpt-realtime (version 2025-08-28 ) gpt-realtime-mini (version 2025-10-06 ) gpt-realtime-mini-2025-12-15 (version 2025-12-15 ) For more information, see the models and versions documentation . For more information, see the models and versions documentation . API support Support for the Realtime API was first added in API version 2024-10-01-preview (retired). Use version 2025-08-28 to access the latest Realtime API fe --- ## Realtime API via WebRTC ### Realtime API via WebRTC URL: https://learn.microsoft.com/en-us/azure/ai-foundry/openai/how-to/realtime-audio-webrtc?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Use the GPT Realtime API via WebRTC Feedback Summarize this article for me In this article Note This document refers to the Microsoft Foundry (classic) portal. 🔄 Switch to the Microsoft Foundry (new) documentation if you're using the new portal. Note This document refers to the Microsoft Foundry (new) portal. Azure OpenAI GPT Realtime API for speech and audio is part of the GPT-4o model family that supports low-latency, "speech in, speech out" conversational interactions. You can use the Realtime API via WebRTC, SIP, or WebSocket to send audio input to the model and receive audio responses in real time. Follow the instructions in this article to get started with the Realtime API via WebRTC. In most cases, use the WebRTC API for real-time audio streaming. The WebRTC API is a web standard that enables real-time communication (RTC) between browsers and mobile applications. Here are some reasons why WebRTC is preferred for real-time audio streaming: Lower latency : WebRTC is designed to minimize delay, making it more suitable for audio and video communication where low latency is critical for maintaining quality and synchronization. Media handling : WebRTC has built-in support for audio and video codecs, providing optimized handling of media streams. Error correction : WebRTC includes mechanisms for handling packet loss and jitter, which are essential for maintaining the quality of audio streams over unpredictable networks. Peer-to-peer communication : WebRTC allows direct communication between clients, reducing the need for a central server to relay audio data, which can further reduce latency. Use the Realtime API via WebSockets i --- ## Realtime API via WebSockets ### Realtime API via WebSockets URL: https://learn.microsoft.com/en-us/azure/ai-foundry/openai/how-to/realtime-audio-websockets?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Use the GPT Realtime API via WebSockets Feedback Summarize this article for me In this article Note This document refers to the Microsoft Foundry (classic) portal. 🔄 Switch to the Microsoft Foundry (new) documentation if you're using the new portal. Note This document refers to the Microsoft Foundry (new) portal. Azure OpenAI GPT Realtime API for speech and audio is part of the GPT-4o model family that supports low-latency, "speech in, speech out" conversational interactions. You can use the Realtime API via WebRTC, SIP, or WebSocket to send audio input to the model and receive audio responses in real time. Follow the instructions in this article to get started with the Realtime API via WebSockets. Use the Realtime API via WebSockets in server-to-server scenarios where low latency isn't a requirement. Tip In most cases, use the Realtime API via WebRTC for real-time audio streaming in client-side applications such as a web application or mobile app. WebRTC is designed for low-latency, real-time audio streaming and is the best choice for most scenarios. Prerequisites Before you can use GPT real-time audio, you need: An Azure subscription. Create one for free . A Microsoft Foundry resource. Create the resource in one of the supported regions . For setup steps, see Create a Microsoft Foundry resource . A deployment of the gpt-4o-realtime-preview , gpt-4o-mini-realtime-preview , gpt-realtime , gpt-realtime-mini , or gpt-realtime-mini-2025-12-15 model in a supported region as described in the supported models section in this article. In the Foundry portal, load your project. Select Build in the upper-right menu, then select the Models --- ## Realtime API via SIP ### Realtime API via SIP URL: https://learn.microsoft.com/en-us/azure/ai-foundry/openai/how-to/realtime-audio-sip?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Use the GPT Realtime API via SIP Feedback Summarize this article for me In this article Note This document refers to the Microsoft Foundry (classic) portal. 🔍 View the Microsoft Foundry (new) documentation to learn about the new portal. Azure OpenAI GPT Realtime API for speech and audio is part of the GPT-4o model family that supports low-latency, "speech in, speech out" conversational interactions. You can use the Realtime API via WebRTC, SIP, or WebSocket to send audio input to the model and receive audio responses in real time. Follow the instructions in this article to get started with the Realtime API via SIP. SIP is a protocol used to make phone calls over the internet. With SIP and the Realtime API you can direct incoming phone calls to the API. Supported models The GPT real-time models are available for global deployments in East US 2 and Sweden Central regions . gpt-4o-mini-realtime-preview (2024-12-17) gpt-4o-realtime-preview (2024-12-17) gpt-realtime (version 2025-08-28) gpt-realtime-mini (version 2025-10-06) gpt-realtime-mini-2025-12-15 (version 2025-12-15) Prerequisites Before you can use GPT real-time audio, you need: An Azure subscription - Create one for free . A Microsoft Foundry resource - Create a Microsoft Foundry resource in one of the supported regions . A deployment of the gpt-4o-realtime-preview , gpt-4o-mini-realtime-preview , gpt-realtime , gpt-realtime-mini , or gpt-realtime-mini-2025-12-15 model in a supported region as described in the supported models section in this article. In the Microsoft Foundry portal, load your project. Select Build in the upper right menu, then select the Models tab on the l --- ## Speech to text with Whisper ### Speech to text with Whisper URL: https://learn.microsoft.com/en-us/azure/ai-foundry/openai/whisper-quickstart?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Quickstart: Speech to text with the Azure OpenAI Whisper model Feedback Summarize this article for me In this article Note This document refers to the Microsoft Foundry (classic) portal. 🔄 Switch to the Microsoft Foundry (new) documentation if you're using the new portal. Note This document refers to the Microsoft Foundry (new) portal. In this quickstart, you use the Azure OpenAI Whisper model for speech to text conversion. The Whisper model can transcribe human speech in numerous languages, and it can also translate other languages into English. Prerequisites An Azure subscription - Create one for free . An Azure OpenAI resource with a speech to text model deployed in a supported region . For more information, see Create a resource and deploy a model with Azure OpenAI . Be sure that you are assigned at least the Cognitive Services Contributor role for the Azure OpenAI resource. A sample audio file. You can get sample audio, such as wikipediaOcelot.wav , from the Azure Speech in Foundry Tools SDK repository at GitHub . Setup Retrieve key and endpoint To successfully make a call against Azure OpenAI, you need an endpoint and a key . Variable name Value AZURE_OPENAI_ENDPOINT The service endpoint can be found in the Keys & Endpoint section when examining your resource from the Azure portal. Alternatively, you can find the endpoint via the Deployments page in Microsoft Foundry portal. An example endpoint is: https://docs-test-001.openai.azure.com/ . AZURE_OPENAI_API_KEY This value can be found in the Keys & Endpoint section when examining your resource from the Azure portal. You can use either KEY1 or KEY2 . Go to your resource in t --- ## Deploy Foundry Models in the portal ### Deploy Foundry Models in the portal URL: https://learn.microsoft.com/en-us/azure/ai-foundry/foundry-models/how-to/deploy-foundry-models?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Deploy Microsoft Foundry Models in the Foundry portal Feedback Summarize this article for me In this article Note This document refers to the Microsoft Foundry (classic) portal. 🔄 Switch to the Microsoft Foundry (new) documentation if you're using the new portal. Note This document refers to the Microsoft Foundry (new) portal. In this article, you learn how to use the Foundry portal to deploy a Foundry Model in a Foundry resource for use in performing inferencing tasks. Foundry Models include models such as Azure OpenAI models, Meta Llama models, and more. Once you deploy a Foundry Model, you can interact with it by using the Foundry Playground and inference it by using code. This article uses a Foundry Model from partners and community Llama-3.2-90B-Vision-Instruct for illustration. Models from partners and community require that you subscribe to Azure Marketplace before deployment. On the other hand, Foundry Models sold directly by Azure, such as Azure Open AI in Foundry Models, don't have this requirement. For more information about Foundry Models, including the regions where they're available for deployment, see Foundry Models sold directly by Azure and Foundry Models from partners and community . Prerequisites To complete this article, you need: An Azure subscription with a valid payment method. If you don't have an Azure subscription, create a paid Azure account to begin. If you're using GitHub Models , you can upgrade to Foundry Models and create an Azure subscription in the process. Access to Microsoft Foundry with appropriate permissions to create and manage resources. A Microsoft Foundry project . This kind of project --- ## Deploy Foundry Models using code ### Deploy Foundry Models using code URL: https://learn.microsoft.com/en-us/azure/ai-foundry/foundry-models/how-to/create-model-deployments?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Deploy models using Azure CLI and Bicep Feedback Summarize this article for me In this article Note This document refers to the Microsoft Foundry (classic) portal. 🔄 Switch to the Microsoft Foundry (new) documentation if you're using the new portal. Note This document refers to the Microsoft Foundry (new) portal. Important If you're currently using an Azure AI Inference beta SDK with Microsoft Foundry Models or Azure OpenAI service, we strongly recommend that you transition to the generally available OpenAI/v1 API , which uses an OpenAI stable SDK. For more information on how to migrate to the OpenAI/v1 API by using an SDK in your programming language of choice, see Migrate from Azure AI Inference SDK to OpenAI SDK . In this article, you'll learn how to add a new model deployment to a Foundry Models endpoint. The deployment is available for inference in your Foundry resource when you specify the deployment name in your requests. Prerequisites To complete this article, you need the following: An Azure subscription. If you're using GitHub Models , you can upgrade your experience and create an Azure subscription in the process. For more information, see Upgrade from GitHub Models to Foundry Models . A Foundry project. This project type is managed under a Foundry resource (formerly known as Azure AI Services resource). If you don't have a Foundry project, see Create a project for Microsoft Foundry . Azure role-based access control (RBAC) permissions to create and manage deployments. You need the Cognitive Services Contributor role or equivalent permissions for the Foundry resource. Foundry Models from partners and community require --- ## Deployment types ### Deployment types URL: https://learn.microsoft.com/en-us/azure/ai-foundry/foundry-models/concepts/deployment-types?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Deployment types for Microsoft Foundry Models Feedback Summarize this article for me In this article When you deploy a model in Microsoft Foundry, you choose a deployment type that determines: Where your data is processed (global, data zone, or single region) How you pay (pay-per-token or reserved capacity) Performance characteristics (latency variance, throughput limits) The service offers two main categories: standard (pay-per-token) and provisioned (reserved capacity). Within each category, you can choose global, data zone, or regional processing based on your compliance requirements. Important Data residency for all deployment types : Data stored at rest remains in the designated Azure geography. However, inferencing data is processed as follows: Global types: May be processed in any Azure region DataZone types: Processed only within the Microsoft-specified data zone (US or EU) Standard/Regional types: Processed in the deployment region Learn more about data residency . Deployment type comparison Deployment type SKU code Data processing Billing Best for Global Standard GlobalStandard Any Azure region Pay-per-token General workloads, highest quota Global Provisioned GlobalProvisionedManaged Any Azure region Reserved PTU Predictable high-throughput Global Batch GlobalBatch Any Azure region 50% discount, 24-hr Large async jobs Data Zone Standard DataZoneStandard Within data zone Pay-per-token EU/US data zone compliance Data Zone Provisioned DataZoneProvisionedManaged Within data zone Reserved PTU Data zone + predictable throughput Data Zone Batch DataZoneBatch Within data zone 50% discount Large async jobs with data zone Standa --- ## Upgrade GitHub Models to Foundry Models ### Upgrade GitHub Models to Foundry Models URL: https://learn.microsoft.com/en-us/azure/ai-foundry/foundry-models/how-to/quickstart-github-models?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Upgrade from GitHub Models to Microsoft Foundry Models Feedback Summarize this article for me In this article Note This document refers to the Microsoft Foundry (classic) portal. 🔄 Switch to the Microsoft Foundry (new) documentation if you're using the new portal. Note This document refers to the Microsoft Foundry (new) portal. In this article, you learn to develop a generative AI application by starting from GitHub Models and then upgrade your experience by deploying a Foundry Tools resource with Microsoft Foundry Models. GitHub Models are useful when you want to find and experiment with AI models for free as you develop a generative AI application. When you're ready to bring your application to production, upgrade your experience by deploying a Foundry Tools resource in an Azure subscription and start using Foundry Models. You don't need to change anything else in your code. The playground and free API usage for GitHub Models are rate limited by requests per minute, requests per day, tokens per request, and concurrent requests. If you get rate limited, you need to wait for the rate limit that you hit to reset before you can make more requests. Prerequisites To complete this tutorial, you need: A GitHub account with access to GitHub Models . An Azure subscription with a valid payment method. If you don't have an Azure subscription, create a paid Azure account to begin. Alternatively, you can wait until you're ready to deploy your model to production, at which point you'll be prompted to create or update your Azure account to a standard account. Foundry Models from partners and community require access to Azure Marketplace . Ens --- ## Azure OpenAI in Foundry Models ### Azure OpenAI in Foundry Models URL: https://learn.microsoft.com/en-us/azure/ai-foundry/openai/supported-languages?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Azure OpenAI supported programming languages Feedback Summarize this article for me In this article Note This document refers to the Microsoft Foundry (classic) portal. 🔄 Switch to the Microsoft Foundry (new) documentation if you're using the new portal. Note This document refers to the Microsoft Foundry (new) portal. Source code | Package (NuGet) Azure OpenAI API version support v1 Generally Available (GA) API now allows access to both GA and Preview operations. To learn more, see the API version lifecycle guide . Installation dotnet add package OpenAI Authentication Microsoft Entra ID API Key A secure, keyless authentication approach is to use Microsoft Entra ID (formerly Azure Active Directory) via the Azure Identity library . To use the library: dotnet add package Azure.Identity Use the desired credential type from the library. For example, DefaultAzureCredential : using Azure.Identity; using OpenAI; using OpenAI.Chat; using System.ClientModel.Primitives; #pragma warning disable OPENAI001 BearerTokenPolicy tokenPolicy = new( new DefaultAzureCredential(), "https://cognitiveservices.azure.com/.default"); ChatClient client = new( model: "gpt-4.1-nano", authenticationPolicy: tokenPolicy, options: new OpenAIClientOptions() { Endpoint = new Uri("https://YOUR-RESOURCE-NAME.openai.azure.com/openai/v1") } ); ChatCompletion completion = client.CompleteChat("Tell me about the bitter lesson.'"); Console.WriteLine($"[ASSISTANT]: {completion.Content[0].Text}"); For more information about Azure OpenAI keyless authentication, see the " Get started with the Azure OpenAI security building block " QuickStart article. using OpenAI; using OpenAI --- ## Claude in Foundry Models ### Claude in Foundry Models URL: https://learn.microsoft.com/en-us/azure/ai-foundry/foundry-models/how-to/use-foundry-models-claude?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Deploy and use Claude models in Microsoft Foundry (preview) Feedback Summarize this article for me In this article Note This document refers to the Microsoft Foundry (classic) portal. 🔄 Switch to the Microsoft Foundry (new) documentation if you're using the new portal. Note This document refers to the Microsoft Foundry (new) portal. Anthropic's Claude models bring advanced conversational AI capabilities to Microsoft Foundry, enabling you to build intelligent applications with state-of-the-art language understanding and generation. Claude models excel at complex reasoning, code generation, and multimodal tasks including image analysis. In this article, you learn how to: Deploy Claude models in Microsoft Foundry Authenticate by using Microsoft Entra ID or API keys Call the Claude Messages API from Python, JavaScript, or REST Choose the right Claude model for your use case. Claude models in Foundry include: Model family Models Claude Opus claude-opus-4-6 (preview), claude-opus-4-5 (preview), claude-opus-4-1 (preview) Claude Sonnet claude-sonnet-4-5 (preview) Claude Haiku claude-haiku-4-5 (preview) To learn more about the individual models, see Available Claude models . Important To use Claude models in Microsoft Foundry, you need a paid Azure subscription with a billing account in a country or region where Anthropic offers the models for purchase. The following paid subscription types are currently restricted: Cloud Solution Providers (CSP), sponsored accounts with Azure credits, enterprise accounts in Singapore and South Korea, and Microsoft accounts. For a list of common subscription-related errors, see Common error messages and --- ## Priority processing ### Priority processing URL: https://learn.microsoft.com/en-us/azure/ai-foundry/openai/concepts/priority-processing?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Priority processing for Microsoft Foundry models (preview) Feedback Summarize this article for me In this article Note This document refers to the Microsoft Foundry (classic) portal. 🔄 Switch to the Microsoft Foundry (new) documentation if you're using the new portal. Note This document refers to the Microsoft Foundry (new) portal. Important Priority processing is in preview and available by invitation only. Register here to be notified when it becomes more broadly available. This preview is provided without a service-level agreement and isn't recommended for production workloads. Certain features might not be supported or might have constrained capabilities. For more information, see Supplemental Terms of Use for Microsoft Azure Previews . Priority processing provides low-latency performance with the flexibility of pay-as-you-go. It operates on a pay-as-you-go token model, offering rapid response times without long-term contract commitments. This article covers the following topics: An overview of priority processing How to enable priority processing How to verify what service tier was used to process requests How to monitor costs Benefits Predictable low latency : Faster, more consistent token generation. Easy-to-use flexibility : Like standard pay-as-you-go processing, access priority processing on a flexible, pay-as-you-go basis instead of requiring provisioning and reservations in advance. Key use cases Consistent, low latency for responsive user experiences. Pay-as-you-go simplicity with no long-term commitments. Business-hour or bursty traffic that benefits from scalable, cost-efficient performance. Optionally, you can co --- ## Provisioned Throughput offering (PTU) ### Provisioned Throughput offering (PTU) URL: https://learn.microsoft.com/en-us/azure/ai-foundry/openai/concepts/provisioned-throughput?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . What is provisioned throughput for Foundry Models? Feedback Summarize this article for me In this article Note This document refers to the Microsoft Foundry (classic) portal. 🔄 Switch to the Microsoft Foundry (new) documentation if you're using the new portal. Note This document refers to the Microsoft Foundry (new) portal. Tip For more information on recent changes to the provisioned throughput offering, see the update article . The Microsoft Foundry provisioned throughput offering is a model deployment type that allows you to specify the amount of throughput you require in a model deployment. Foundry then allocates the necessary model processing capacity and ensures it's ready for you. Use the provisioned throughput you requested across a diverse portfolio of models that are sold directly by Azure . These models include Azure OpenAI models and newly introduced flagship model families like Azure DeepSeek, Azure Grok, Azure Llama, and more within Foundry Models. Provisioned throughput provides: Benefit Description Broader model choice Access to the latest flagship models Flexibility Switch models and deployments with given provisioned throughput quota Significant discounts Boost your reservation utilization with a more flexible reservation choice Predictable performance Stable max latency and throughput for uniform workloads Allocated processing capacity Throughput is available whether used or not once deployed Cost savings High throughput workloads might provide cost savings versus token-based consumption Tip Take advantage of more cost savings when you buy Microsoft Foundry Provisioned Throughput reservations . Provisioned thr --- ## Understanding and calculating PTU costs ### Understanding and calculating PTU costs URL: https://learn.microsoft.com/en-us/azure/ai-foundry/openai/how-to/provisioned-throughput-onboarding?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Understanding costs associated with provisioned throughput units (PTU) Feedback Summarize this article for me In this article Note This document refers to the Microsoft Foundry (classic) portal. 🔄 Switch to the Microsoft Foundry (new) documentation if you're using the new portal. Note This document refers to the Microsoft Foundry (new) portal. Use this article to learn about calculating and understanding costs associated with PTU. For an overview of the provisioned throughput offering, see What is provisioned throughput? . When you're ready to sign up for the provisioned throughput offering, see the getting started guide . Note In function calling and agent use cases, token usage can be variable. You should understand your expected Tokens Per Minute (TPM) usage in detail before migrating workloads to PTU. Provisioned throughput units Provisioned throughput units (PTUs) are generic units of model processing capacity that you can use to size provisioned deployments to achieve the required throughput for processing prompts and generating completions. Provisioned throughput units are granted to a subscription as quota. Each quota is specific to a region and defines the maximum number of PTUs that can be assigned to deployments in that subscription and region. Understanding provisioned throughput billing Microsoft Foundry Regional Provisioned Throughput , Data Zone Provisioned Throughput , and Global Provisioned Throughput are billed hourly based on the number of deployed PTUs, with substantial term discount available via the purchase of Azure reservations. The hourly billing model is useful for short-term deployment needs, such as v --- ## Get started with Provisioned Deployments ### Get started with Provisioned Deployments URL: https://learn.microsoft.com/en-us/azure/ai-foundry/openai/how-to/provisioned-get-started?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Get started using provisioned deployments on the Azure OpenAI in Microsoft Foundry Models Feedback Summarize this article for me In this article Note This document refers to the Microsoft Foundry (classic) portal. 🔄 Switch to the Microsoft Foundry (new) documentation if you're using the new portal. Note This document refers to the Microsoft Foundry (new) portal. The following guide walks you through key steps in creating a provisioned deployment with your Microsoft Foundry resource. For more details on the concepts discussed here, see: Foundry Provisioned Throughput Onboarding Guide Foundry Provisioned Throughput Concepts Prerequisites An Azure subscription - Create one for free Azure Contributor or Cognitive Services Contributor role Obtain/verify PTU quota availability Provisioned throughput deployments are sized in units called Provisioned Throughput Units (PTUs). PTU quota for each provisioned deployment type is granted to a subscription regionally and limits the total number of PTUs that can be deployed in that region across all models and versions. Creating a new deployment requires available (unused) quota to cover the desired size of the deployment. For example: If a subscription has the following in South Central US: Total PTU Quota = 500 PTUs Deployments: 100 PTUs: GPT-4o, 2024-05-13 100 PTUs: DeepSeek-R1, 1 Then 200 PTUs of quota are considered used, and there are 300 PTUs available for use to create new deployments. A default amount of global, data zone, and regional provisioned quota is assigned to eligible subscriptions in several regions. You can view the quota available to you in a region by visiting the Quotas p --- ## Provisioned spillover ### Provisioned spillover URL: https://learn.microsoft.com/en-us/azure/ai-foundry/openai/how-to/spillover-traffic-management?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Manage traffic with spillover for provisioned deployments Feedback Summarize this article for me In this article Note This document refers to the Microsoft Foundry (classic) portal. 🔄 Switch to the Microsoft Foundry (new) documentation if you're using the new portal. Note This document refers to the Microsoft Foundry (new) portal. This article describes how to manage traffic with spillover for provisioned deployments in Azure OpenAI. Spillover manages traffic fluctuations by routing overage traffic to a corresponding standard deployment. This optional capability can be set for all requests on a deployment or managed on a per-request basis, helping you reduce disruptions during traffic bursts. Prerequisites An Azure subscription. Create one for free . A provisioned managed deployment and a standard deployment in the same Azure OpenAI resource The data processing level of your standard deployment must match your provisioned deployment. For example, use a global provisioned deployment with a global standard spillover deployment. Azure CLI installed for REST API examples, or access to the Foundry portal The AZURE_OPENAI_ENDPOINT environment variable set to your Azure OpenAI endpoint URL Cognitive Services Contributor role or higher on the Azure OpenAI resource to create or modify deployments Enable spillover for all requests on a provisioned deployment Foundry portal REST API To deploy a model with the spillover capability, go to the Foundry portal . On the left navigation menu, then select Deployments . Select Deploy model . In the menu that appears, select Customize . Specify one of the provisioned options as the Deployment type , --- ## Chat completions API ### Chat completions API URL: https://learn.microsoft.com/en-us/azure/ai-foundry/openai/how-to/chatgpt?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Work with chat completions models Feedback Summarize this article for me In this article Chat models are language models that are optimized for conversational interfaces. The models behave differently than the older completion API models. Previous models were text-in and text-out, which means they accepted a prompt string and returned a completion to append to the prompt. However, the latest models are conversation-in and message-out. The models expect input formatted in a specific chat-like transcript format. They return a completion that represents a model-written message in the chat. This format was designed specifically for multi-turn conversations, but it can also work well for nonchat scenarios. This article walks you through getting started with chat completions models. To get the best results, use the techniques described here. Don't try to interact with the models the same way you did with the older model series because the models are often verbose and provide less useful responses. Work with chat completion models The following code snippet shows the most basic way to interact with models that use the Chat Completion API. Note The responses API uses the same chat style of interaction, but supports the latest features which are not supported with the older chat completions API. Microsoft Entra ID API Key from openai import OpenAI from azure.identity import DefaultAzureCredential, get_bearer_token_provider token_provider = get_bearer_token_provider( DefaultAzureCredential(), "https://cognitiveservices.azure.com/.default" ) client = OpenAI( base_url = "https://YOUR-RESOURCE-NAME.openai.azure.com/openai/v1/", api_key=token --- ## Responses API ### Responses API URL: https://learn.microsoft.com/en-us/azure/ai-foundry/openai/how-to/responses?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Azure OpenAI Responses API Feedback Summarize this article for me In this article The Responses API is a new stateful API from Azure OpenAI. It brings together the best capabilities from the chat completions and assistants API in one unified experience. The Responses API also adds support for the new computer-use-preview model which powers the Computer use capability. Getting started with the responses API To access the responses API commands, you need to upgrade your version of the OpenAI library. pip install --upgrade openai Generate a text response Python (API Key) Python (Microsoft Entra ID) REST API Output import os from openai import OpenAI client = OpenAI( api_key=os.getenv("AZURE_OPENAI_API_KEY"), base_url="https://YOUR-RESOURCE-NAME.openai.azure.com/openai/v1/", ) response = client.responses.create( model="gpt-4.1-nano", # Replace with your model deployment name input="This is a test.", ) print(response.model_dump_json(indent=2)) Important Use API keys with caution. Don't include the API key directly in your code, and never post it publicly. If you use an API key, store it securely in Azure Key Vault. For more information about using API keys securely in your apps, see API keys with Azure Key Vault . For more information about AI services security, see Authenticate requests to Azure AI services . from openai import OpenAI from azure.identity import DefaultAzureCredential, get_bearer_token_provider token_provider = get_bearer_token_provider( DefaultAzureCredential(), "https://cognitiveservices.azure.com/.default" ) client = OpenAI( base_url = "https://YOUR-RESOURCE-NAME.openai.azure.com/openai/v1/", api_key=token_provide --- ## Reasoning models ### Reasoning models URL: https://learn.microsoft.com/en-us/azure/ai-foundry/openai/how-to/reasoning?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Azure OpenAI reasoning models Feedback Summarize this article for me In this article Azure OpenAI reasoning models are designed to tackle reasoning and problem-solving tasks with increased focus and capability. These models spend more time processing and understanding the user's request, making them exceptionally strong in areas like science, coding, and math compared to previous iterations. Key capabilities of reasoning models: Complex Code Generation: Capable of generating algorithms and handling advanced coding tasks to support developers. Advanced Problem Solving: Ideal for comprehensive brainstorming sessions and addressing multifaceted challenges. Complex Document Comparison: Perfect for analyzing contracts, case files, or legal documents to identify subtle differences. Instruction Following and Workflow Management: Particularly effective for managing workflows requiring shorter contexts. Usage These models don't currently support the same set of parameters as other models that use the chat completions API. Chat completions API C# Python REST Output using Azure.Identity; using OpenAI; using OpenAI.Chat; using System.ClientModel.Primitives; #pragma warning disable OPENAI001 //currently required for token based authentication BearerTokenPolicy tokenPolicy = new( new DefaultAzureCredential(), "https://cognitiveservices.azure.com/.default"); ChatClient client = new( model: "o4-mini", authenticationPolicy: tokenPolicy, options: new OpenAIClientOptions() { Endpoint = new Uri("https://YOUR-RESOURCE-NAME.openai.azure.com/openai/v1") } ); ChatCompletionOptions options = new ChatCompletionOptions { MaxOutputTokenCount = 100000 }; Ch --- ## Batch processing ### Batch processing URL: https://learn.microsoft.com/en-us/azure/ai-foundry/openai/how-to/batch?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Getting started with Azure OpenAI batch deployments Feedback Summarize this article for me In this article The Azure OpenAI Batch API is designed to handle large-scale and high-volume processing tasks efficiently. Process asynchronous groups of requests with separate quota, with 24-hour target turnaround, at 50% less cost than global standard . With batch processing, rather than send one request at a time you send a large number of requests in a single file. Global batch requests have a separate enqueued token quota avoiding any disruption of your online workloads. Key use cases include: Large-Scale Data Processing: Quickly analyze extensive datasets in parallel. Content Generation: Create large volumes of text, such as product descriptions or articles. Document Review and Summarization: Automate the review and summarization of lengthy documents. Customer Support Automation: Handle numerous queries simultaneously for faster responses. Data Extraction and Analysis: Extract and analyze information from vast amounts of unstructured data. Natural Language Processing (NLP) Tasks: Perform tasks like sentiment analysis or translation on large datasets. Marketing and Personalization: Generate personalized content and recommendations at scale. Tip If your batch jobs are so large that you are hitting the enqueued token limit even after maxing out the quota for your deployment, certain regions now support a new feature that allows you to queue multiple batch jobs with exponential backoff. Once your enqueued token quota is available, the next batch job can be created and kicked off automatically. To learn more, see automating retries of lar --- ## Deep research ### Deep research URL: https://learn.microsoft.com/en-us/azure/ai-foundry/openai/how-to/deep-research?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Deep research Feedback Summarize this article for me In this article The o3-deep-research model is designed for advanced research tasks. It can browse, analyze, and synthesize information from hundreds of sources to produce a comprehensive, citation-rich report. This model uses multi-step reasoning, web search, and remote Model Context Protocol (MCP) servers to gather and process data. It can also run code for complex analysis. Use deep research when you need: Legal or scientific research Market and competitive analysis Reports based on large sets of internal or public data To start, call the Responses API with the model set to o3-deep-research . Include at least one data source: web search or a remote MCP server. Optionally, add the code interpreter tool for advanced analysis. Start a deep research task curl https://YOUR-RESOURCE-NAME.openai.azure.com/openai/v1/responses \ -H "Content-Type: application/json" \ -H "Authorization: Bearer $AZURE_OPENAI_AUTH_TOKEN" \ -d '{ "model": "o3-deep-research", "background": true, "tools": [ { "type": "web_search_preview" }, { "type": "code_interpreter", "container": {"type": "auto"} } ], "input": "Research the economic impact of semaglutide on global healthcare systems. Include specific figures, trends, statistics, and measurable outcomes. Prioritize reliable, up-to-date sources: peer-reviewed research, health organizations (e.g., WHO, CDC), regulatory agencies, or pharmaceutical earnings reports. Include inline citations and return all source metadata. Be analytical, avoid generalities, and ensure that each section supports data-backed reasoning that could inform healthcare policy or finan --- ## Structured outputs ### Structured outputs URL: https://learn.microsoft.com/en-us/azure/ai-foundry/openai/how-to/structured-outputs?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Structured outputs Feedback Summarize this article for me In this article Structured outputs make a model follow a JSON Schema definition that you provide as part of your inference API call. This is in contrast to the older JSON mode feature, which guaranteed valid JSON would be generated, but was unable to ensure strict adherence to the supplied schema. Structured outputs are recommended for function calling, extracting structured data, and building complex multi-step workflows. Getting started Python (Microsoft Entra ID) Python (key-based auth) You can use Pydantic to define object schemas in Python. Depending on what version of the OpenAI and Pydantic libraries you're running you might need to upgrade to a newer version. These examples were tested against openai 1.42.0 and pydantic 2.8.2 . pip install openai pydantic --upgrade If you are new to using Microsoft Entra ID for authentication see How to configure Azure OpenAI in Microsoft Foundry Models with Microsoft Entra ID authentication . import os from pydantic import BaseModel from openai import OpenAI from azure.identity import DefaultAzureCredential, get_bearer_token_provider token_provider = get_bearer_token_provider( DefaultAzureCredential(), "https://cognitiveservices.azure.com/.default" ) client = OpenAI( base_url = "https://YOUR-RESOURCE-NAME.openai.azure.com/openai/v1/", api_key=token_provider, ) class CalendarEvent(BaseModel): name: str date: str participants: list[str] completion = client.beta.chat.completions.parse( model="MODEL_DEPLOYMENT_NAME", # replace with the model deployment name of your gpt-4o 2024-08-06 deployment messages=[ {"role": "system", "content": --- ## JSON mode ### JSON mode URL: https://learn.microsoft.com/en-us/azure/ai-foundry/openai/how-to/json-mode?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Learn how to use JSON mode Feedback Summarize this article for me In this article JSON mode allows you to set the models response format to return a valid JSON object as part of a chat completion. While generating valid JSON was possible previously, there could be issues with response consistency that would lead to invalid JSON objects being generated. Note While JSON mode is still supported, when possible we recommend using structured outputs . Like JSON mode structured outputs generate valid JSON, but with the added benefit that you can constrain the model to use a specific JSON schema. Note Currently Structured outputs are not supported on bring your own data scenario. JSON mode support JSON mode is only currently supported with the following models: API support Support for JSON mode was first added in API version 2023-12-01-preview Example Python PowerShell import os from openai import OpenAI client = OpenAI( base_url="https://YOUR-RESOURCE-NAME.openai.azure.com/openai/v1/", api_key=os.getenv("AZURE_OPENAI_API_KEY") ) response = client.chat.completions.create( model="YOUR-MODEL_DEPLOYMENT_NAME", # Model = should match the deployment name you chose for your model deployment response_format={ "type": "json_object" }, messages=[ {"role": "system", "content": "You are a helpful assistant designed to output JSON."}, {"role": "user", "content": "Who won the world series in 2020?"} ] ) print(response.choices[0].message.content) Output { "winner": "Los Angeles Dodgers", "event": "World Series", "year": 2020 } In this example, the user asks for historical information in a specific JSON schema because they plan to use the output for f --- ## Web search ### Web search URL: https://learn.microsoft.com/en-us/azure/ai-foundry/openai/how-to/web-search?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Web search (preview) Feedback Summarize this article for me In this article Web search enables models to retrieve and ground responses with real-time information from the public web before generating output. When enabled, the model can return up-to-date answers with inline citations. Web search is available via the web_search_preview tool in the Responses API . Note Some SDKs may expose both web_search_preview and web_search tool types. Only web_search_preview is currently supported for Web search in the Azure OpenAI Responses API. The web_search tool type is not supported at this time and should not be used. Important Web Search (preview) uses Grounding with Bing Search and/or Grounding with Bing Custom Search, which are First Party Consumption Services governed by these Grounding with Bing terms of use and the Microsoft Privacy Statement . The Microsoft Data Protection Addendum does not apply to data sent to Grounding with Bing Search and/or Grounding with Bing Custom Search. When Customer uses Grounding with Bing Search and/or Grounding with Bing Custom Search, Customer Data will flow outside Customer’s compliance and Geo boundary. Use of Grounding with Bing Search and Grounding with Bing Custom Search will incur costs; learn more about pricing . Learn more about how Azure admins can manage access to the use of Web Search (preview). Options to use web search Web search supports three modes. Choose the mode based on the depth and speed you need. Web search without reasoning The model forwards the user query directly to the web search tool and uses top-ranked sources to ground the response. There's no multi-step planning. This --- ## Predicted outputs ### Predicted outputs URL: https://learn.microsoft.com/en-us/azure/ai-foundry/openai/how-to/predicted-outputs?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Predicted outputs (preview) Feedback Summarize this article for me In this article Predicted outputs can improve model response latency for chat completions calls where minimal changes are needed to a larger body of text. If you're asking the model to provide a response where a large portion of the expected response is already known, predicted outputs can significantly reduce the latency of this request. This capability is particularly well-suited for coding scenarios, including autocomplete, error detection, and real-time editing, where speed and responsiveness are critical for developers and end-users. Rather than have the model regenerate all the text from scratch, you can indicate to the model that most of the response is already known by passing the known text to the prediction parameter. Model support gpt-4o-mini version: 2024-07-18 gpt-4o version: 2024-08-06 gpt-4o version: 2024-11-20 gpt-4.1 version: 2025-04-14 gpt-4.1-nano version: 2025-04-14 gpt-4.1-mini version: 2025-04-14 API support First introduced in 2025-01-01-preview . Supported in all subsequent releases. Unsupported features Predicted outputs is currently text-only. These features can't be used in conjunction with the prediction parameter and predicted outputs. Tools/Function calling audio models/inputs and outputs n values higher than 1 logprobs presence_penalty values greater than 0 frequency_penalty values greater than 0 max_completion_tokens Note The predicted outputs feature is currently unavailable for models in the South East Asia region. Getting started To demonstrate the basics of predicted outputs, we'll start by asking a model to refactor the code f --- ## Prompt caching ### Prompt caching URL: https://learn.microsoft.com/en-us/azure/ai-foundry/openai/how-to/prompt-caching?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Prompt caching Feedback Summarize this article for me In this article Prompt caching allows you to reduce overall request latency and cost for longer prompts that have identical content at the beginning of the prompt. "Prompt" in this context is referring to the input you send to the model as part of your chat completions request. Rather than reprocess the same input tokens over and over again, the service is able to retain a temporary cache of processed input token computations to improve overall performance. Prompt caching has no impact on the output content returned in the model response beyond a reduction in latency and cost. For supported models, cached tokens are billed at a discount on input token pricing for Standard deployment types and up to 100% discount on input tokens for Provisioned deployment types. Azure AI Foundry Model prompt caches are cleared within 24 hours. Prompt caches aren't shared between Azure subscriptions. Supported models Prompt caching is supported with all Azure OpenAI models GPT-4o or newer. Prompt caching applies to models that have chat-completion, completion, responses, or real-time operations. For models which don't have these operations, this feature isn't available. Getting started For a request to take advantage of prompt caching the request must be both: A minimum of 1,024 tokens in length. The first 1,024 tokens in the prompt must be identical. Requests are routed based on a hash of the initial prefix of a prompt. The hash typically uses the first 256 tokens, though the exact length varies depending on the model. When a match is found between the token computations in a prompt and the cu --- ## Embeddings ### Embeddings URL: https://learn.microsoft.com/en-us/azure/ai-foundry/openai/how-to/embeddings?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Learn how to generate embeddings Feedback Summarize this article for me In this article An embedding is a special format of data representation that can be easily utilized by machine learning models and algorithms. The embedding is an information dense representation of the semantic meaning of a piece of text. Each embedding is a vector of floating point numbers, such that the distance between two embeddings in the vector space is correlated with semantic similarity between two inputs in the original format. For example, if two texts are similar, then their vector representations should also be similar. Embeddings power vector similarity search in Azure Databases such as Azure Cosmos DB for NoSQL , Azure Cosmos DB for MongoDB vCore , Azure SQL Database or Azure Database for PostgreSQL - Flexible Server . How to get embeddings To obtain an embedding vector for a piece of text, make a request to the embeddings endpoint as shown in the following code snippets: Note The Azure OpenAI embeddings API does not currently support Microsoft Entra ID with the v1 API. C# Go JavaScript Python PowerShell REST using OpenAI; using OpenAI.Embeddings; using System.ClientModel; EmbeddingClient client = new( "text-embedding-3-small", credential: new ApiKeyCredential("API-KEY"), options: new OpenAIClientOptions() { Endpoint = new Uri("https://YOUR-RESOURCE-NAME.openai.azure.com/openai/v1") } ); string input = "This is a test"; OpenAIEmbedding embedding = client.GenerateEmbedding(input); ReadOnlyMemory vector = embedding.ToFloats(); Console.WriteLine($"Embeddings: [{string.Join(", ", vector.ToArray())}]"); package main import ( "context" "fmt" --- ### Embeddings URL: https://learn.microsoft.com/en-us/azure/ai-foundry/openai/latest#create-embedding?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Azure OpenAI in Microsoft Foundry Models v1 REST API reference Feedback Summarize this article for me In this article Only a subset of operations are currently supported with the v1 API. To learn more, see the API version lifecycle guide . v1 OpenAPI 3.0 spec Create chat completion POST {endpoint}/openai/v1/chat/completions Creates a chat completion. Parameters Name In Required Type Description endpoint path Yes string url Supported Azure OpenAI endpoints (protocol and hostname, for example: https://aoairesource.openai.azure.com . Replace "aoairesource" with your Azure OpenAI resource name). https://{your-resource-name}.openai.azure.com api-version query No The explicit Microsoft Foundry Models API version to use for this request. v1 if not otherwise specified. Request Header Use either token based authentication or API key. Authenticating with token based authentication is recommended and more secure. Name Required Type Description Authorization True string Example: Authorization: Bearer {Azure_OpenAI_Auth_Token} To generate an auth token using Azure CLI: az account get-access-token --resource https://cognitiveservices.azure.com Type: oauth2 Authorization Url: https://login.microsoftonline.com/common/oauth2/v2.0/authorize scope: https://cognitiveservices.azure.com/.default api-key True string Provide Azure OpenAI API key here Request Body Content-Type : application/json Name Type Description Required Default audio object Parameters for audio output. Required when audio output is requested with modalities: ["audio"] . No └─ format enum Specifies the output audio format. Must be one of wav , mp3 , flac , opus , or pcm16 . Possibl --- ## Embeddings tutorial ### Embeddings tutorial URL: https://learn.microsoft.com/en-us/azure/ai-foundry/openai/tutorials/embeddings?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Tutorial: Explore Azure OpenAI in Microsoft Foundry Models embeddings and document search Feedback Summarize this article for me In this article This tutorial will walk you through using the Azure OpenAI embeddings API to perform document search where you'll query a knowledge base to find the most relevant document. In this tutorial, you learn how to: Download a sample dataset and prepare it for analysis. Create environment variables for your resources endpoint and API key. Use one of the following models: text-embedding-ada-002 (Version 2), text-embedding-3-large, text-embedding-3-small models. Use cosine similarity to rank search results. Prerequisites An Azure subscription - Create one for free A Microsoft Foundry or Azure OpenAI resource with the text-embedding-ada-002 (Version 2) model deployed. This model is currently only available in certain regions . Python 3.10 or later version The following Python libraries: openai , num2words , matplotlib , plotly , scipy , scikit-learn , pandas , tiktoken . Jupyter Notebooks Set up Python libraries If you haven't already, you need to install the following libraries: pip install openai num2words matplotlib plotly scipy scikit-learn pandas tiktoken Download the BillSum dataset BillSum is a dataset of United States Congressional and California state bills. For illustration purposes, we'll look only at the US bills. The corpus consists of bills from the 103rd-115th (1993-2018) sessions of Congress. The data was split into 18,949 train bills and 3,269 test bills. The BillSum corpus focuses on mid-length legislation from 5,000 to 20,000 characters in length. More information on the projec --- ## Audio generation ### Audio generation URL: https://learn.microsoft.com/en-us/azure/ai-foundry/openai/audio-completions-quickstart?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Quickstart: Get started with Azure OpenAI audio generation Feedback Summarize this article for me In this article Note This document refers to the Microsoft Foundry (classic) portal. 🔄 Switch to the Microsoft Foundry (new) documentation if you're using the new portal. Note This document refers to the Microsoft Foundry (new) portal. The gpt-4o-audio-preview and gpt-4o-mini-audio-preview models introduce the audio modality into the existing /chat/completions API. The audio model expands the potential for AI applications in text and voice-based interactions and audio analysis. Modalities supported in gpt-4o-audio-preview and gpt-4o-mini-audio-preview models include: text, audio, and text + audio. Here's a table of the supported modalities with example use cases: Modality input Modality output Example use case Text Text + audio Text to speech, audio book generation Audio Text + audio Audio transcription, audio book generation Audio Text Audio transcription Text + audio Text + audio Audio book generation Text + audio Text Audio transcription By using audio generation capabilities, you can achieve more dynamic and interactive AI applications. Models that support audio inputs and outputs allow you to generate spoken audio responses to prompts and use audio inputs to prompt the model. Supported models Currently only gpt-4o-audio-preview and gpt-4o-mini-audio-preview version: 2024-12-17 supports audio generation. For more information about region availability, see the models and versions documentation . Currently the following voices are supported for audio out: Alloy, Echo, and Shimmer. The maximum audio file size is 20 MB. Note The Rea --- ## Speech to text ### Speech to text URL: https://learn.microsoft.com/en-us/azure/ai-services/speech-service/get-started-speech-to-text?context=/azure/ai-foundry/context/context?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Quickstart: Recognize and convert speech to text Feedback Summarize this article for me In this article In this quickstart, you try real-time speech to text in Microsoft Foundry . Prerequisites An Azure subscription. Create one for free . A Foundry project. If you need to create a project, see Create a Microsoft Foundry project . Try real-time speech to text Foundry (new) portal Foundry (classic) portal Sign in to Microsoft Foundry . Make sure the New Foundry toggle is on. These steps refer to Foundry (new) . Select Build from the top right menu. Select Models on the left pane. The AI Services tab shows the Foundry models that can be used out of the box in the Foundry portal. Select Azure Speech - Speech to text to open the Speech to Text playground. Optionally use the Parameters section to change the task, language, profanity policy, and other settings. You can also add special instructions for the LLM. Use the Upload files section to select your audio file. Then select Start . View the transcription output in the Transcript tab. Optionally view the raw API response output in the JSON tab. Switch to the Code tab to get the sample code for using the speech to text feature in your application. Other Foundry (new) features The following Speech features are available in the Foundry (new) portal: Speech MCP server Speech to text quickstart Text to speech quickstart Text to speech avatar quickstart Voice live quickstart Sign in to Microsoft Foundry . Make sure the New Foundry toggle is off. These steps refer to Foundry (classic) . Select Playgrounds from the left pane and then select a playground to use. In this example, select Try t --- ## Text to speech ### Text to speech URL: https://learn.microsoft.com/en-us/azure/ai-services/speech-service/get-started-text-to-speech?context=/azure/ai-foundry/context/context?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Quickstart: Convert text to speech Feedback Summarize this article for me In this article In this quickstart, try out the text to speech model from Azure Speech in Foundry Tools, using Microsoft Foundry . Prerequisites An Azure subscription. Create one for free . A Foundry project. If you need to create a project, see Create a Microsoft Foundry project . Try text to speech Try text to speech in the Foundry portal by following these steps: Foundry (new) portal Foundry (classic) portal Sign in to Microsoft Foundry . Make sure the New Foundry toggle is on. These steps refer to Foundry (new) . Select Build from the top right menu. Select Models on the left pane. The AI Services tab shows the Azure AI models that can be used out of the box in the Foundry portal. Select Azure Speech - Text to Speech to open the Text to Speech playground. Choose a prebuilt voice from the dropdown menu, and optionally tune it with the provider parameter sliders. Enter your sample text in the text box. Select Play to hear the synthetic voice read your text. Other Foundry (new) features The following Speech features are available in the Foundry (new) portal: Speech MCP server Speech to text quickstart Text to speech quickstart Text to speech avatar quickstart Voice live quickstart Sign in to Microsoft Foundry . Make sure the New Foundry toggle is off. These steps refer to Foundry (classic) . Select Playgrounds from the left pane and then select a playground to use. In this example, select Try the Speech playground . Select Voice gallery . Select a voice from the gallery. Optionally filter voices by keyword or supported languages. On the right pane, select --- ## Text to speech avatar ### Text to speech avatar URL: https://learn.microsoft.com/en-us/azure/ai-services/speech-service/text-to-speech-avatar/batch-synthesis-avatar?context=/azure/ai-foundry/context/context?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Use batch synthesis for text to speech avatar Feedback Summarize this article for me In this article The batch synthesis API for text to speech avatar lets you synthesize text asynchronously into a talking avatar as a video file. Publishers and video content platforms can use this API to create avatar video content in a batch. That approach can be suitable for different use cases like training materials, presentations, or advertisements. The synthetic avatar video will be generated asynchronously after the system receives text input. The generated video output can be downloaded in batch mode synthesis. You submit text for synthesis, poll for the synthesis status, and download the video output when the status shows success. The text input formats must be plain text or Speech Synthesis Markup Language (SSML) text. This diagram provides a high-level overview of the workflow. Try out the text to speech avatar feature in Microsoft Foundry . Prerequisites An Azure subscription. A Foundry project. If you need to create a project, see Create a Microsoft Foundry project . Try text to speech avatar Try text to speech in the Foundry portal by following these steps: Go to Microsoft Foundry . Select Build from the top right menu. Select Models on the left pane. The AI Services tab shows the Azure AI models that can be used out of the box in the Foundry portal. Select Azure Speech - Text to Speech Avatar to open the Text to Speech Avatar playground. Choose a prebuilt avatar from the grid, and select a voice from the Voice dropdown menu. Enter your sample text in the text box on the right. Select Play to hear the synthetic voice read your text --- ## Voice Live ### Voice Live URL: https://learn.microsoft.com/en-us/azure/ai-services/speech-service/voice-live-quickstart?context=/azure/ai-foundry/context/context?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Quickstart: Create a Voice Live real-time voice agent Feedback Summarize this article for me In this article In this article, you learn how to use Voice Live with generative AI and Azure Speech in Foundry Tools in the Microsoft Foundry portal . You create and run an application to use Voice Live directly with generative AI models for real-time voice agents. Using models directly allows specifying custom instructions (prompts) for each session, offering more flexibility for dynamic or experimental use cases. Models may be preferable when you want fine-grained control over session parameters or need to frequently adjust the prompt or configuration without updating an agent in the portal. The code for model-based sessions is simpler in some respects, as it does not require managing agent IDs or agent-specific setup. Direct model use is suitable for scenarios where agent-level abstraction or built-in logic is unnecessary. To instead use the Voice Live API with agents, see the Voice Live API agents quickstart . Prerequisites An Azure subscription. Create one for free . A Foundry project. If you need to create a project, see Create a Microsoft Foundry project . For more information about region availability, see the Voice Live overview documentation . Tip To use Voice Live, you don't need to deploy an audio model with your Microsoft Foundry resource. Voice Live is fully managed, and the model is automatically deployed for you. For more information about models availability, see the Voice Live overview documentation . Try out Voice Live in the Speech playground Foundry (new) portal Foundry (classic) portal To try out the Voice Live dem --- ## Image prompt transformation ### Image prompt transformation URL: https://learn.microsoft.com/en-us/azure/ai-foundry/openai/concepts/prompt-transformation?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Prompt transformation Feedback Summarize this article for me In this article Note This document refers to the Microsoft Foundry (classic) portal. 🔄 Switch to the Microsoft Foundry (new) documentation if you're using the new portal. Note This document refers to the Microsoft Foundry (new) portal. Prompt transformation is a process included in the DALL-E 3 models that applies a safety and quality system message to your original prompt. It uses a large language model (LLM) to add a message before sending your prompt to the image generation model. This system message enriches your original prompt with the goal of generating more diverse and higher-quality images while maintaining intent. After prompt transformation is applied to the original prompt, content filtering is applied as a secondary step before image generation. For more information, see Content filtering . Tip Learn more about image generation prompting in OpenAI's Image generation documentation . Prompt transformation example Example text prompt Example generated image without prompt transformation Example generated image with prompt transformation "Watercolor painting of the Seattle skyline" Why is prompt transformation needed? Prompt transformation is essential for responsible and high-quality generations. Not only does prompt transformation improve the safety of your generated image, but it also enriches your prompt in a more descriptive manner, leading to higher quality and descriptive imagery. Default prompt transformation contains safety enhancements that steer the model away from generating images of Copyright Studio characters and artwork, public figures, and oth --- ## Video generation (preview) ### Video generation (preview) URL: https://learn.microsoft.com/en-us/azure/ai-foundry/openai/concepts/video-generation?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Video generation with Sora (preview) Feedback Summarize this article for me In this article Note This document refers to the Microsoft Foundry (classic) portal. 🔄 Switch to the Microsoft Foundry (new) documentation if you're using the new portal. Note This document refers to the Microsoft Foundry (new) portal. Sora is an AI model from OpenAI that creates realistic and imaginative video scenes from text instructions and/or input images or video. The model can generate a wide range of video content, including realistic scenes, animations, and special effects. It supports several video resolutions and durations. Azure OpenAI supports two versions of Sora: Sora (or Sora 1): Azure OpenAI–specific implementation released as an API in early preview. Sora 2: The latest OpenAI-based API, now available with the Azure OpenAI v1 API . Capabilities Modalities: text → video, image → video, video (generated) → video Audio: Sora 2 supports audio generation in output videos (similar to the Sora app). Remix: Sora 2 introduces the ability to remix existing videos by making targeted adjustments instead of regenerating from scratch. Responsible AI and video generation: Azure OpenAI's video generation models include built-in Responsible AI (RAI) protections to help ensure safe and compliant use. Sora 2 blocks all IP and photorealistic content. In addition, Azure provides input and output moderation across all image generation models, along with Azure-specific safeguards such as content filtering and abuse monitoring. These systems help detect and prevent the generation or misuse of harmful, unsafe, or policy-violating content. Customers can learn mor --- ## Video translation (preview) ### Video translation (preview) URL: https://learn.microsoft.com/en-us/azure/ai-services/speech-service/video-translation-get-started?context=/azure/ai-foundry/context/context?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . How to use video translation Feedback Summarize this article for me In this article Reference documentation | Samples In this article, you learn how to use video translation with Azure Speech in Foundry Tools in the Microsoft Foundry portal . Tip Try out video translation in the Microsoft Foundry portal before using the API. Use the video translation REST API to integrate video translation into your applications. For more information about the API, see Video translation REST API . Prerequisites An Azure subscription. If you don't have an Azure subscription, create a free account before you begin. a Foundry resource for Speech in a supported region . If you don't have a Speech resource, create one in the Azure portal . An Azure Blob Storage account. You need a video file in .mp4 format, less than 5 GB, and shorter than 4 hours. For testing purposes, you can use the sample video file provided by Microsoft at https://ai.azure.com/speechassetscache/ttsvoice/VideoTranslation/PublicDoc/SampleData/es-ES-TryOutOriginal.mp4 . Make sure video translation supports your source and target language . Try video translation To try out the video translation demo, follow these steps: Go to the model catalog in Microsoft Foundry portal . Enter and search for "Azure-AI-Speech" in the catalog search box. Select Azure-AI-Speech and you're taken to the Azure-AI-Speech try out page. Select Speech capabilities by scenario > Video translation . Under the Sample option to the right, select personal or standard voice. Select the Play button to hear the translated audio. Select the original video tab to play the original audio. The voice type options are: S --- ## Codex ### Codex URL: https://learn.microsoft.com/en-us/azure/ai-foundry/openai/how-to/codex?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Codex with Azure OpenAI in Microsoft Foundry Models Feedback Summarize this article for me In this article OpenAI’s Codex CLI is the same coding agent that powers ChatGPT’s Codex. You can run this coding agent entirely on Azure infrastructure, while keeping your data inside your compliance boundary with the added advantages of enterprise-grade security, private networking, role-based access control, and predictable cost management. Codex is more than a chat with your code agent – it's an asynchronous coding agent that can be triggered from your terminal, VS Code, or from a GitHub Actions runner. Codex allows you to automatically open pull requests, refactor files, and write tests with the credentials of your Foundry project and Azure OpenAI deployments. Prerequisites An Azure subscription - Create one for free Contributor permissions in Microsoft Foundry . homebrew or npm for installing the Codex CLI or VS Code with the Codex extension. Requirements Details Operating systems macOS 12+, Ubuntu 20.04+/Debian 10+, or Windows 11 via WSL2 Git (optional, recommended) 2.23+ for built-in pull request helpers RAM 4-GB minimum (8-GB recommended) Deploy a model in Foundry Go to Foundry and create a new project. From the model catalog select a reasoning model such as gpt-5.1-codex-max , gpt-5.1-codex , gpt-5.1-codex-mini , gpt-5-codex , gpt-5 , gpt-5-mini , or gpt-5-nano . To deploy the model from the model catalog select Use this model , or if using the Azure OpenAI Deployments pane select deploy model . Copy the endpoint URL and the API Key . Install the Codex CLI From the terminal, run the following commands to install Codex CLI npm brew --- ## Webhooks ### Webhooks URL: https://learn.microsoft.com/en-us/azure/ai-foundry/openai/how-to/webhooks?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Azure OpenAI in Foundry Models webhooks Feedback Summarize this article for me In this article Azure OpenAI webhooks enable your applications to receive real-time notifications about API events, such as batch completions or incoming calls. By subscribing to webhook events, you can automate workflows, trigger alerts, and integrate with other systems seamlessly. This guide walks you through setting up a webhook server, securing your endpoints, deploying, and troubleshooting common issues. Prerequisites Install the required Python packages: pip install flask openai websockets requests Webhook server setup A webhook server is an application that listens for and processes automated messages (webhooks) sent by Azure OpenAI when specific events occur. Create the webhook listener application Create a file called app.py with the following Flask application that receives and processes webhook events: from flask import Flask, request, Response from openai import OpenAI, InvalidWebhookSignatureError import os import logging app = Flask(__name__) # Configure logging for Azure App Service logging.basicConfig(level=logging.INFO) logger = logging.getLogger(__name__) client = OpenAI( # api-key parameter is required, but If you are only using the client for webhooks the key can be a placeholder string api_key=os.environ.get("OPENAI_API_KEY", "placeholder-key-for-webhooks-only"), webhook_secret=os.environ["OPENAI_WEBHOOK_SECRET"] # This will be created later ) @app.route("/webhook", methods=["POST"]) def webhook(): """Webhook endpoint to receive and process OpenAI events.""" try: # Unwrap and verify the message using the webhook secret event = cli --- ## Playgrounds and quick evaluation ### Playgrounds and quick evaluation URL: https://learn.microsoft.com/en-us/azure/ai-foundry/concepts/concept-playgrounds?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Microsoft Foundry Playgrounds Feedback Summarize this article for me In this article Note This document refers to the Microsoft Foundry (classic) portal. 🔄 Switch to the Microsoft Foundry (new) documentation if you're using the new portal. Note This document refers to the Microsoft Foundry (new) portal. Important Items marked (preview) in this article are currently in public preview. This preview is provided without a service-level agreement, and we don't recommend it for production workloads. Certain features might not be supported or might have constrained capabilities. For more information, see Supplemental Terms of Use for Microsoft Azure Previews . As you build with state-of-the-art models and create agents and apps with them, Microsoft Foundry playgrounds provide an on-demand, zero-setup environment designed for rapid prototyping, API exploration, and technical validation before you commit a single line of code to your production codebase. Highlights of the Foundry playgrounds experience Some highlights of the Foundry playgrounds experience include: AgentOps support for evaluations and tracing in the Agents playground. Open in VS Code for Chat and Agents playground. This feature saves you time by automatically importing your endpoint and key from Foundry to VS Code for multilingual code samples. Images playground 2.0 for models such as gpt-image-1 , Stable Diffusion 3.5 Large , and FLUX.1-Kontext-pro models. Video playground for Azure OpenAI Sora-2. Audio playground for models such as gpt-4o-audio-preview , gpt-4o-transcribe , and gpt-4o-mini-tts models. Tip In the screenshot of the playground landing page, the left pane o --- ### Playgrounds and quick evaluation URL: https://learn.microsoft.com/en-us/azure/ai-foundry/concepts/concept-playgrounds?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Microsoft Foundry Playgrounds Feedback Summarize this article for me In this article Note This document refers to the Microsoft Foundry (classic) portal. 🔄 Switch to the Microsoft Foundry (new) documentation if you're using the new portal. Note This document refers to the Microsoft Foundry (new) portal. Important Items marked (preview) in this article are currently in public preview. This preview is provided without a service-level agreement, and we don't recommend it for production workloads. Certain features might not be supported or might have constrained capabilities. For more information, see Supplemental Terms of Use for Microsoft Azure Previews . As you build with state-of-the-art models and create agents and apps with them, Microsoft Foundry playgrounds provide an on-demand, zero-setup environment designed for rapid prototyping, API exploration, and technical validation before you commit a single line of code to your production codebase. Highlights of the Foundry playgrounds experience Some highlights of the Foundry playgrounds experience include: AgentOps support for evaluations and tracing in the Agents playground. Open in VS Code for Chat and Agents playground. This feature saves you time by automatically importing your endpoint and key from Foundry to VS Code for multilingual code samples. Images playground 2.0 for models such as gpt-image-1 , Stable Diffusion 3.5 Large , and FLUX.1-Kontext-pro models. Video playground for Azure OpenAI Sora-2. Audio playground for models such as gpt-4o-audio-preview , gpt-4o-transcribe , and gpt-4o-mini-tts models. Tip In the screenshot of the playground landing page, the left pane o --- ## Model leaderboards ### Model leaderboards URL: https://learn.microsoft.com/en-us/azure/ai-foundry/concepts/model-benchmarks?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Model leaderboards in Microsoft Foundry portal (preview) Feedback Summarize this article for me In this article Note This document refers to the Microsoft Foundry (classic) portal. 🔄 Switch to the Microsoft Foundry (new) documentation if you're using the new portal. Note This document refers to the Microsoft Foundry (new) portal. Important Items marked (preview) in this article are currently in public preview. This preview is provided without a service-level agreement, and we don't recommend it for production workloads. Certain features might not be supported or might have constrained capabilities. For more information, see Supplemental Terms of Use for Microsoft Azure Previews . Model leaderboards (preview) in Microsoft Foundry portal help you compare models in the Foundry model catalog using industry-standard benchmarks. From the model leaderboards section of the model catalog, you can browse leaderboards to compare available models by: Quality, safety, cost, and performance leaderboards to identify leading models on a single metric (quality, safety, cost, or throughput) Trade-off charts to compare performance across two metrics, such as quality versus cost Leaderboards by scenario to find models aligned to specific use cases Model leaderboards (preview) in Foundry portal help you compare models in the Foundry model catalog using industry-standard benchmarks. You can review detailed benchmarking methodology for each leaderboard category: Quality benchmarking of language models to understand how well models perform on cores tasks including reasoning, knowledge, question answering, math, and coding; Safety benchmarking of langua --- ## Compare models ### Compare models URL: https://learn.microsoft.com/en-us/azure/ai-foundry/how-to/benchmark-model-in-catalog?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Compare and select models using the model leaderboard in Microsoft Foundry portal (preview) Feedback Summarize this article for me In this article Note This document refers to the Microsoft Foundry (classic) portal. 🔄 Switch to the Microsoft Foundry (new) documentation if you're using the new portal. Note This document refers to the Microsoft Foundry (new) portal. This article shows you how to streamline model selection in the Microsoft Foundry model catalog by comparing models in the model leaderboards (preview) available in Foundry portal. This comparison can help you make informed decisions about which models meet the requirements for your particular use case or application. This article shows you how to streamline model selection in the Microsoft Foundry model catalog by using the model leaderboards (preview) and side-by-side comparison features in Microsoft Foundry portal. These features enable you to understand model performance through comprehensive leaderboards and direct comparisons, helping you make informed decisions about which models best meet your specific use case or application requirements. You can analyze and compare models using: Model leaderboard to quickly identify top-performing models for quality, safety, estimated cost, and throughput leaderboards Trade-off charts to visually compare model performance across two metrics, such as quality versus cost Leaderboards by scenario to find the most relevant benchmark leaderboard for your specific scenario Compare models to evaluate features, performance, and estimated cost in a side-by-side view Important Items marked (preview) in this article are currently in pub --- ## Upgrade/Switch Models with Ask AI ### Upgrade/Switch Models with Ask AI URL: https://learn.microsoft.com/en-us/azure/ai-foundry/observability/how-to/optimization-model-upgrade?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Upgrade/switch models in Microsoft Foundry with Ask AI (preview) Feedback Summarize this article for me In this article Important Items marked (preview) in this article are currently in public preview. This preview is provided without a service-level agreement, and we don't recommend it for production workloads. Certain features might not be supported or might have constrained capabilities. For more information, see Supplemental Terms of Use for Microsoft Azure Previews . When new model versions are released or older versions are deprecated, use Ask AI—the built-in chat assistant—to detect, evaluate, and upgrade models without leaving the Foundry portal. Ask AI provides conversational guidance, context-aware recommendations, and one-click actions for upgrading, evaluating, and deploying models. This article describes the integrated user experience and system behavior when you initiate or manage model upgrades through Ask AI. Prerequisites A Foundry project with one or more deployed models or agents . Access to Ask AI (the chat assistant) in the Foundry portal. At least one evaluation dataset in CSV or JSONL format. Sufficient permission to deploy and evaluate. Start a chat with Ask AI You can start a chat with Ask AI from any page in the Foundry portal. Select the Ask AI icon at the top of the page. Select one of the predefined prompts from the Ask AI banner under Build/Model/Monitor or Build/Agent/Monitor page, or in Ask AI. Get recommendations on model replacement or upgrade Ask AI questions like: "Is any model I’m using deprecated?" "Should I upgrade my model?" "What’s new in the latest GPT-5 version?" It gives responses and --- ## When to use fine-tuning ### When to use fine-tuning URL: https://learn.microsoft.com/en-us/azure/ai-foundry/openai/concepts/fine-tuning-considerations?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Microsoft Foundry fine-tuning considerations Feedback Summarize this article for me In this article Fine-tuning is the process of taking a pretrained language model and adapting it to perform a specific task or improve its performance on a particular dataset. This involves training the model on a smaller, task-specific dataset while adjusting the model's weights slightly. Fine-tuning leverages the knowledge the model acquired during its initial training on a large, diverse dataset, allowing it to specialize without starting from scratch. This approach is often more efficient than training a new model from scratch; for example, many teams fine-tune with hundreds to thousands of labeled examples instead of retraining on millions of samples. Key benefits of fine-tuning Enhanced accuracy and relevance Fine-tuning improves the model's performance on particular tasks by training it with task-specific data. This often results in more accurate and relevant high-quality outputs compared to using general prompts. Unlike few-shot learning, where only a limited number of examples can be included in a prompt, fine-tuning allows you to train the model on an additional dataset. Fine-tuning helps the model learn more nuanced patterns and improves task performance. Efficiency and potential cost savings Fine-tuned models require shorter prompts because they are trained on relevant examples. This process reduces the number of tokens needed in each request, which can lead to cost savings depending on the use case. Since fine-tuned models need fewer examples in the prompt, they process requests faster, resulting in quicker response times. Scalabilit --- ## Fine-tune models ### Fine-tune models URL: https://learn.microsoft.com/en-us/azure/ai-foundry/openai/how-to/fine-tuning?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Customize a model with fine-tuning Feedback Summarize this article for me In this article Learn how to fine-tune models in Microsoft Foundry for your datasets and use cases. Fine-tuning enables: Higher-quality results than what you can get just from prompt engineering . The ability to train on more examples than what can fit into a model's request context limit. Token savings due to shorter prompts. Lower-latency requests, particularly when you're using smaller models. In contrast to few-shot learning, fine-tuning improves the model by training on more examples than what fits in a prompt. Because weights adapt to your task, you include fewer examples or instructions. Including less reduces tokens per call and potentially lowers cost and latency. We use low-rank adaptation (LoRA) to fine-tune models in a way that reduces their complexity without significantly affecting their performance. This method works by approximating the original high-rank matrix with a lower-rank one. Fine-tuning a smaller subset of important parameters during the supervised training phase makes the model more manageable and efficient. For users, it also makes training faster and more affordable than other techniques. In this article, you learn how to: Choose appropriate datasets and formats for fine-tuning. Trigger a fine-tuning job, monitor the status, and fetch results. Deploy and evaluate a fine-tuned model. Iterate based on evaluation feedback. Prerequisites Read the guide on when to use Foundry fine-tuning . You need an Azure subscription. Create one for free . You need a Foundry project resource. To create one, sign in to the Foundry portal . Fine-tu --- ## Deploy your fine-tuned model ### Deploy your fine-tuned model URL: https://learn.microsoft.com/en-us/azure/ai-foundry/openai/how-to/fine-tuning-deploy?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Deploy a fine-tuned model for inferencing Feedback Summarize this article for me In this article Once your model is fine-tuned, you can deploy the model and can use it in your own application. When you deploy the model, you make the model available for inferencing, and that incurs an hourly hosting charge. Fine-tuned models, however, can be stored in Microsoft Foundry at no cost until you're ready to use them. Azure OpenAI provides choices of deployment types for fine-tuned models on the hosting structure that fits different business and usage patterns: Standard , Global Standard (preview) and Provisioned Throughput (preview). Learn more about deployment types for fine-tuned models and the concepts of all deployment types . Deploy your fine-tuned model Portal Python REST CLI [!IMPROTANT] To deploy models, you need to be assigned as Azure AI owner role or any role with `Microsfot.CognitiveServices/accounts/deployments/write" action. To deploy your custom model, select the custom model to deploy, and then select Deploy . The Deploy model dialog box opens. In the dialog box, enter your Deployment name and then select Create to start the deployment of your custom model. You can monitor the progress of your deployment on the Deployments pane in Foundry portal. The UI does not support cross region deployment, while Python SDK or REST supports. import json import os import requests token = os.getenv("") subscription = "" resource_group = "" resource_name = "" model_deployment_name = "gpt-4.1-mini-ft" # custom deployment name that you will use to ref --- ## Synthetic Data Generation ### Synthetic Data Generation URL: https://learn.microsoft.com/en-us/azure/ai-foundry/fine-tuning/data-generation?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Generate synthetic data for fine-tuning in Microsoft Foundry (Preview) Feedback Summarize this article for me In this article Learn to generate synthetic data in Microsoft Foundry for fine-tuning. Synthetic data helps you: Create large, diverse datasets when real data is scarce Preserve privacy while retaining useful structure Generate business‑specific data from your documents or code Cut cost vs manual collection Create domain-specific data that refines foundation models for your needs—enabling more accurate Q&A in regulated fields and more dependable tool-use by agents—without relying on scarce proprietary datasets. This article covers: Generate synthetic data in the Foundry portal. Prepare and upload a reference file. Configure generation parameters. Review and download results. Integrate generated data into fine-tuning. Apply best practices . Note Preview: Functionality, formats, and limits may change. Always validate outputs before production use. Prerequisites An Azure subscription. Create one for free A Foundry project. For more information, see create a project with Foundry A minimum role assignment of Azure AI User or optionally Azure AI Project Manager on the Foundry resource. For more information, see Manage access with role-based access control (RBAC) Use one of the supported regions for synthetic data generation: eastus2 eastus westus northcentralus southcentralus swedencentral germanywestcentral francecentral uksouth uaenorth japaneast australiaeast Generate synthetic data for fine-tuning Foundry provides generators that turn a reference file into task‑ready training data aligned to your fine‑tuning goal. Overview --- ## Vision fine-tuning ### Vision fine-tuning URL: https://learn.microsoft.com/en-us/azure/ai-foundry/openai/how-to/fine-tuning-vision?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Vision fine-tuning Feedback Summarize this article for me In this article Fine-tuning is also possible with images in your JSONL files. Just as you can send one or many image inputs to chat completions, you can include those same message types within your training data. Images can be provided either as publicly accessible URLs or data URIs containing base64 encoded images . Model support Vision fine-tuning is supported for gpt-4o version 2024-08-06 and gpt-4.1 version 2025-04-14 models only. Image dataset requirements Your training file can contain a maximum of 50,000 examples that contain images (not including text examples). Each example can have at most 64 images. Each image can be at most 10 MB. Format Images must be: JPEG PNG WEBP Images must be in the RGB or RGBA image mode. You cannot include images as output from messages with the assistant role. As with all fine-tuning training your example file requires at least 10 examples. Example file format { "messages": [ { "role": "system", "content": "You are a helpful AI assistant." }, { "role": "user", "content": "Describe the image?" }, { "role": "user", "content": [ { "type": "image_url", "image_url": { "url": "https://raw.githubusercontent.com/MicrosoftDocs/azure-ai-docs/main/articles/ai-services/openai/media/how-to/generated-seattle.png" } } ] }, { "role": "assistant", "content": "The image appears to be a watercolor painting of a city skyline, featuring tall buildings and a recognizable structure often associated with Seattle, like the Space Needle. The artwork uses soft colors and brushstrokes to create a somewhat abstract and artistic representation of the cityscape" } --- ## Preference fine-tuning ### Preference fine-tuning URL: https://learn.microsoft.com/en-us/azure/ai-foundry/openai/how-to/fine-tuning-direct-preference-optimization?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Direct preference optimization (preview) Feedback Summarize this article for me In this article Direct preference optimization (DPO) is an alignment technique for large language models, used to adjust model weights based on human preferences. It differs from reinforcement learning from human feedback (RLHF) in that it does not require fitting a reward model and uses simpler binary data preferences for training. It is computationally lighter weight and faster than RLHF, while being equally effective at alignment. Why is DPO useful? DPO is especially useful in scenarios where there's no clear-cut correct answer, and subjective elements like tone, style, or specific content preferences are important. This approach also enables the model to learn from both positive examples (what's considered correct or ideal) and negative examples (what's less desired or incorrect). DPO is believed to be a technique that will make it easier for customers to generate high-quality training data sets. While many customers struggle to generate sufficient large data sets for supervised fine-tuning, they often have preference data already collected based on user logs, A/B tests, or smaller manual annotation efforts. Direct preference optimization dataset format Direct preference optimization files have a different format than supervised fine-tuning. Customers provide a "conversation" containing the system message and the initial user message, and then "completions" with paired preference data. Users can only provide two completions. Three top-level fields: input , preferred_output and non_preferred_output Each element in the preferred_output/non_preferre --- ## Reinforcement fine-tuning ### Reinforcement fine-tuning URL: https://learn.microsoft.com/en-us/azure/ai-foundry/openai/how-to/reinforcement-fine-tuning?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Reinforcement fine-tuning Feedback Summarize this article for me In this article Reinforcement fine-tuning (RFT) is a technique for improving reasoning models by training them through a reward-based process, rather than relying only on labeled data. RFT helps models develop better reasoning and problem-solving skills, especially in cases where labeled examples are limited or complex behaviors are desired. Note The fine-tuning service automatically pauses RFT jobs once they hit $5,000 in total training costs (training + grading). Users can deploy the most recent checkpoint or resume the training job. If the user decides to resume the job, billing continues for the job with no further cost-based limits. Model support Reinforcement fine-tuning is supported for the following models: o4-mini version 2025-04-16 gpt-5 version 2025-08-07 (preview) GPT-5 support for reinforcement fine tuning is in private preview and might not be available in your subscription. Requirements Reinforcement fine-tuning (RFT) requires training and validation data formatted as JSONL and containing a messages array using the chat completions format. However, RFT has more requirements: Data The final "message" in the data must be assigned a user role. The data can contain extra fields and values for use by a grader. Both a training and a validation dataset must be provided. Graders A grader must be defined to score the quality of your fine-tuned model and guide learning. Only a single grader can be provided, but multiple graders can be combined using a multigrader. Example training data The following example shows how to present prompts to the model and include --- ## Tool calling ### Tool calling URL: https://learn.microsoft.com/en-us/azure/ai-foundry/openai/how-to/fine-tuning-functions?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Fine-tuning and tool calling Feedback Summarize this article for me In this article Models that use the chat completions API support tool calling . Unfortunately, functions defined in your chat completion calls don't always perform as expected. Fine-tuning your model with tool calling examples can improve model output by enabling you to: Get similarly formatted responses even when the full function definition isn't present. (Allowing you to potentially save money on prompt tokens.) Get more accurate and consistent outputs. Note function_call and functions have been deprecated in favor of tools . It is recommended to use the tools parameter instead. Tool calling (recommended) Constructing a training file When constructing a training file of tool calling examples, you would take a function definition like this: { "messages": [ { "role": "user", "content": "What is the weather in San Francisco?" }, { "role": "assistant", "tool_calls": [ { "id": "call_id", "type": "function", "function": { "name": "get_current_weather", "arguments": "{\"location\": \"San Francisco, USA\", \"format\": \"celsius\"}" } } ] } ], "tools": [ { "type": "function", "function": { "name": "get_current_weather", "description": "Get the current weather", "parameters": { "type": "object", "properties": { "location": { "type": "string", "description": "The city and country/region, eg. San Francisco, USA" }, "format": { "type": "string", "enum": ["celsius", "fahrenheit"] } }, "required": ["location", "format"] } } } ] } And express the information as a single line within your .jsonl training file as below: {"messages":[{"role":"user","content":"What is the weather --- ## Safety evaluation ### Safety evaluation URL: https://learn.microsoft.com/en-us/azure/ai-foundry/openai/how-to/fine-tuning-safety-evaluation?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Safety evaluation for fine-tuning (preview) Feedback Summarize this article for me In this article The advanced capabilities of fine-tuned models come with increased responsible AI challenges related to harmful content, manipulation, human-like behavior, privacy issues, and more. Learn more about risks, capabilities, and limitations in the Overview of Responsible AI practices and Transparency Note . To help mitigate the risks associated with advanced fine-tuned models, we have implemented additional evaluation steps to help detect and prevent harmful content in the training and outputs of fine-tuned models. These steps are grounded in the Microsoft Responsible AI Standard and Azure Microsoft Foundry Models content filtering . Evaluations are conducted in dedicated, customer specific, private workspaces; Evaluation endpoints are in the same geography as the Foundry resource; Training data isn't stored in connection with performing evaluations; only the final model assessment (deployable or not deployable) is persisted; and Fine-tuned model evaluation filters are set to predefined thresholds and can't be modified by customers; they aren't tied to any custom content filtering configuration you might have created. Data evaluation Before training starts, your data is evaluated for potentially harmful content (violence, sexual, hate, and fairness, self-harm – see category definitions here ). If harmful content is detected above the specified severity level, your training job will fail, and you'll receive a message informing you of the categories of failure. Sample message: The provided training data failed RAI checks for harm types: [ --- ## Fine-tuning cost management ### Fine-tuning cost management URL: https://learn.microsoft.com/en-us/azure/ai-foundry/openai/how-to/fine-tuning-cost-management?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Cost management for fine-tuning Feedback Summarize this article for me In this article Fine-tuning can be intimidating: unlike base models, where you're just paying for input and output tokens for inferencing, fine-tuning requires training your custom models and dealing with hosting. This guide is intended to help you better understand the costs associated with fine-tuning and how to manage them. Important The numbers in this article are for example purposes only. You should always refer to the official pricing page for pricing details to use in the formulas provided in this article. Upfront investment - training your model This is the one-time, fixed cost associated with teaching a base model your specific requirements using your training data. Below, we explain how training costs can be calculated. For all training jobs, you have the option to use global standard (10-30% discount from regional standard training) or developer (50% discount from global). Neither Global nor Developer training guarantee data residency; developer training will run on pre-emptible spot capacity so may take longer to complete. Developer tier jobs may be paused by the system but will automatically resume; users do not incur charges for jobs in paused states. The calculation formula Supervised Fine-Tuning (SFT) & Preference Fine-Tuning (DPO) It's straightforward to estimate the costs for SFT & DPO. You're charged based on the number of tokens in your training file, and the number of epochs for your training job. $$ \text{price} = \text{# training tokens} \times \text{# epochs} \times \text{training price per token} $$ In general, smaller models and mor --- ## What is the Foundry control plane? ### What is the Foundry control plane? URL: https://learn.microsoft.com/en-us/azure/ai-foundry/control-plane/overview?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . What is the Microsoft Foundry Control Plane? Feedback Summarize this article for me In this article The Microsoft Foundry Control Plane is a unified management interface that provides visibility, governance, and control for AI agents, models, and tools across your Foundry enterprise. Use the Foundry Control Plane as your central location for managing every aspect of your AI fleet, from build to production. As your organization evolves from isolated copilots to autonomous multi-agent fleets, you need unified oversight. The Foundry Control Plane provides the visibility, governance, and control you need to scale with confidence. In this article, you learn what the Foundry Control Plane offers, including fleet management, observability, compliance enforcement, and security capabilities. Important Items marked (preview) in this article are currently in public preview. This preview is provided without a service-level agreement, and we don't recommend it for production workloads. Certain features might not be supported or might have constrained capabilities. For more information, see Supplemental Terms of Use for Microsoft Azure Previews . Core functionalities The Foundry Control Plane consolidates inventory, observability, compliance, and security into one role-aware interface. It integrates seamlessly with Microsoft security and governance systems (Defender, Purview, Microsoft Entra) to deliver trust at scale . The following diagram shows how the Foundry Control Plane provides unified fleet visibility with agents, models, and tools listed across projects in a subscription: The Foundry Control Plane allows you to: Manage your fleet ac --- ## Monitor fleet health and performance ### Monitor fleet health and performance URL: https://learn.microsoft.com/en-us/azure/ai-foundry/control-plane/monitoring-across-fleet?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Monitor agent health and performance across your fleet Feedback Summarize this article for me In this article As your organization scales from isolated copilots to autonomous multi-agent fleets, maintaining visibility and control becomes critical. The Foundry Control Plane provides a unified command center where you can monitor all agents, models, and tools across your enterprise from build to production. Fleet monitoring serves multiple roles: Team managers gain oversight of agent operations and team productivity. Administrators enforce governance policies and track compliance posture. Cost managers optimize spending and identify resource inefficiencies. Security teams monitor for prohibited behaviors and policy violations. This article shows you how to use the Foundry Control Plane's capabilities to track agent health, performance, compliance, and cost efficiency at scale. By using centralized monitoring, you can identify problems early, optimize resource consumption, and ensure your AI systems operate safely and reliably. Important Items marked (preview) in this article are currently in public preview. This preview is provided without a service-level agreement, and we don't recommend it for production workloads. Certain features might not be supported or might have constrained capabilities. For more information, see Supplemental Terms of Use for Microsoft Azure Previews . Prerequisites An Azure account with an active subscription. If you don't have one, create a free Azure account, which includes a free trial subscription . A Foundry project. If you don't have one, create a project . You need the following permissions: Read a --- ## Manage agents at scale ### Manage agents at scale URL: https://learn.microsoft.com/en-us/azure/ai-foundry/control-plane/how-to-manage-agents?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Manage agents at scale Feedback Summarize this article for me In this article The Microsoft Foundry Control Plane provides centralized management and observability for agents running across different platforms and infrastructures. This article explains how to manage agents across a subscription using Microsoft Foundry Control Plane. Agents inventory The Assets page provides a unified, searchable table of all AI assets across projects within a subscription. It brings together critical metadata and health indicators, so you can assess and act on your AI estate efficiently. Control Plane automatically discovers supported agents within resources in the selected subscription and displays them in the Operate > Assets > Agents page. The following information is displayed: Column Description Agent platform Name The name of the agent or the agentic resource. All Source The source platform from where the agent or resource was discovered. See the list of supported platforms . All Project The Foundry project associated with the agent. For custom agents, it's the project where the agent was registered to. Foundry Custom Status It refers to a broad range of conditions, including operational, health, or lifecycle, status of the agent. Agents transition to different values depending on the platform and lifecycle operations . Possible values are: Running Stopped Blocked Unblocked Unknown All Version The version of the agent asset. Foundry Published as Indicates if the agent was published as an agent application . Published agents in Foundry have their own endpoint for invocation. Foundry Error rate The proportion of failed runs compared to succe --- ## Register and manage custom agents ### Register and manage custom agents URL: https://learn.microsoft.com/en-us/azure/ai-foundry/control-plane/register-custom-agent?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Register and manage custom agents Feedback Summarize this article for me In this article The Microsoft Foundry Control Plane provides centralized management and observability for agents running across different platforms and infrastructures. You can register custom agents running in Azure compute services or other cloud environments to gain visibility into their operations and control their behavior. This article shows you how to register a custom agent in the Foundry Control Plane. You learn how to configure your agent for registration, set up telemetry collection, and use the Control Plane's management capabilities. Important Items marked (preview) in this article are currently in public preview. This preview is provided without a service-level agreement, and we don't recommend it for production workloads. Certain features might not be supported or might have constrained capabilities. For more information, see Supplemental Terms of Use for Microsoft Azure Previews . Prerequisites Before getting started, make sure you have: An Azure account with an active subscription. If you don't have one, create a free Azure account, which includes a free trial subscription . A Foundry project. If you don't have one, create a project . Foundry uses Azure API Management to register agents as APIs. Configure AI Gateway in your Foundry resource . An agent that you deploy and expose through a reachable endpoint (either a public endpoint or an endpoint reachable from the network where you deploy the Foundry resource). Note This capability is available only in the Foundry (new) portal. Look for in the portal banner to confirm you're using Foundry --- ## Enforce token limits ### Enforce token limits URL: https://learn.microsoft.com/en-us/azure/ai-foundry/control-plane/how-to-enforce-limits-models?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Enforce token limits for models Feedback Summarize this article for me In this article Microsoft Foundry Control Plane enforces tokens-per-minute (TPM) rate limits and total token quotas for model deployments at the project scope to prevent runaway token consumption and align usage with organizational guardrails. Control Planes integrates with AI Gateway to provide advance policy enforcement for models. This article explains how to configure token rate limiting and token quotas. Prerequisites Before getting started, make sure you have: An Azure account with an active subscription. If you don't have one, create a free Azure account, which includes a free trial subscription . A Foundry resource with AI Gateway configured. Learn more about how to enable AI Gateway for a Foundry resource . A Foundry project added to the configured AI Gateway. Tip You need API Management Service Contributor (or Owner ) on the Azure API Management resource to enable AI Gateway at a given project. Understand AI Gateway Control Planes integrates with AI Gateway to provide advanced policy enforcement for models. AI Gateway sits between clients and model deployments, making all requests flow through the API Management instance associated with it. Limits apply at the project level (each project can have its own TPM and quota settings). Use AI Gateway for: Multi-team token containment (prevent one project from monopolizing capacity). Cost control by capping aggregate usage. Compliance boundaries for regulated workloads (enforce predictable usage ceilings). Configure token limits You can configure token limits for specific model deployments within your proje --- ## Apply a guardrail policy for models ### Apply a guardrail policy for models URL: https://learn.microsoft.com/en-us/azure/ai-foundry/control-plane/quickstart-create-guardrail-policy?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Quickstart: Create a guardrail policy Feedback Summarize this article for me In this article In this quickstart, you create an Azure Policy in Microsoft Foundry to govern the use of guardrail controls for model deployments across your subscription. Important Items marked (preview) in this article are currently in public preview. This preview is provided without a service-level agreement, and we don't recommend it for production workloads. Certain features might not be supported or might have constrained capabilities. For more information, see Supplemental Terms of Use for Microsoft Azure Previews . Prerequisites An Azure account with an active subscription. If you don't have one, create a free Azure account, which includes a free trial subscription . A Foundry project. If you don't have one, create a project . Required permissions: You must have the appropriate roles to create an Azure Policy for your subscription. You can learn more about Azure Policy roles here: Overview of Azure Policy - Azure Policy Note This capability is available only in the Microsoft Foundry (new) portal . Create the guardrail policy Sign in to Microsoft Foundry . Make sure the New Foundry toggle is on. These steps refer to Foundry (new) . Select Operate from the upper-right navigation. Select Compliance in the left pane. Select Create policy . Select the controls to be added to the policy. Guardrail controls include content safety filters, prompt shields, and groundedness checks that help ensure your AI models operate safely and responsibly. These controls represent the minimum settings required for a model deployment to be considered compliant with the --- ## Optimize cost and performance ### Optimize cost and performance URL: https://learn.microsoft.com/en-us/azure/ai-foundry/control-plane/how-to-optimize-cost-performance?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Optimize model cost and performance Feedback Summarize this article for me In this article When your model or agent costs start increasing, use the Ask AI agent — the built-in chat assistant — to quickly diagnose issues, take action, and verify improvements. You can access the Ask AI agent from the top navigation bar in the Foundry portal. This article walks you through a recommended workflow, from identifying cost spikes to switching models and validating performance improvements — all within the Foundry portal. Tip The Operate > Overview page includes pre-built prompts specific to agent optimization and performance. Select one of these prompts to start a conversation with the Ask AI agent, or open the Ask AI agent from the top navigation bar and type your own question. Prerequisites An Azure account with an active subscription. If you don't have one, create a free Azure account, which includes a free trial subscription . A Foundry project. If you don't have one, create a project . At least one deployed or published agent with cost data. For meaningful trend analysis, you need a minimum of 7 days of usage data. You have access to the Ask AI agent (the chat assistant). An evaluation dataset configured for your project. To set one up, see Create and manage evaluation datasets . Detect cost increases Start by opening the Ask AI agent from the top navigation bar, or go to Operate > Overview to use one of the pre-built prompts. Ask the assistant to provide a summary of your metrics and cost data from the Foundry Control Plane dashboard. You can select a predefined prompt on the Operate overview page, or type your own question, such --- ## Manage compliance and security ### Manage compliance and security URL: https://learn.microsoft.com/en-us/azure/ai-foundry/control-plane/how-to-manage-compliance-security?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Manage compliance and security in Microsoft Foundry Feedback Summarize this article for me In this article Learn how Foundry Control Plane helps you manage compliance, enforce guardrail controls, and integrate security tooling such as Microsoft Defender for Cloud across subscriptions. Important Items marked (preview) in this article are currently in public preview. This preview is provided without a service-level agreement, and we don't recommend it for production workloads. Certain features might not be supported or might have constrained capabilities. For more information, see Supplemental Terms of Use for Microsoft Azure Previews . Use the compliance workspace tabs to reach the right surface quickly. Tab Navigation Outcome Policies Operate > Compliance > Policies Review guardrail policies, check compliance, and create or edit enforcement rules. Assets Operate > Compliance > Assets Inspect individual model deployments, view policy violations, and jump to remediation. Guardrails Operate > Compliance > Guardrails Compare guardrail configurations across deployments and spot coverage gaps. Security Operate > Compliance > Security Review Microsoft Defender for Cloud recommendations and manage Microsoft Purview enablement. Prerequisites An Azure account with an active subscription. If you don't have one, create a free Azure account, which includes a free trial subscription . A Foundry project. If you don't have one, create a project . If you use agents, you need Agents v2 or later for full compliance feature support. Appropriate permissions based on the tasks you want to perform: To view compliance status and guardrail policies: No --- ## Configure AI Gateway ### Configure AI Gateway URL: https://learn.microsoft.com/en-us/azure/ai-foundry/configuration/enable-ai-api-management-gateway-portal?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Configure AI Gateway in your Foundry resources Feedback Summarize this article for me In this article This article shows you how to enable AI Gateway for a Microsoft Foundry resource using the Foundry portal. AI Gateway uses Azure API Management behind the scenes to provide token limits, quotas, and governance for model deployments. Prerequisites Azure subscription ( create one for free ). Permissions to create or reuse an Azure API Management (APIM) instance: To create an APIM instance: Contributor or Owner on the target resource group (or subscription). To manage an existing APIM instance: API Management Service Contributor (or Owner ) on the APIM instance. For more information, see How to use role-based access control in Azure API Management . Access to the Foundry portal ( Admin console ) for the target Foundry resource. For example: Azure AI Account Owner or Azure AI Owner on the Foundry resource. For more information, see Role-based access control for Microsoft Foundry . Decision on whether to create a dedicated APIM instance or reuse an existing one. Create an AI Gateway Follow these steps in the Foundry portal to enable AI Gateway for a resource. Sign in to Microsoft Foundry . Make sure the New Foundry toggle is on. These steps refer to Foundry (new) . Select Operate > Admin console . Open the AI Gateway tab. Select Add AI Gateway . Select the Foundry resource you want to connect with the gateway. Select Create new or Use existing APIM. Create new : Creates a Basic v2 SKU instance. Basic v2 is designed for development and testing with SLA support. Use existing : Select an instance that meets your organization's governanc --- ## Observability basics ### Observability basics URL: https://learn.microsoft.com/en-us/azure/ai-foundry/concepts/observability?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Observability in generative AI Feedback Summarize this article for me In this article Note This document refers to the Microsoft Foundry (classic) portal. 🔄 Switch to the Microsoft Foundry (new) documentation if you're using the new portal. Note This document refers to the Microsoft Foundry (new) portal. Important Items marked (preview) in this article are currently in public preview. This preview is provided without a service-level agreement, and we don't recommend it for production workloads. Certain features might not be supported or might have constrained capabilities. For more information, see Supplemental Terms of Use for Microsoft Azure Previews . In today's AI-driven world, Generative AI Operations (GenAIOps) is revolutionizing how organizations build and deploy intelligent systems. As companies increasingly use AI agents and applications to transform decision-making, enhance customer experiences, and fuel innovation, one element stands paramount: robust evaluation frameworks. Evaluation isn't just a checkpoint. It's the foundation of quality and trust in AI applications. Without rigorous assessment and monitoring, AI systems can produce content that's: Fabricated or ungrounded in reality Irrelevant or incoherent Harmful in perpetuating content risks and stereotypes Dangerous in spreading misinformation Vulnerable to security exploits This is where observability becomes essential. These capabilities measure both the frequency and severity of risks in AI outputs, enabling teams to systematically address quality, safety, and security concerns throughout the entire AI development journey—from selecting the right model to mo --- ## Transparency Note for safety evaluations ### Transparency Note for safety evaluations URL: https://learn.microsoft.com/en-us/azure/ai-foundry/concepts/safety-evaluations-transparency-note?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Microsoft Foundry risk and safety evaluations (preview) Transparency Note Feedback Summarize this article for me In this article Note This document refers to the Microsoft Foundry (classic) portal. 🔄 Switch to the Microsoft Foundry (new) documentation if you're using the new portal. Note This document refers to the Microsoft Foundry (new) portal. Important Items marked (preview) in this article are currently in public preview. This preview is provided without a service-level agreement, and we don't recommend it for production workloads. Certain features might not be supported or might have constrained capabilities. For more information, see Supplemental Terms of Use for Microsoft Azure Previews . What is a Transparency Note An AI system includes not only the technology, but also the people who will use it, the people who will be affected by it, and the environment in which it's deployed. Creating a system that is fit for its intended purpose requires an understanding of how the technology works, what its capabilities and limitations are, and how to achieve the best performance. Microsoft's Transparency Notes are intended to help you understand how our AI technology works, the choices system owners can make that influence system performance and behavior, and the importance of thinking about the whole system, including the technology, the people, and the environment. You can use Transparency Notes when developing or deploying your own system, or share them with the people who will use or be affected by your system. Microsoft's Transparency Notes are part of a broader effort at Microsoft to put our AI Principles into practice. To f --- ## General purpose evaluators ### General purpose evaluators URL: https://learn.microsoft.com/en-us/azure/ai-foundry/concepts/evaluation-evaluators/general-purpose-evaluators?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . General purpose evaluators Feedback Summarize this article for me In this article Note This document refers to the Microsoft Foundry (classic) portal. 🔄 Switch to the Microsoft Foundry (new) documentation if you're using the new portal. Note This document refers to the Microsoft Foundry (new) portal. AI systems might generate textual responses that are incoherent, or lack the general writing quality beyond minimum grammatical correctness. To address these issues, Microsoft Foundry supports evaluating: Coherence Fluency Note The Microsoft Foundry SDK for evaluation and Foundry portal are in public preview, but the APIs are generally available for model and dataset evaluation (agent evaluation remains in public preview). The Azure AI Evaluation SDK and evaluators marked (preview) in this article are currently in public preview everywhere. Note The Microsoft Foundry SDK for evaluation and Foundry portal are in public preview, but the APIs are generally available for model and dataset evaluation (agent evaluation remains in public preview). Evaluators marked (preview) in this article are currently in public preview everywhere. If you have a question-answering (QA) scenario with both context and ground truth data in addition to query and response , you can also use our QAEvaluator , which is a composite evaluator that uses relevant evaluators for judgment. Model configuration for AI-assisted evaluators For reference in the following code snippet, the AI-assisted evaluators use a model configuration as follows: import os from azure.ai.evaluation import AzureOpenAIModelConfiguration from dotenv import load_dotenv load_dotenv() model_co --- ## Textual similarity evaluators ### Textual similarity evaluators URL: https://learn.microsoft.com/en-us/azure/ai-foundry/concepts/evaluation-evaluators/textual-similarity-evaluators?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Textual similarity evaluators Feedback Summarize this article for me In this article Note This document refers to the Microsoft Foundry (classic) portal. 🔄 Switch to the Microsoft Foundry (new) documentation if you're using the new portal. Note This document refers to the Microsoft Foundry (new) portal. Note The Microsoft Foundry SDK for evaluation and Foundry portal are in public preview, but the APIs are generally available for model and dataset evaluation (agent evaluation remains in public preview). The Azure AI Evaluation SDK and evaluators marked (preview) in this article are currently in public preview everywhere. Note The Microsoft Foundry SDK for evaluation and Foundry portal are in public preview, but the APIs are generally available for model and dataset evaluation (agent evaluation remains in public preview). Evaluators marked (preview) in this article are currently in public preview everywhere. It's important to compare how closely the textual response generated by your AI system matches the response you would expect. The expected response is called the ground truth . Use a LLM-judge metric like Similarity with a focus on the semantic similarity between the generated response and the ground truth. Or, use metrics from the field of natural language processing (NLP), including F1 score , BLEU , GLEU , ROUGE , and METEOR with a focus on the overlaps of tokens or n-grams between the two. Model configuration for AI-assisted evaluators For reference in the following code snippets, the AI-assisted evaluators use a model configuration for the LLM-judge: import os from azure.ai.evaluation import AzureOpenAIModelConfiguration --- ## Retrieval Augmented Generation (RAG) evaluators ### Retrieval Augmented Generation (RAG) evaluators URL: https://learn.microsoft.com/en-us/azure/ai-foundry/concepts/evaluation-evaluators/rag-evaluators?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Retrieval-Augmented Generation (RAG) evaluators Feedback Summarize this article for me In this article Note This document refers to the Microsoft Foundry (classic) portal. 🔄 Switch to the Microsoft Foundry (new) documentation if you're using the new portal. Note This document refers to the Microsoft Foundry (new) portal. A Retrieval-Augmented Generation (RAG) system tries to generate the most relevant answer consistent with grounding documents in response to a user's query. A user's query triggers a search retrieval in the corpus of grounding documents to provide grounding context for the AI model to generate a response. Note The Microsoft Foundry SDK for evaluation and Foundry portal are in public preview, but the APIs are generally available for model and dataset evaluation (agent evaluation remains in public preview). The Azure AI Evaluation SDK and evaluators marked (preview) in this article are currently in public preview everywhere. Note The Microsoft Foundry SDK for evaluation and Foundry portal are in public preview, but the APIs are generally available for model and dataset evaluation (agent evaluation remains in public preview). Evaluators marked (preview) in this article are currently in public preview everywhere. It's important to evaluate: Document Retrieval Retrieval Groundedness Groundedness Pro (preview) Relevance Response Completeness These evaluators focus on three aspects: The relevance of the retrieval results to the user's query: use Document Retrieval if you have labels for query-specific document relevance, or query relevance judgement (qrels) for more accurate measurements. Use Retrieval if you only have --- ## Risk and safety evaluators ### Risk and safety evaluators URL: https://learn.microsoft.com/en-us/azure/ai-foundry/concepts/evaluation-evaluators/risk-safety-evaluators?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Risk and safety evaluators Feedback Summarize this article for me In this article Note This document refers to the Microsoft Foundry (classic) portal. 🔄 Switch to the Microsoft Foundry (new) documentation if you're using the new portal. Note This document refers to the Microsoft Foundry (new) portal. Note The Microsoft Foundry SDK for evaluation and Foundry portal are in public preview, but the APIs are generally available for model and dataset evaluation (agent evaluation remains in public preview). The Azure AI Evaluation SDK and evaluators marked (preview) in this article are currently in public preview everywhere. Note The Microsoft Foundry SDK for evaluation and Foundry portal are in public preview, but the APIs are generally available for model and dataset evaluation (agent evaluation remains in public preview). Evaluators marked (preview) in this article are currently in public preview everywhere. Risk and safety evaluators draw on insights gained from our previous large language model (LLM) projects such as GitHub Copilot and Bing. This approach ensures a comprehensive approach to evaluating generated responses for risk and safety severity scores. These evaluators are generated through the Microsoft Foundry Evaluation service, which employs a set of language models. Each model assesses specific risks that could be present in the response from your AI system. Specific risks include sexual content, violent content, and other content. These evaluator models are provided with risk definitions and annotate accordingly. Currently, we support the following risks for assessment: Hateful and unfair content Sexual content Violent --- ## Agent evaluators ### Agent evaluators URL: https://learn.microsoft.com/en-us/azure/ai-foundry/concepts/evaluation-evaluators/agent-evaluators?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Agent evaluators (preview) Feedback Summarize this article for me In this article Note This document refers to the Microsoft Foundry (classic) portal. 🔄 Switch to the Microsoft Foundry (new) documentation if you're using the new portal. Note This document refers to the Microsoft Foundry (new) portal. Important Items marked (preview) in this article are currently in public preview. This preview is provided without a service-level agreement, and we don't recommend it for production workloads. Certain features might not be supported or might have constrained capabilities. For more information, see Supplemental Terms of Use for Microsoft Azure Previews . Note The Microsoft Foundry SDK for evaluation and Foundry portal are in public preview, but the APIs are generally available for model and dataset evaluation (agent evaluation remains in public preview). The Azure AI Evaluation SDK and evaluators marked (preview) in this article are currently in public preview everywhere. Note The Microsoft Foundry SDK for evaluation and Foundry portal are in public preview, but the APIs are generally available for model and dataset evaluation (agent evaluation remains in public preview). Evaluators marked (preview) in this article are currently in public preview everywhere. Agents are powerful productivity assistants. They plan, make decisions, and execute actions. Agents typically reason through user intents in conversations , select the correct tools to satisfy user requests, and complete tasks according to instructions. Microsoft Foundry supports these agent-specific evaluators for agentic workflows: Intent resolution (preview) Tool call accurac --- ## Azure OpenAI evaluators ### Azure OpenAI evaluators URL: https://learn.microsoft.com/en-us/azure/ai-foundry/concepts/evaluation-evaluators/azure-openai-graders?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Azure OpenAI graders Feedback Summarize this article for me In this article Note This document refers to the Microsoft Foundry (classic) portal. 🔄 Switch to the Microsoft Foundry (new) documentation if you're using the new portal. Note This document refers to the Microsoft Foundry (new) portal. Azure OpenAI graders are a new set of evaluation tools in the Microsoft Foundry SDK that evaluate the performance of AI models and their outputs. These graders include: Label grader String checker Text similarity Python grader Name Type What it does label_grader label_model Classifies sentiment as positive , neutral , or negative using an LLM. text_check_grader text_similarity Compares ground truth and response using BLEU score for similarity. string_check_grader string_check Performs a string equality check between two values. score score_model Assigns a similarity score (1–5) based on semantic and structural comparison. You can run graders locally or remotely. Each grader assesses specific aspects of AI models and their outputs. Note The Microsoft Foundry SDK for evaluation and Foundry portal are in public preview, but the APIs are generally available for model and dataset evaluation (agent evaluation remains in public preview). The Azure AI Evaluation SDK and evaluators marked (preview) in this article are currently in public preview everywhere. Note The Microsoft Foundry SDK for evaluation and Foundry portal are in public preview, but the APIs are generally available for model and dataset evaluation (agent evaluation remains in public preview). Evaluators marked (preview) in this article are currently in public preview everywhere. Mod --- ## Custom evaluators ### Custom evaluators URL: https://learn.microsoft.com/en-us/azure/ai-foundry/concepts/evaluation-evaluators/custom-evaluators?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Custom evaluators Feedback Summarize this article for me In this article Note This document refers to the Microsoft Foundry (classic) portal. 🔄 Switch to the Microsoft Foundry (new) documentation if you're using the new portal. Note This document refers to the Microsoft Foundry (new) portal. Built-in evaluators provide an easy way to monitor the quality of your application's generations. To customize your evaluations, you can create your own code-based or prompt-based evaluators. Note The Microsoft Foundry SDK for evaluation and Foundry portal are in public preview, but the APIs are generally available for model and dataset evaluation (agent evaluation remains in public preview). The Azure AI Evaluation SDK and evaluators marked (preview) in this article are currently in public preview everywhere. Note The Microsoft Foundry SDK for evaluation and Foundry portal are in public preview, but the APIs are generally available for model and dataset evaluation (agent evaluation remains in public preview). Evaluators marked (preview) in this article are currently in public preview everywhere. Code-based evaluators You don't need a large language model for certain evaluation metrics. Code-based evaluators give you the flexibility to define metrics based on functions or callable classes. You can build your own code-based evaluator, for example, by creating a simple Python class that calculates the length of an answer in answer_length.py under the directory answer_len/ , as in the following example. Code-based evaluator example: Answer length class AnswerLengthEvaluator: def __init__(self): pass # A class is made callable by implementing th --- ## Run evaluations in the cloud ### Run evaluations in the cloud URL: https://learn.microsoft.com/en-us/azure/ai-foundry/how-to/develop/cloud-evaluation?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Run evaluations in the cloud by using the Microsoft Foundry SDK Feedback Summarize this article for me In this article Note This document refers to the Microsoft Foundry (classic) portal. 🔄 Switch to the Microsoft Foundry (new) documentation if you're using the new portal. Note This document refers to the Microsoft Foundry (new) portal. Important Items marked (preview) in this article are currently in public preview. This preview is provided without a service-level agreement, and we don't recommend it for production workloads. Certain features might not be supported or might have constrained capabilities. For more information, see Supplemental Terms of Use for Microsoft Azure Previews . In this article, you learn how to run evaluations in the cloud (preview) for predeployment testing on a test dataset. The Azure AI Evaluation SDK lets you run evaluations locally on your machine and in the cloud. For example, run local evaluations on small test data to assess your generative AI application prototypes, and then move into predeployment testing to run evaluations on a large dataset. Use cloud evaluations for most scenarios—especially when testing at scale, integrating evaluations into continuous integration and continuous delivery (CI/CD) pipelines, or performing predeployment testing. Running evaluations in the cloud eliminates the need to manage local compute infrastructure and supports large scale, automated testing workflows. After deployment, you can choose to continuously evaluate your agents for post-deployment monitoring. When you use the Foundry SDK, it logs evaluation results in your Foundry project for better observabilit --- ## Evaluate your AI agents ### Evaluate your AI agents URL: https://learn.microsoft.com/en-us/azure/ai-foundry/how-to/develop/agent-evaluate-sdk?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Evaluate your AI agents (preview) Feedback Summarize this article for me In this article Note This document refers to the Microsoft Foundry (classic) portal. 🔄 Switch to the Microsoft Foundry (new) documentation if you're using the new portal. Note This document refers to the Microsoft Foundry (new) portal. Important Items marked (preview) in this article are currently in public preview. This preview is provided without a service-level agreement, and we don't recommend it for production workloads. Certain features might not be supported or might have constrained capabilities. For more information, see Supplemental Terms of Use for Microsoft Azure Previews . AI agents are powerful productivity assistants that create workflows for business needs. However, observability is challenging because of their complex interaction patterns. In this article, you learn how to evaluate Microsoft Foundry agents or other agents using built-in evaluators. To build production-ready agentic applications and ensure observability and transparency, developers need tools to assess not only the final output of an agent's workflows but also the quality and efficiency of the workflows. An event like a user querying "weather tomorrow" triggers an agentic workflow. To produce a final response, the workflow includes reasoning through user intents, calling tools, and using retrieval-augmented generation. Note The Microsoft Foundry SDK for evaluation and Foundry portal are in public preview, but the APIs are generally available for model and dataset evaluation (agent evaluation remains in public preview). The Azure AI Evaluation SDK and evaluators marked (previ --- ## Run evaluations from the portal ### Run evaluations from the portal URL: https://learn.microsoft.com/en-us/azure/ai-foundry/how-to/evaluate-generative-ai-app?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Evaluate generative AI models and applications by using Microsoft Foundry Feedback Summarize this article for me In this article Note This document refers to the Microsoft Foundry (classic) portal. 🔄 Switch to the Microsoft Foundry (new) documentation if you're using the new portal. Note This document refers to the Microsoft Foundry (new) portal. To thoroughly assess the performance of your generative AI models and applications on a substantial dataset, initiate an evaluation process. During this evaluation, the model or application is tested with the given dataset, and its performance is measured using mathematical metrics and AI-assisted metrics. This evaluation run provides comprehensive insights into the application's capabilities and limitations. Use the evaluation functionality in the Microsoft Foundry portal, a platform that offers tools and features for assessing the performance and safety of generative AI models. In the Foundry portal, log, view, and analyze detailed evaluation metrics. This article explains how to create an evaluation run against a model, agent, or test dataset using built-in evaluation metrics from the Foundry UI. For greater flexibility, you can establish a custom evaluation flow and employ the custom evaluation feature. Use the custom evaluation feature to conduct a batch run without evaluation. Prerequisites A test dataset in one of these formats: CSV or JSON Lines (JSONL). An Azure OpenAI connection with a deployment of one of these models: a GPT-3.5 model, a GPT-4 model, or a Davinci model. This is required only for AI-assisted quality evaluations. A test dataset in one of these formats: A model, --- ## View evaluation results in the portal ### View evaluation results in the portal URL: https://learn.microsoft.com/en-us/azure/ai-foundry/how-to/evaluate-results?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . See evaluation results in the Microsoft Foundry portal Feedback Summarize this article for me In this article Note This document refers to the Microsoft Foundry (classic) portal. 🔄 Switch to the Microsoft Foundry (new) documentation if you're using the new portal. Note This document refers to the Microsoft Foundry (new) portal. Learn how to see evaluation results in the Microsoft Foundry portal. View and interpret AI model evaluation data, performance metrics, and quality assessments. Access results from flows, playground sessions, and SDK to make data-driven decisions. After visualizing your evaluation results, examine them thoroughly. View individual results, compare them across multiple evaluation runs, and identify trends, patterns, and discrepancies to gain insights into your AI system's performance under various conditions. In this article, you learn to: Locate and open evaluation runs. View aggregate and sample-level metrics. Compare results across runs. Interpret metric categories and calculations. Troubleshoot missing or partial metrics. See your evaluation results After you submit an evaluation, locate the run on the Evaluation page. Filter or adjust columns to focus on runs of interest. Review high-level metrics at a glance before drilling in. Tip You can view an evaluation run with any version of the promptflow-evals SDK or azure-ai-evaluation versions 1.0.0b1, 1.0.0b2, 1.0.0b3. Enable the Show all runs toggle to locate the run. Select Learn more about metrics for definitions and formulas. Select a run to open details (dataset, task type, prompt, parameters) plus per-sample metrics. The metrics dashboard visualizes p --- ## Evaluation Cluster Analysis ### Evaluation Cluster Analysis URL: https://learn.microsoft.com/en-us/azure/ai-foundry/observability/how-to/cluster-analysis?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Evaluation cluster analysis (preview) Feedback Summarize this article for me In this article Important Items marked (preview) in this article are currently in public preview. This preview is provided without a service-level agreement, and we don't recommend it for production workloads. Certain features might not be supported or might have constrained capabilities. For more information, see Supplemental Terms of Use for Microsoft Azure Previews . After you run one or more evaluation runs, you can generate an evaluation cluster analysis to understand your evaluation results. This analysis provides an intuitive way to identify the top patterns and errors in your evaluation runs, along with recommended next steps to improve evaluator scores. This article explains how to generate and interact with an evaluation cluster analysis. Prerequisites A Foundry project . One or more completed evaluation runs . Generate an evaluation cluster analysis On the evaluation detail page, select one or more runs and then select Cluster analysis . A window appears and prompts you to select a model to generate the cluster analysis. It also shows an estimated time and tokens based on the samples selected from the evaluation runs. View cluster analysis Cluster analysis provides an intuitive visualization of performance by grouping evaluation result samples with similar issues or response patterns. It helps you quickly identify recurring failure types, understand the distribution across error categories, and prioritize areas for improvement. At the top of the view, summary statistics for the evaluation run are displayed: Total samples: Total number of eval --- ## Human evaluation for agents ### Human evaluation for agents URL: https://learn.microsoft.com/en-us/azure/ai-foundry/observability/how-to/human-evaluation?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Set up human evaluation for your agents (preview) Feedback Summarize this article for me In this article Important Items marked (preview) in this article are currently in public preview. This preview is provided without a service-level agreement, and we don't recommend it for production workloads. Certain features might not be supported or might have constrained capabilities. For more information, see Supplemental Terms of Use for Microsoft Azure Previews . In this article, you’ll learn how to set up human evaluation for your Foundry agent. As an agent builder, you can create evaluation question templates focused on key aspects of interest and enable them to be answered for each agent response in the agent’s preview experience. This enables human evaluations by peers, data scientists, or compliance team members based on the defined templates. Once evaluations are completed, you can view and download the results directly from the Foundry portal for further analysis. Prerequisites A Foundry project with one or more agents . Application Insights configured for your project. Create a human evaluation template To begin human evaluation for your Foundry agent, you’ll first define a template that contains the set of questions you want human reviewers to complete based on agent responses. Steps to create a template Select the agent you want to evaluate from the agent table in the Agents tab. Navigate to the Human Evaluation tab under Evaluation . Select Create new template to start the template creation process. In the Create Human Evaluation Template pop-up, assign a name and description, edit or delete sample questions, and add new qu --- ## AI red teaming agent overview ### AI red teaming agent overview URL: https://learn.microsoft.com/en-us/azure/ai-foundry/concepts/ai-red-teaming-agent?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . AI Red Teaming Agent (preview) Feedback Summarize this article for me In this article Note This document refers to the Microsoft Foundry (classic) portal. 🔄 Switch to the Microsoft Foundry (new) documentation if you're using the new portal. Note This document refers to the Microsoft Foundry (new) portal. Important Items marked (preview) in this article are currently in public preview. This preview is provided without a service-level agreement, and we don't recommend it for production workloads. Certain features might not be supported or might have constrained capabilities. For more information, see Supplemental Terms of Use for Microsoft Azure Previews . The AI Red Teaming Agent is a powerful tool designed to help organizations proactively find safety risks associated with generative AI systems during design and development of generative AI models and applications. Traditional red teaming involves exploiting the cyber kill chain and describes the process by which a system is tested for security vulnerabilities. However, with the rise of generative AI, the term AI red teaming has been coined to describe probing for novel risks (both content and security related) that these systems present and refers to simulating the behavior of an adversarial user who is trying to cause your AI system to misbehave in a particular way. The AI Red Teaming Agent leverages Microsoft's open-source framework for Python Risk Identification Tool's ( PyRIT ) AI red teaming capabilities along with Microsoft Foundry's Risk and Safety Evaluations to help you automatically assess safety issues in three ways: Automated scans for content risks: Firstly, you ca --- ## Run red teaming scans locally ### Run red teaming scans locally URL: https://learn.microsoft.com/en-us/azure/ai-foundry/how-to/develop/run-scans-ai-red-teaming-agent?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Run AI Red Teaming Agent locally (preview) Feedback Summarize this article for me In this article Note This document refers to the Microsoft Foundry (classic) portal. 🔄 Switch to the Microsoft Foundry (new) documentation if you're using the new portal. Note This document refers to the Microsoft Foundry (new) portal. Important Items marked (preview) in this article are currently in public preview. This preview is provided without a service-level agreement, and we don't recommend it for production workloads. Certain features might not be supported or might have constrained capabilities. For more information, see Supplemental Terms of Use for Microsoft Azure Previews . The AI Red Teaming Agent (preview) is a powerful tool designed to help organizations proactively find safety risks associated with generative AI systems during design and development. The AI red teaming capabilities of Microsoft's open-source framework for Python Risk Identification Tool ( PyRIT ) are integrated directly into Microsoft Foundry. Teams can automatically scan their model and application endpoints for risks, simulate adversarial probing, and generate detailed reports. This article explains how to: Create an AI Red Teaming Agent locally with the Azure AI Evaluation SDK Run automated scans locally and view the results in Foundry Prerequisites A Foundry project or hubs based project. To learn more, see Create a project . If this is your first time running evaluations and logging it to your Microsoft Foundry project, you might need to do a few additional steps: Create and connect your storage account to your Foundry project at the resource level. There are t --- ## Run red teaming scans in the cloud ### Run red teaming scans in the cloud URL: https://learn.microsoft.com/en-us/azure/ai-foundry/how-to/develop/run-ai-red-teaming-cloud?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Run AI Red Teaming Agent in the cloud (preview) Feedback Summarize this article for me In this article Note This document refers to the Microsoft Foundry (classic) portal. 🔄 Switch to the Microsoft Foundry (new) documentation if you're using the new portal. Note This document refers to the Microsoft Foundry (new) portal. Important Items marked (preview) in this article are currently in public preview. This preview is provided without a service-level agreement, and we don't recommend it for production workloads. Certain features might not be supported or might have constrained capabilities. For more information, see Supplemental Terms of Use for Microsoft Azure Previews . Though the AI Red Teaming Agent (preview) can be run locally during prototyping and development to help identify safety risks, running them in the cloud allows for pre-deployment AI red teaming runs on larger combinations of attack strategies and risk categories for a fuller analysis. Though the AI Red Teaming Agent can be run locally during prototyping and development to help identify safety risks, running them in the cloud allows for the following scenarios: Pre-deployment AI red teaming runs on larger combinations of attack strategies and risk categories for a fuller analysis, Post-deployment continuous AI red teaming runs that can be scheduled to run at set time intervals Agentic-specific risk scenarios to support a minimally sandboxed environment for the AI red teaming run Prerequisites Note You must use a Foundry project for this feature. A hub-based project isn't supported. See How do I know which type of project I have? and Create a Foundry project . To --- ## Evaluation in GitHub Actions ### Evaluation in GitHub Actions URL: https://learn.microsoft.com/en-us/azure/ai-foundry/how-to/evaluation-github-action?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . How to run an evaluation in GitHub Action (preview) Feedback Summarize this article for me In this article Note This document refers to the Microsoft Foundry (classic) portal. 🔄 Switch to the Microsoft Foundry (new) documentation if you're using the new portal. Note This document refers to the Microsoft Foundry (new) portal. Important Items marked (preview) in this article are currently in public preview. This preview is provided without a service-level agreement, and we don't recommend it for production workloads. Certain features might not be supported or might have constrained capabilities. For more information, see Supplemental Terms of Use for Microsoft Azure Previews . This GitHub Action enables offline evaluation of AI models and agents within your CI/CD pipelines. It streamlines the evaluation process, so you can assess model performance and make informed decisions before deploying to production. Offline evaluation involves testing AI models and agents by using test datasets to measure their performance on various quality and safety metrics such as fluency, coherence, and appropriateness. After you select a model in the Foundry model catalog or GitHub Model marketplace , perform offline pre-production evaluation to validate the AI application during integration testing. This process allows developers to identify potential problems and make improvements before deploying the model or application to production, such as when creating and updating agents. This GitHub Action enables offline evaluation of Microsoft Foundry Agents within your CI/CD pipelines. It's designed to streamline the offline evaluation process, so you can --- ## Evaluation in Azure DevOps ### Evaluation in Azure DevOps URL: https://learn.microsoft.com/en-us/azure/ai-foundry/how-to/evaluation-azure-devops?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . How to run an evaluation in Azure DevOps (preview) Feedback Summarize this article for me In this article Note This document refers to the Microsoft Foundry (classic) portal. 🔄 Switch to the Microsoft Foundry (new) documentation if you're using the new portal. Note This document refers to the Microsoft Foundry (new) portal. Important Items marked (preview) in this article are currently in public preview. This preview is provided without a service-level agreement, and we don't recommend it for production workloads. Certain features might not be supported or might have constrained capabilities. For more information, see Supplemental Terms of Use for Microsoft Azure Previews . Similar to the Azure AI evaluation in GitHub Actions , an Azure DevOps extension is also available in the Azure DevOps Marketplace. By using this extension, you can evaluate AI agents offline within your CI/CD pipelines. This Azure DevOps extension enables offline evaluation of Microsoft Foundry Agents within your CI/CD pipelines. It streamlines the offline evaluation process, so you can identify potential problems and make improvements before releasing an update to production. To use this extension, provide a data set with test queries and a list of evaluators. This task invokes your agents with the queries, evaluates them, and generates a summary report. Features Automated Evaluation : Integrate offline evaluation into your CI/CD workflows to automate the pre-production assessment of AI models. Built-in Evaluators : Leverage existing evaluators provided by the Azure AI Evaluation SDK . The following evaluators are supported: Category Evaluator class/Metrics --- ## Monitoring dashboard insights with Foundry agent ### Monitoring dashboard insights with Foundry agent URL: https://learn.microsoft.com/en-us/azure/ai-foundry/observability/how-to/optimization-dashboard?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Monitoring dashboard insights in Microsoft Foundry with Ask AI (preview) Feedback Summarize this article for me In this article Important Items marked (preview) in this article are currently in public preview. This preview is provided without a service-level agreement, and we don't recommend it for production workloads. Certain features might not be supported or might have constrained capabilities. For more information, see Supplemental Terms of Use for Microsoft Azure Previews . After your agent is in production, set up and view various metrics in the monitoring dashboard or control plane dashboard. Use Ask AI—the built-in chat assistant—to get a summary of your dashboard data and recommendations for next steps without leaving the Foundry portal. This article describes the integrated user experience and system behavior for getting a dashboard summary or insights through Ask AI. Prerequisites A Foundry project with one or more published agents . Access to Ask AI (the chat assistant) in the Foundry portal. Start a chat with Ask AI You can start a chat with Ask AI from any page in the Foundry portal. Select the Ask AI icon at the top of the page. Select one of the predefined prompts from the Ask AI banner under Build/Model/Monitor , Build/Agent/Monitor , or overview page. Get a summary or insight of your dashboard You can ask questions like: "Give me a summary of the dashboard" "Analyze the performance trend of my dashboard" Or you can select a predefined prompt under the Ask AI banner, and the question is passed to Ask AI. Ask AI provides highlights and abnormal behavior insights on your dashboard for the selected time period and --- ## Monitor model deployments ### Monitor model deployments URL: https://learn.microsoft.com/en-us/azure/ai-foundry/foundry-models/how-to/monitor-models?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Monitor model deployments in Microsoft Foundry Models Feedback Summarize this article for me In this article Important Items marked (preview) in this article are currently in public preview. This preview is provided without a service-level agreement, and we don't recommend it for production workloads. Certain features might not be supported or might have constrained capabilities. For more information, see Supplemental Terms of Use for Microsoft Azure Previews . When you have critical applications and business processes that rely on Azure resources, you need to monitor and get alerts for your system. The Azure Monitor service collects and aggregates metrics and logs from every component of your system, including Foundry Models deployments. You can use this information to view availability, performance, and resilience, and get notifications of issues. This article explains how you can use metrics and logs to monitor model deployments in Foundry Models. Note Monitoring is only supported for OpenAI, Globalbatch sku & non-whisper models. Prerequisites To use monitoring capabilities for model deployments in Foundry Models, you need the following: A Microsoft Foundry resource . Tip If you're using serverless API endpoints and you want to take advantage of monitoring capabilities explained in this article, migrate your serverless API endpoints to Foundry Models . At least one model deployment. Access to diagnostic information for the resource. Metrics Azure Monitor collects metrics from Foundry Models automatically. No configuration is required . These metrics are: Stored in the Azure Monitor time-series metrics database. Lightweight an --- ## Agent tracing overview ### Agent tracing overview URL: https://learn.microsoft.com/en-us/azure/ai-foundry/observability/concepts/trace-agent-concept?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Agent tracing overview (preview) Feedback Summarize this article for me In this article Important Items marked (preview) in this article are currently in public preview. This preview is provided without a service-level agreement, and we don't recommend it for production workloads. Certain features might not be supported or might have constrained capabilities. For more information, see Supplemental Terms of Use for Microsoft Azure Previews . Microsoft Foundry provides an observability platform for monitoring and tracing AI agents. It captures key details during an agent run, such as inputs, outputs, tool usage, retries, latencies, and costs. Understanding the reasoning behind your agent's executions is important for troubleshooting and debugging. However, understanding complex agents presents challenges for several reasons: There could be a high number of steps involved in generating a response, making it hard to keep track of all of them. The sequence of steps might vary based on user input. The inputs/outputs at each stage might be long and deserve more detailed inspection. Each step of an agent's runtime might also involve nesting. For example, an agent might invoke a tool, which uses another process, which then invokes another tool. If you notice strange or incorrect output from a top-level agent run, it might be difficult to determine exactly where in the execution the issue was introduced. Trace results solve this by allowing you to view the inputs and outputs of each primitive involved in a particular agent run, displayed in the order they were invoked, making it easy to understand and debug your AI agent's behavior. Prere --- ## Set up your developer environment ### Set up your developer environment URL: https://learn.microsoft.com/en-us/azure/ai-foundry/how-to/develop/install-cli-sdk?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Prepare your development environment Feedback Summarize this article for me In this article Set up your development environment to use the Microsoft Foundry SDK. You also need Azure CLI for authentication so that your code can access your user credentials. Important This article covers general prerequisites only, such as language runtimes, global tools, and VS Code and extension setup. It doesn't cover scenario-specific steps like SDK installation or authentication. When your environment is ready, continue to the quickstart for those instructions. Important This article covers general prerequisites only, such as language runtimes, global tools, and VS Code and extension setup. It doesn't cover scenario-specific steps like SDK installation or authentication. When your environment is ready, continue to the quickstart for those instructions. Prerequisites An Azure account with an active subscription. If you don't have one, create a free Azure account, which includes a free trial subscription . Download, install, and configure Visual Studio Code, or the IDE of your choice. For more information, see Download Visual Studio Code . To create and manage Foundry resources, one of the following Azure RBAC roles Azure AI Project Manager (for managing Foundry projects) Contributor or Owner (for subscription-level permissions) To use project but not create new resources, you need at least: Azure AI User on the projects you use (least-privilege role for development) For details on each role's permissions, see Role-based access control for Microsoft Foundry . Install your programming language In Visual Studio Code, create a new folder for your --- ## Work in VS Code ### Work in VS Code URL: https://learn.microsoft.com/en-us/azure/ai-foundry/how-to/develop/get-started-projects-vs-code?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Work with the Microsoft Foundry for Visual Studio Code extension (Preview) Feedback Summarize this article for me In this article Microsoft Foundry provides a unified platform for enterprise AI operations, model builders, and application development. This foundation combines production-grade infrastructure with friendly interfaces, ensuring organizations can build and operate AI applications with confidence. With Foundry, you can: Deploy the latest language models from Microsoft, OpenAI, Meta, DeepSeek, and more using the robust model catalog Test the deployed models in a model playground Quickly get started with developing generative AI applications using a collection of Azure curated code templates Configure and deploy agents with Foundry Agent Service With the Foundry for Visual Studio Code extension, you can accomplish much of this workflow directly from Visual Studio Code. It also comes with other features, such as code templates, playgrounds, and integration with other VS Code extensions and features. This article shows you how to quickly get started using the features of the Foundry for Visual Studio Code extension. Important Items marked (preview) in this article are currently in public preview. This preview is provided without a service-level agreement, and we don't recommend it for production workloads. Certain features might not be supported or might have constrained capabilities. For more information, see Supplemental Terms of Use for Microsoft Azure Previews . Prerequisites Before using the Foundry for Visual Studio Code extension, you must: Download, install, and configure Visual Studio Code. More information: Down --- ## Start with an AI template ### Start with an AI template URL: https://learn.microsoft.com/en-us/azure/ai-foundry/how-to/develop/ai-template-get-started?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Get started with an AI template Feedback Summarize this article for me In this article Note This document refers to the Microsoft Foundry (classic) portal. 🔄 Switch to the Microsoft Foundry (new) documentation if you're using the new portal. Note This document refers to the Microsoft Foundry (new) portal. Find, explore, and deploy AI solution templates from the Foundry portal. Streamline your code-first development with prebuilt, task-specific Azure AI templates. These ready-to-use, customizable templates help you skip setup, reduce friction, and deliver value faster with trusted, scalable infrastructure. Built on insights from over 2,000 customer engagements, AI solution templates significantly reduce time from concept to production at scale. AI solution templates include customizable code samples, pre-integrated Azure services, and GitHub-hosted quick-start guides. Development teams can focus on outcomes instead of setup, building solutions for popular use cases like live voice agents, release management, and data unification. These templates harness the power of multi-agent, agentic AI, enabling you to automate workflows, optimize operations, reduce costs, and make faster, data-driven decisions. Important Starter templates, manifests, code samples, and other resources made available by Microsoft or its partners ("samples") are designed to assist in accelerating development of agents and AI solutions for specific scenarios. Review all provided resources and carefully test output behavior in the context of your use case. AI responses might be inaccurate and AI actions should be monitored with human oversight. Learn more in the --- ## Use the Visual Studio Code extension for agent development ### Use the Visual Studio Code extension for agent development URL: https://learn.microsoft.com/en-us/azure/ai-foundry/how-to/develop/vs-code-agents?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Work with Foundry Agent Service in Visual Studio Code (preview) Feedback Summarize this article for me In this article After you get started with the Microsoft Foundry for Visual Studio Code extension , use Foundry Agent Service to build agents. Agents are microservices that: Answer questions by using their training data or search other sources with retrieval-augmented generation (RAG). Perform specific actions. Automate complete workflows. Agents combine AI models with tools to access and interact with your data. Foundry developers can stay productive by developing, testing, and deploying agents in the familiar environment of Visual Studio Code (VS Code). Important Items marked (preview) in this article are currently in public preview. This preview is provided without a service-level agreement, and we don't recommend it for production workloads. Certain features might not be supported or might have constrained capabilities. For more information, see Supplemental Terms of Use for Microsoft Azure Previews . Create and edit an Azure AI agent within the designer view Follow these steps to create an Azure AI agent: Sign in to your Azure resources . Set your default project . Deploy a model to use with your agent. In the Foundry Extension view, find the Resources section. Select the plus ( + ) icon next to the Agents subsection to create a new AI agent. Interact with your agent in the designer After you choose your save location, both the agent .yaml file and the designer view open so that you can edit your AI agent. Perform the following tasks in the agent designer: In the prompt, enter a name for your agent. In the dropdown list, s --- ## Use MCP Tools in the Visual Studio Code extension ### Use MCP Tools in the Visual Studio Code extension URL: https://learn.microsoft.com/en-us/azure/ai-foundry/how-to/develop/vs-code-agents-mcp?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Work with Foundry Agent Service and MCP server tools in Visual Studio Code (preview) Feedback Summarize this article for me In this article In this article, you learn how to add and use Model Context Protocol (MCP) tools with Azure AI agents by using the Microsoft Foundry for Visual Studio Code extension . After you build an agent in Foundry Agent Service by using this Visual Studio Code (VS Code) extension, you can add MCP tools to your agent. Using or building an MCP server allows your agent to: Access up-to-date information from your APIs and services. Retrieve relevant context to enhance the quality of responses from your AI models. Agents combine AI models with tools to access and interact with your data. Foundry developers can stay productive by developing, testing, and deploying MCP tool-calling agents in the familiar environment of VS Code. Important Items marked (preview) in this article are currently in public preview. This preview is provided without a service-level agreement, and we don't recommend it for production workloads. Certain features might not be supported or might have constrained capabilities. For more information, see Supplemental Terms of Use for Microsoft Azure Previews . Create an Azure AI agent within the designer view To create an Azure AI agent, follow the steps in Create and edit Azure AI agents within the designer view . Add an existing MCP server tool to the AI agent After you create your agent, you can add tools to it, including MCP tools. For more information about available tools, see Tools for Azure AI agents . You can bring multiple remote MCP servers by adding them as tools. For each tool, --- ## Use declarative agent workflows in the Visual Studio Code extension ### Use declarative agent workflows in the Visual Studio Code extension URL: https://learn.microsoft.com/en-us/azure/ai-foundry/agents/how-to/vs-code-agents-workflow-low-code?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Work with Declarative (Low-code) Agent workflows in Visual Studio Code (preview) Feedback Summarize this article for me In this article In this article, you learn how to add and use Foundry Agent workflows with Azure AI agents by using the Microsoft Foundry for Visual Studio Code extension . After you build an agent in Foundry Agent Service by using this Visual Studio Code (VS Code) extension, you can add workflows to your agent. Foundry Workflows is a UI-based tool in Foundry that creates declarative, predefined sequences of actions including agents, as in Microsoft Agent Framework Workflows. Workflows let you build intelligent automation systems that blend AI agents with business processes in a visual way. Traditional single-agent systems struggle to handle complex tasks with many parts. When you orchestrate multiple agents, each with specialized skills or roles, you create systems that are more robust, adaptive, and capable of solving real-world problems together. What is a declarative agent? A declarative agent is an AI agent that operates based on predefined rules, workflows, or configurations instead of explicit programming logic. This approach lets users define what the agent should do and how it should behave through high-level specifications. Declarative agents make it easier to create and manage complex interactions without deep coding knowledge. View a declarative agent workflow For declarative agent workflows start by creating a workflow in Microsoft Foundry . The following sections guide you through the steps to view and test a simple workflow that uses an agent to process user input. In Foundry, navigate to your pr --- ## Use hosted agent workflows in the Visual Studio Code extension ### Use hosted agent workflows in the Visual Studio Code extension URL: https://learn.microsoft.com/en-us/azure/ai-foundry/agents/how-to/vs-code-agents-workflow-pro-code?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Work with Hosted (Pro-code) Agents in Visual Studio Code (preview) Feedback Summarize this article for me In this article In this article, you learn how to add and use hosted Foundry Agent workflows with Azure AI agents by using the Microsoft Foundry for Visual Studio Code extension . After you build an agent in Foundry Agent Service by using this Visual Studio Code (VS Code) extension, you can add hosted agent workflows to your agent. Foundry developers can stay productive by developing, testing, and deploying hosted agent workflows in the familiar environment of VS Code. Create a hosted agent workflow You can use the Foundry for Visual Studio Code extension to create hosted agent workflows. A hosted agent workflow is a sequence of agents that work together to accomplish a task. Each agent in the workflow can have its own model, tools, and instructions. Open the command palette ( Ctrl + Shift + P ). Run this command: >Microsoft Foundry: Create a New Hosted Agent . Select a programming language. Select a folder where you want to save your new workflow. Enter a name for your workflow project. A new folder is created with the necessary files for your hosted agent project, including a sample code file to get you started. Install dependencies Install the required dependencies for your hosted agent project. The dependencies vary based on the programming language that you selected when you created the project. Prerequisites To run the sample hosted agent Python project, make sure you install Python 3.10 or higher. You also need a Foundry project with a deployed model, or an Azure OpenAI resource. Give the project's managed identity th --- ## Get started with Foundry MCP Server ### Get started with Foundry MCP Server URL: https://learn.microsoft.com/en-us/azure/ai-foundry/mcp/get-started?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Get started with Foundry MCP Server (preview) using Visual Studio Code Feedback Summarize this article for me In this article Foundry MCP Server (preview) is a Microsoft-managed, cloud-hosted implementation of the Model Context Protocol (MCP). It exposes curated tools that let your agents perform read and write operations against Foundry services without calling backend APIs directly. Use an MCP-compliant client such as Visual Studio Code to connect to the public endpoint, authenticate with Entra ID, and let LLMs access the tools. After you connect, you can build agents that invoke these tools with natural language prompts. In this article, you learn how to: Connect to Foundry MCP Server with GitHub Copilot in Visual Studio Code Run prompts to test Foundry MCP Server tools and interact with Azure resources Note This feature is currently in public preview. This preview is provided without a service-level agreement, and we don't recommend it for production workloads. Certain features might not be supported or might have constrained capabilities. For more information, see Supplemental Terms of Use for Microsoft Azure Previews . Prerequisites Azure account with an active subscription. If you don't have one, create a free Azure account . Foundry project. If you don't have a project, create one with the Microsoft Foundry SDK Quickstart . Visual Studio Code . GitHub Copilot Visual Studio Code extension. Benefits of Foundry MCP Server Cloud-hosted interface for AI tool orchestration : Foundry MCP Server (preview) provides a secure, scalable endpoint for MCP-compliant clients. You don't need to deploy infrastructure, enabling seamless in --- ## Best practices and security guidance ### Best practices and security guidance URL: https://learn.microsoft.com/en-us/azure/ai-foundry/mcp/security-best-practices?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Foundry MCP Server best practices and security guidance Feedback Summarize this article for me In this article Use Foundry MCP Server (preview) tools to automate read and write operations across Foundry resources (deployments, datasets, evaluations, monitoring, analytics). This guidance helps you verify intent, reduce risk, and apply security and governance practices before you run MCP tools. Note This feature is currently in public preview. This preview is provided without a service-level agreement, and we don't recommend it for production workloads. Certain features might not be supported or might have constrained capabilities. For more information, see Supplemental Terms of Use for Microsoft Azure Previews . Interpreting the response MCP Server provides output that is passed to the language model selected for your agent (for example, Visual Studio Code with GitHub Copilot). The language model combines this output with the conversation context to generate a final response based on its capabilities. Always verify the accuracy of the language model’s response. It may include details that are inferred or generated beyond the MCP Server’s original output. Impact of write operations Write operations have a critical impact on Foundry resources. Proceed with caution and proper planning when you interact with Foundry MCP Server (preview), just as you would when using the portal, SDKs, or REST APIs. For example: Deployments: Immediately affect live apps and billing. Deletions: Permanently remove resources and can break dependent services. Evaluations: Consume compute quota and incur costs. Datasets: Can overwrite existing versions. Exa --- ## Available tools and sample prompts ### Available tools and sample prompts URL: https://learn.microsoft.com/en-us/azure/ai-foundry/mcp/available-tools?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Available tools and example prompts for Foundry MCP Server (preview) Feedback Summarize this article for me In this article Use the following sections to find available tools and example prompts for Foundry MCP Server (preview). Foundry MCP Server lets you use conversational prompts instead of API calls to interact with Foundry services. Note This feature is currently in public preview. This preview is provided without a service-level agreement, and we don't recommend it for production workloads. Certain features might not be supported or might have constrained capabilities. For more information, see Supplemental Terms of Use for Microsoft Azure Previews . Dataset management evaluation_dataset_create (write) Create or update a dataset version in a Foundry project. Example prompts include: "Upload my customer support Q&A dataset from this Azure Blob Storage URL." "Create a new dataset version 2.0 for my training data located at ." "Register a new evaluation dataset called product-reviews-v1 from my blob storage." evaluation_dataset_get (read) Get a dataset by name and version, or list all datasets in the project. Example prompts include: "Show me all datasets in my Foundry project" "Get details for the 'customer-support-qa' dataset version 2" "List all available datasets I can use for evaluation" Evaluation operations evaluation_create (write) Create an evaluation run for a dataset using one or more evaluators. Example prompts include: "Create an evaluation run for my customer service dataset using Relevance, Groundedness, and Coherence evaluators." "Run an evaluation on dataset-456 with Violence, HateU --- ## Microsoft Foundry SDKs ### Microsoft Foundry SDKs URL: https://learn.microsoft.com/en-us/azure/ai-foundry/how-to/develop/sdk-overview?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Microsoft Foundry SDKs and Endpoints Feedback Summarize this article for me In this article Note This document refers to the Microsoft Foundry (classic) portal. 🔄 Switch to the Microsoft Foundry (new) documentation if you're using the new portal. Note This document refers to the Microsoft Foundry (new) portal. Creating a Foundry resource unlocks access to models, agents, and tools through a unified set of SDKs and endpoints. This article covers what each SDK is for and which endpoint to use. SDK What it's for Endpoint Foundry SDK Foundry-specific capabilities with OpenAI-compatible interfaces. Includes access to Foundry direct models through the Responses API (not Chat Completions). https://.services.ai.azure.com/api/projects/ OpenAI SDK Latest OpenAI SDK models and features with the full OpenAI API surface. Foundry direct models available through Chat Completions API (not Responses). https://.openai.azure.com/openai/v1 Foundry Tools SDKs Prebuilt solutions (Vision, Speech, Content Safety, and more). Tool-specific endpoints (varies by service). Agent Framework Multi-agent orchestration in code. Cloud-agnostic. Uses the project endpoint via the Foundry SDK. Note Resource types: A Foundry resource provides all endpoints previously listed. An Azure OpenAI resource provides only the /openai/v1 endpoint. Authentication: Samples here use Microsoft Entra ID ( DefaultAzureCredential ). API keys work on /openai/v1 . Pass the key as api_key instead of a token provider. Prerequisites An Azure account with an active subscription. If you don't have one, create a free Azure account, which includes a --- ## Endpoints for Foundry Models ### Endpoints for Foundry Models URL: https://learn.microsoft.com/en-us/azure/ai-foundry/foundry-models/concepts/endpoints?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Endpoints for Microsoft Foundry Models Feedback Summarize this article for me In this article Note This document refers to the Microsoft Foundry (classic) portal. 🔄 Switch to the Microsoft Foundry (new) documentation if you're using the new portal. Note This document refers to the Microsoft Foundry (new) portal. Microsoft Foundry Models enables you to access the most powerful models from leading model providers through a single endpoint and set of credentials. This capability lets you switch between models and use them in your application without changing any code. This article explains how the Foundry services organize models and how to use the inference endpoint to access them. Important If you're currently using an Azure AI Inference beta SDK with Microsoft Foundry Models or Azure OpenAI service, we strongly recommend that you transition to the generally available OpenAI/v1 API , which uses an OpenAI stable SDK. For more information on how to migrate to the OpenAI/v1 API by using an SDK in your programming language of choice, see Migrate from Azure AI Inference SDK to OpenAI SDK . Deployments Foundry uses deployments to make models available. Deployments give a model a name and set specific configurations. You can access a model by using its deployment name in your requests. A deployment includes: A model name A model version A provisioning or capacity type 1 A content filtering configuration 1 A rate limiting configuration 1 1 These configurations can change depending on the selected model. A Foundry resource can have many model deployments. You only pay for inference performed on model deployments. Deployments are Azure res --- ## C# ### C# URL: https://learn.microsoft.com/dotnet/api/overview/azure/ai.projects-readme?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Azure AI Projects client library for .NET - version 1.1.0 Feedback Summarize this article for me In this article The AI Projects client library is part of the Azure AI Foundry SDK and provides easy access to resources in your Azure AI Foundry Project. Use it to: Create and run Agents using the GetPersistentAgentsClient method on the client. Enumerate AI Models deployed to your Foundry Project using the Deployments operations. Enumerate connected Azure resources in your Foundry project using the Connections operations. Upload documents and create Datasets to reference them using the Datasets operations. Create and enumerate Search Indexes using the Indexes operations. The client library uses version v1 of the AI Foundry data plane REST APIs . Product documentation | Samples | API reference documentation | Package (NuGet) | SDK source code Table of contents Getting started Prerequisites Install the package Key concepts Create and authenticate the client Examples Performing Agent operations Get an authenticated AzureOpenAI client Get an authenticated ChatCompletionsClient Deployments operations Connections operations Dataset operations Indexes operations Troubleshooting Next steps Contributing Getting started Prerequisites To use Azure AI Projects capabilities, you must have an Azure subscription . This will allow you to create an Azure AI resource and get a connection URL. Install the package Install the client library for .NET with NuGet : dotnet add package Azure.AI.Projects --prerelease Authenticate the client A secure, keyless authentication approach is to use Microsoft Entra ID (formerly Azure Active Directory) via the Azure --- ## JavaScript ### JavaScript URL: https://learn.microsoft.com/javascript/api/overview/azure/ai-projects-readme?view=azure-node-preview&preserve-view=true Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Azure AI Projects client library for JavaScript - version 2.0.0-beta.4 Feedback Summarize this article for me In this article The AI Projects client library (in preview) is part of the Microsoft Foundry SDK, and provides easy access to resources in your Microsoft Foundry Project. Use it to: Create and run Agents using the .agents property on the client. Enhance Agents with specialized tools : Agent Memory Search Agent-to-Agent (A2A) Azure AI Search Bing Custom Search Bing Grounding Browser Automation Code Interpreter Computer Use File Search Function Tool Image Generation Microsoft Fabric Model Context Protocol (MCP) OpenAPI SharePoint Web Search Get an OpenAI client using the .getOpenAIClient. method to run Responses, Conversations, Evals and FineTuning operations with your Agent. Manage memory stores for Agent conversations, using the .memoryStores operations. Explore additional evaluation tools to assess the performance of your generative AI application, using the .evaluationRules , .evaluationTaxonomies , .evaluators , .insights , and .schedules operations. Run Red Team scans to identify risks associated with your generative AI application, using the ".redTeams" operations. Fine tune AI Models on your data. Enumerate AI Models deployed to your Foundry Project using the .deployments operations. Enumerate connected Azure resources in your Foundry project using the .connections operations. Upload documents and create Datasets to reference them using the .datasets operations. Create and enumerate Search Indexes using the .indexes operations. The client library uses version 2025-11-15-preview of the Microsoft Foundry data plane R --- ## Python ### Python URL: https://learn.microsoft.com/python/api/overview/azure/ai-projects-readme?view=azure-python-preview&preserve-view=true Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Azure AI Projects client library for Python - version 2.0.0b3 Feedback Summarize this article for me In this article The AI Projects client library (in preview) is part of the Microsoft Foundry SDK, and provides easy access to resources in your Microsoft Foundry Project. Use it to: Create and run Agents using methods on methods on the .agents client property. Enhance Agents with specialized tools : Agent Memory Search Agent-to-Agent (A2A) Azure AI Search Bing Custom Search Bing Grounding Browser Automation Code Interpreter Computer Use File Search Function Tool Image Generation Microsoft Fabric Model Context Protocol (MCP) OpenAPI SharePoint Web Search Get an OpenAI client using .get_openai_client() method to run Responses, Conversations, Evals and FineTuning operations with your Agent. Manage memory stores for Agent conversations, using the .memory_stores operations. Explore additional evaluation tools to assess the performance of your generative AI application, using the .evaluation_rules , .evaluation_taxonomies , .evaluators , .insights , and .schedules operations. Run Red Team scans to identify risks associated with your generative AI application, using the ".red_teams" operations. Fine tune AI Models on your data. Enumerate AI Models deployed to your Foundry Project using the .deployments operations. Enumerate connected Azure resources in your Foundry project using the .connections operations. Upload documents and create Datasets to reference them using the .datasets operations. Create and enumerate Search Indexes using methods the .indexes operations. The client library uses version 2025-11-15-preview of the AI Foundry da --- ## REST API ### REST API URL: https://learn.microsoft.com/rest/api/aifoundry/aiprojects/?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Azure AI Projects REST API reference Feedback Summarize this article for me In this article The Azure AI Projects REST API lets you work with AI projects , the central resource that brings together models, data, tools, and telemetry in Azure AI Foundry. Using familiar HTTP verbs you can create a project, configure secure connections to other Azure services, and retrieve ready-to-use endpoints for inferencing, agents, evaluation, and more. Because every service you attach lives behind a single project connection string, applications only need to know one credential and one base URL to reach everything they need. Authentication The Azure AI Projects REST API supports Microsoft Entra ID tokens for authentication. Azure resource keys are not yet supported. Access is controlled by Azure role-based access control (RBAC). For more details, see What is Azure role-based access control (Azure RBAC)? and Role-based access control for Azure AI Foundry . Versioning Azure AI Projects REST API provides two types of versioning: Evolving Version : These versions follow the format v# (e.g., v1, v2, v3). As the name suggests, evolving versions incorporate non-breaking changes to the contract, such as the addition of optional parameters to requests or new properties to responses. Clients using these versions should be designed to handle additive changes. Pinned Version : These are date-based versions that do not include contract changes. The following table illustrates how these versions work. Timeline Evolving Version Pinned Version Notes 25-May v1 5/1/2025 Customers who do not want any changes to the contract can use the date-based version. Custo --- ## REST API (preview) ### REST API (preview) URL: https://learn.microsoft.com/en-us/azure/ai-foundry/reference/foundry-project-rest-preview?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Microsoft Foundry Project REST reference Feedback Summarize this article for me In this article API Version: 2025-11-15-preview Agents - create agent POST {endpoint}/agents?api-version=2025-11-15-preview Creates the agent. URI Parameters Name In Required Type Description endpoint path Yes string url Foundry Project endpoint in the form https://{ai-services-account-name}.services.ai.azure.com/api/projects/{project-name} . If you only have one Project in your Foundry Hub, or to target the default Project in your Hub, use the form https://{ai-services-account-name}.services.ai.azure.com/api/projects/_project api-version query Yes string The API version to use for this operation. Request Header Name Required Type Description Authorization True string Example: Authorization: Bearer {Azure_AI_Foundry_Project_Auth_Token} To generate an auth token using Azure CLI: az account get-access-token --resource https://ai.azure.com/ Type: oauth2 Authorization Url: https://login.microsoftonline.com/common/oauth2/v2.0/authorize scope: https://ai.azure.com/.default Request Body Content-Type : application/json Name Type Description Required Default definition object Yes └─ kind AgentKind No └─ rai_config RaiConfig Configuration for Responsible AI (RAI) content filtering and safety features. No description string A human-readable description of the agent. No metadata object Set of 16 key-value pairs that can be attached to an object. This can be useful for storing additional information about the object in a structured format, and querying for objects via API or the dashboard. Keys are strings with a maximum length of 64 characters. Values are string --- ## API lifecycle ### API lifecycle URL: https://learn.microsoft.com/en-us/azure/ai-foundry/openai/api-version-lifecycle?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Azure OpenAI in Microsoft Foundry Models API lifecycle Feedback Summarize this article for me In this article This article is to help you understand the support lifecycle for Azure OpenAI APIs. Note New API response objects may be added to the API response at any time. We recommend you only parse the response objects you require. API evolution Previously, Azure OpenAI received monthly updates of new API versions. Taking advantage of new features required constantly updating code and environment variables with each new API release. Azure OpenAI also required the extra step of using Azure specific clients which created overhead when migrating code between OpenAI and Azure OpenAI. Starting in August 2025, you can now opt in to our next generation v1 Azure OpenAI APIs which add support for: Ongoing access to the latest features with no need to specify new api-version 's each month. Faster API release cycle with new features launching more frequently. OpenAI client support with minimal code changes to swap between OpenAI and Azure OpenAI when using key-based authentication. OpenAI client support for token based authentication and automatic token refresh without the need to take a dependency on a separate Azure OpenAI client. Make chat completions calls with models from other providers like DeepSeek and Grok which support the v1 chat completions syntax. Access to new API calls that are still in preview will be controlled by passing feature specific preview headers allowing you to opt in to the features you want, without having to swap API versions. Alternatively, some features will indicate preview status through their API path and do --- ## Dataplane SDK language support ### Dataplane SDK language support URL: https://learn.microsoft.com/en-us/azure/ai-foundry/openai/supported-languages?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Azure OpenAI supported programming languages Feedback Summarize this article for me In this article Note This document refers to the Microsoft Foundry (classic) portal. 🔄 Switch to the Microsoft Foundry (new) documentation if you're using the new portal. Note This document refers to the Microsoft Foundry (new) portal. Source code | Package (NuGet) Azure OpenAI API version support v1 Generally Available (GA) API now allows access to both GA and Preview operations. To learn more, see the API version lifecycle guide . Installation dotnet add package OpenAI Authentication Microsoft Entra ID API Key A secure, keyless authentication approach is to use Microsoft Entra ID (formerly Azure Active Directory) via the Azure Identity library . To use the library: dotnet add package Azure.Identity Use the desired credential type from the library. For example, DefaultAzureCredential : using Azure.Identity; using OpenAI; using OpenAI.Chat; using System.ClientModel.Primitives; #pragma warning disable OPENAI001 BearerTokenPolicy tokenPolicy = new( new DefaultAzureCredential(), "https://cognitiveservices.azure.com/.default"); ChatClient client = new( model: "gpt-4.1-nano", authenticationPolicy: tokenPolicy, options: new OpenAIClientOptions() { Endpoint = new Uri("https://YOUR-RESOURCE-NAME.openai.azure.com/openai/v1") } ); ChatCompletion completion = client.CompleteChat("Tell me about the bitter lesson.'"); Console.WriteLine($"[ASSISTANT]: {completion.Content[0].Text}"); For more information about Azure OpenAI keyless authentication, see the " Get started with the Azure OpenAI security building block " QuickStart article. using OpenAI; using OpenAI --- ## v1 API ### v1 API URL: https://learn.microsoft.com/en-us/azure/ai-foundry/openai/latest?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Azure OpenAI in Microsoft Foundry Models v1 REST API reference Feedback Summarize this article for me In this article Only a subset of operations are currently supported with the v1 API. To learn more, see the API version lifecycle guide . v1 OpenAPI 3.0 spec Create chat completion POST {endpoint}/openai/v1/chat/completions Creates a chat completion. Parameters Name In Required Type Description endpoint path Yes string url Supported Azure OpenAI endpoints (protocol and hostname, for example: https://aoairesource.openai.azure.com . Replace "aoairesource" with your Azure OpenAI resource name). https://{your-resource-name}.openai.azure.com api-version query No The explicit Microsoft Foundry Models API version to use for this request. v1 if not otherwise specified. Request Header Use either token based authentication or API key. Authenticating with token based authentication is recommended and more secure. Name Required Type Description Authorization True string Example: Authorization: Bearer {Azure_OpenAI_Auth_Token} To generate an auth token using Azure CLI: az account get-access-token --resource https://cognitiveservices.azure.com Type: oauth2 Authorization Url: https://login.microsoftonline.com/common/oauth2/v2.0/authorize scope: https://cognitiveservices.azure.com/.default api-key True string Provide Azure OpenAI API key here Request Body Content-Type : application/json Name Type Description Required Default audio object Parameters for audio output. Required when audio output is requested with modalities: ["audio"] . No └─ format enum Specifies the output audio format. Must be one of wav , mp3 , flac , opus , or pcm16 . Possibl --- ## Chat ### Chat URL: https://learn.microsoft.com/en-us/azure/ai-foundry/openai/latest#create-chat-completion?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Azure OpenAI in Microsoft Foundry Models v1 REST API reference Feedback Summarize this article for me In this article Only a subset of operations are currently supported with the v1 API. To learn more, see the API version lifecycle guide . v1 OpenAPI 3.0 spec Create chat completion POST {endpoint}/openai/v1/chat/completions Creates a chat completion. Parameters Name In Required Type Description endpoint path Yes string url Supported Azure OpenAI endpoints (protocol and hostname, for example: https://aoairesource.openai.azure.com . Replace "aoairesource" with your Azure OpenAI resource name). https://{your-resource-name}.openai.azure.com api-version query No The explicit Microsoft Foundry Models API version to use for this request. v1 if not otherwise specified. Request Header Use either token based authentication or API key. Authenticating with token based authentication is recommended and more secure. Name Required Type Description Authorization True string Example: Authorization: Bearer {Azure_OpenAI_Auth_Token} To generate an auth token using Azure CLI: az account get-access-token --resource https://cognitiveservices.azure.com Type: oauth2 Authorization Url: https://login.microsoftonline.com/common/oauth2/v2.0/authorize scope: https://cognitiveservices.azure.com/.default api-key True string Provide Azure OpenAI API key here Request Body Content-Type : application/json Name Type Description Required Default audio object Parameters for audio output. Required when audio output is requested with modalities: ["audio"] . No └─ format enum Specifies the output audio format. Must be one of wav , mp3 , flac , opus , or pcm16 . Possibl --- ## Evals (preview) ### Evals (preview) URL: https://learn.microsoft.com/en-us/azure/ai-foundry/openai/latest#list-evals?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Azure OpenAI in Microsoft Foundry Models v1 REST API reference Feedback Summarize this article for me In this article Only a subset of operations are currently supported with the v1 API. To learn more, see the API version lifecycle guide . v1 OpenAPI 3.0 spec Create chat completion POST {endpoint}/openai/v1/chat/completions Creates a chat completion. Parameters Name In Required Type Description endpoint path Yes string url Supported Azure OpenAI endpoints (protocol and hostname, for example: https://aoairesource.openai.azure.com . Replace "aoairesource" with your Azure OpenAI resource name). https://{your-resource-name}.openai.azure.com api-version query No The explicit Microsoft Foundry Models API version to use for this request. v1 if not otherwise specified. Request Header Use either token based authentication or API key. Authenticating with token based authentication is recommended and more secure. Name Required Type Description Authorization True string Example: Authorization: Bearer {Azure_OpenAI_Auth_Token} To generate an auth token using Azure CLI: az account get-access-token --resource https://cognitiveservices.azure.com Type: oauth2 Authorization Url: https://login.microsoftonline.com/common/oauth2/v2.0/authorize scope: https://cognitiveservices.azure.com/.default api-key True string Provide Azure OpenAI API key here Request Body Content-Type : application/json Name Type Description Required Default audio object Parameters for audio output. Required when audio output is requested with modalities: ["audio"] . No └─ format enum Specifies the output audio format. Must be one of wav , mp3 , flac , opus , or pcm16 . Possibl --- ## Files ### Files URL: https://learn.microsoft.com/en-us/azure/ai-foundry/openai/latest#create-file?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Azure OpenAI in Microsoft Foundry Models v1 REST API reference Feedback Summarize this article for me In this article Only a subset of operations are currently supported with the v1 API. To learn more, see the API version lifecycle guide . v1 OpenAPI 3.0 spec Create chat completion POST {endpoint}/openai/v1/chat/completions Creates a chat completion. Parameters Name In Required Type Description endpoint path Yes string url Supported Azure OpenAI endpoints (protocol and hostname, for example: https://aoairesource.openai.azure.com . Replace "aoairesource" with your Azure OpenAI resource name). https://{your-resource-name}.openai.azure.com api-version query No The explicit Microsoft Foundry Models API version to use for this request. v1 if not otherwise specified. Request Header Use either token based authentication or API key. Authenticating with token based authentication is recommended and more secure. Name Required Type Description Authorization True string Example: Authorization: Bearer {Azure_OpenAI_Auth_Token} To generate an auth token using Azure CLI: az account get-access-token --resource https://cognitiveservices.azure.com Type: oauth2 Authorization Url: https://login.microsoftonline.com/common/oauth2/v2.0/authorize scope: https://cognitiveservices.azure.com/.default api-key True string Provide Azure OpenAI API key here Request Body Content-Type : application/json Name Type Description Required Default audio object Parameters for audio output. Required when audio output is requested with modalities: ["audio"] . No └─ format enum Specifies the output audio format. Must be one of wav , mp3 , flac , opus , or pcm16 . Possibl --- ## Fine-tuning ### Fine-tuning URL: https://learn.microsoft.com/en-us/azure/ai-foundry/openai/latest#create-fine-tuning-job?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Azure OpenAI in Microsoft Foundry Models v1 REST API reference Feedback Summarize this article for me In this article Only a subset of operations are currently supported with the v1 API. To learn more, see the API version lifecycle guide . v1 OpenAPI 3.0 spec Create chat completion POST {endpoint}/openai/v1/chat/completions Creates a chat completion. Parameters Name In Required Type Description endpoint path Yes string url Supported Azure OpenAI endpoints (protocol and hostname, for example: https://aoairesource.openai.azure.com . Replace "aoairesource" with your Azure OpenAI resource name). https://{your-resource-name}.openai.azure.com api-version query No The explicit Microsoft Foundry Models API version to use for this request. v1 if not otherwise specified. Request Header Use either token based authentication or API key. Authenticating with token based authentication is recommended and more secure. Name Required Type Description Authorization True string Example: Authorization: Bearer {Azure_OpenAI_Auth_Token} To generate an auth token using Azure CLI: az account get-access-token --resource https://cognitiveservices.azure.com Type: oauth2 Authorization Url: https://login.microsoftonline.com/common/oauth2/v2.0/authorize scope: https://cognitiveservices.azure.com/.default api-key True string Provide Azure OpenAI API key here Request Body Content-Type : application/json Name Type Description Required Default audio object Parameters for audio output. Required when audio output is requested with modalities: ["audio"] . No └─ format enum Specifies the output audio format. Must be one of wav , mp3 , flac , opus , or pcm16 . Possibl --- ## Models ### Models URL: https://learn.microsoft.com/en-us/azure/ai-foundry/openai/latest#list-models?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Azure OpenAI in Microsoft Foundry Models v1 REST API reference Feedback Summarize this article for me In this article Only a subset of operations are currently supported with the v1 API. To learn more, see the API version lifecycle guide . v1 OpenAPI 3.0 spec Create chat completion POST {endpoint}/openai/v1/chat/completions Creates a chat completion. Parameters Name In Required Type Description endpoint path Yes string url Supported Azure OpenAI endpoints (protocol and hostname, for example: https://aoairesource.openai.azure.com . Replace "aoairesource" with your Azure OpenAI resource name). https://{your-resource-name}.openai.azure.com api-version query No The explicit Microsoft Foundry Models API version to use for this request. v1 if not otherwise specified. Request Header Use either token based authentication or API key. Authenticating with token based authentication is recommended and more secure. Name Required Type Description Authorization True string Example: Authorization: Bearer {Azure_OpenAI_Auth_Token} To generate an auth token using Azure CLI: az account get-access-token --resource https://cognitiveservices.azure.com Type: oauth2 Authorization Url: https://login.microsoftonline.com/common/oauth2/v2.0/authorize scope: https://cognitiveservices.azure.com/.default api-key True string Provide Azure OpenAI API key here Request Body Content-Type : application/json Name Type Description Required Default audio object Parameters for audio output. Required when audio output is requested with modalities: ["audio"] . No └─ format enum Specifies the output audio format. Must be one of wav , mp3 , flac , opus , or pcm16 . Possibl --- ## Responses ### Responses URL: https://learn.microsoft.com/en-us/azure/ai-foundry/openai/latest#create-response?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Azure OpenAI in Microsoft Foundry Models v1 REST API reference Feedback Summarize this article for me In this article Only a subset of operations are currently supported with the v1 API. To learn more, see the API version lifecycle guide . v1 OpenAPI 3.0 spec Create chat completion POST {endpoint}/openai/v1/chat/completions Creates a chat completion. Parameters Name In Required Type Description endpoint path Yes string url Supported Azure OpenAI endpoints (protocol and hostname, for example: https://aoairesource.openai.azure.com . Replace "aoairesource" with your Azure OpenAI resource name). https://{your-resource-name}.openai.azure.com api-version query No The explicit Microsoft Foundry Models API version to use for this request. v1 if not otherwise specified. Request Header Use either token based authentication or API key. Authenticating with token based authentication is recommended and more secure. Name Required Type Description Authorization True string Example: Authorization: Bearer {Azure_OpenAI_Auth_Token} To generate an auth token using Azure CLI: az account get-access-token --resource https://cognitiveservices.azure.com Type: oauth2 Authorization Url: https://login.microsoftonline.com/common/oauth2/v2.0/authorize scope: https://cognitiveservices.azure.com/.default api-key True string Provide Azure OpenAI API key here Request Body Content-Type : application/json Name Type Description Required Default audio object Parameters for audio output. Required when audio output is requested with modalities: ["audio"] . No └─ format enum Specifies the output audio format. Must be one of wav , mp3 , flac , opus , or pcm16 . Possibl --- ## Vector stores ### Vector stores URL: https://learn.microsoft.com/en-us/azure/ai-foundry/openai/latest#list-vector-stores?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Azure OpenAI in Microsoft Foundry Models v1 REST API reference Feedback Summarize this article for me In this article Only a subset of operations are currently supported with the v1 API. To learn more, see the API version lifecycle guide . v1 OpenAPI 3.0 spec Create chat completion POST {endpoint}/openai/v1/chat/completions Creates a chat completion. Parameters Name In Required Type Description endpoint path Yes string url Supported Azure OpenAI endpoints (protocol and hostname, for example: https://aoairesource.openai.azure.com . Replace "aoairesource" with your Azure OpenAI resource name). https://{your-resource-name}.openai.azure.com api-version query No The explicit Microsoft Foundry Models API version to use for this request. v1 if not otherwise specified. Request Header Use either token based authentication or API key. Authenticating with token based authentication is recommended and more secure. Name Required Type Description Authorization True string Example: Authorization: Bearer {Azure_OpenAI_Auth_Token} To generate an auth token using Azure CLI: az account get-access-token --resource https://cognitiveservices.azure.com Type: oauth2 Authorization Url: https://login.microsoftonline.com/common/oauth2/v2.0/authorize scope: https://cognitiveservices.azure.com/.default api-key True string Provide Azure OpenAI API key here Request Body Content-Type : application/json Name Type Description Required Default audio object Parameters for audio output. Required when audio output is requested with modalities: ["audio"] . No └─ format enum Specifies the output audio format. Must be one of wav , mp3 , flac , opus , or pcm16 . Possibl --- ## REST API (resource creation & deployment) ### REST API (resource creation & deployment) URL: https://learn.microsoft.com/rest/api/aiservices/accountmanagement/deployments/create-or-update?tabs=HTTP?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Summarize this article for me Deployments - Create Or Update Service: Azure AI Services API Version: 2024-10-01 Update the state of specified deployments associated with the Cognitive Services account. PUT https://management.azure.com/subscriptions/{subscriptionId}/resourceGroups/{resourceGroupName}/providers/Microsoft.CognitiveServices/accounts/{accountName}/deployments/{deploymentName}?api-version=2024-10-01 URI Parameters Name In Required Type Description account Name path True string minLength: 2 maxLength: 64 pattern: ^[a-zA-Z0-9][a-zA-Z0-9_.-]*$ The name of Cognitive Services account. deployment Name path True string The name of the deployment associated with the Cognitive Services Account resource Group Name path True string minLength: 1 maxLength: 90 The name of the resource group. The name is case insensitive. subscription Id path True string minLength: 1 The ID of the target subscription. api-version query True string minLength: 1 The API version to use for this operation. Request Body Name Type Description properties Deployment Properties Properties of Cognitive Services account deployment. sku Sku The resource model definition representing SKU tags object Resource tags. Responses Name Type Description 200 OK Deployment Create/Update the deployment successfully. 201 Created Deployment Create the deployment successfully. Other Status Codes Error Response Error response describing why the operation failed. Examples Put Deployment Sample request HTTP Java Python Go JavaScript dotnet PUT https://management.azure.com/subscriptions/subscriptionId/resourceGroups/resourceGroupName/providers/Microsoft.CognitiveServices/account --- ## Azure OpenAI monitoring data reference ### Azure OpenAI monitoring data reference URL: https://learn.microsoft.com/en-us/azure/ai-foundry/openai/monitor-openai-reference?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Azure OpenAI monitoring data reference Feedback Summarize this article for me In this article This article contains all the monitoring reference information for this service. See Monitor Azure OpenAI for details on the data you can collect for Azure OpenAI in Microsoft Foundry Models and how to use it. Metrics This section lists all the automatically collected platform metrics for this service. These metrics are also part of the global list of all platform metrics supported in Azure Monitor . For information on metric retention, see Azure Monitor Metrics overview . Supported metrics for Microsoft.CognitiveServices/accounts Here are the most important metrics we think you should monitor for Azure OpenAI. Later in this article is a longer list of all available metrics for this namespace which contains more details on metrics in this shorter list. Please see below list for most up to date information. We're working on refreshing the tables in the following sections. Azure OpenAI Requests Active Tokens Generated Completion Tokens Processed FineTuned Training Hours Processed Inference Tokens Processed Prompt Tokens Provisioned-managed Utilization V2 Prompt Token Cache Match Rate Time to Response Time Between Tokens Time to Last Byte Normalized Time to First Byte Tokens per Second You can also monitor Content Safety metrics that are used by other related services. Blocked Volume Harmful Volume Detected Potential Abusive User Count Safety System Event Total Volume Sent for Safety Check Note The Provisioned-managed Utilization metric is now deprecated and is no longer recommended. This metric has been replaced by the Provisioned-managed --- ## Audio events reference ### Audio events reference URL: https://learn.microsoft.com/en-us/azure/ai-foundry/openai/realtime-audio-reference?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Audio events reference Feedback Summarize this article for me In this article Note This document refers to the Microsoft Foundry (classic) portal. 🔄 Switch to the Microsoft Foundry (new) documentation if you're using the new portal. Note This document refers to the Microsoft Foundry (new) portal. Realtime events are used to communicate between the client and server in real-time audio applications. The events are sent as JSON objects over various endpoints, such as WebSockets or WebRTC. The events are used to manage the conversation, audio buffers, and responses in real-time. You can use audio client and server events with these APIs: Azure OpenAI Realtime API Azure AI Voice Live API Unless otherwise specified, the events described in this document are applicable to both APIs. Client events There are nine client events that can be sent from the client to the server: Event Description RealtimeClientEventConversationItemCreate The client conversation.item.create event is used to add a new item to the conversation's context, including messages, function calls, and function call responses. RealtimeClientEventConversationItemDelete The client conversation.item.delete event is used to remove an item from the conversation history. RealtimeClientEventConversationItemRetrieve The client conversation.item.retrieve event is used to retrieve an item from the conversation history. RealtimeClientEventConversationItemTruncate The client conversation.item.truncate event is used to truncate a previous assistant message's audio. RealtimeClientEventInputAudioBufferAppend The client input_audio_buffer.append event is used to append audio bytes to th --- ## Foundry Tools SDKs ### Foundry Tools SDKs URL: https://learn.microsoft.com/en-us/azure/ai-services/reference/sdk-package-resources?context=/azure/ai-foundry/context/context?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Foundry Tools SDK reference Feedback Summarize this article for me In this article This article provides an overview of available Foundry Tools client libraries and packages with links to service and feature level reference documentation. Available Foundry Tools Select a service from the table and learn how Azure AI can help you meet your development goals. Supported services Service Description Reference documentation Speech Add speech to text, text to speech, translation, and speaker recognition capabilities to applications. Speech SDK for C++ Supported services Service Description Reference documentation Azure AI Search Bring AI-powered cloud search to your mobile and web apps. • Azure AI Search SDK for .NET • Azure AI Search NuGet package Azure OpenAI Perform a wide variety of natural language tasks. • Azure OpenAI SDK for .NET • Azure OpenAI NuGet package Bot Service Create bots and connect them across channels. • Bot service SDK for .NET • Bot Builder (NuGet package) Content Safety Detect harmful content in applications and services. • Content Safety SDK for .NET • Content Safety NuGet package Custom Vision Customize image recognition for your applications and models. • Custom Vision SDK for .NET • Custom Vision NuGet package (prediction) • Custom Vision NuGet package (training) Document Intelligence Turn documents into intelligent data-driven solutions. • Document Intelligence SDK for .NET • Document Intelligence NuGet package Face Detect, recognize, and identify human faces in images. • Face SDK for .NET • Face NuGet package Immersive Reader Help users with text readability and comprehension. • Immersive Reader C# quicks --- ## Foundry Tools REST APIs ### Foundry Tools REST APIs URL: https://learn.microsoft.com/en-us/azure/ai-services/reference/rest-api-resources?context=/azure/ai-foundry/context/context?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Foundry Tools REST API reference Feedback Summarize this article for me In this article This article lists the REST API reference documentation for Foundry Tools. Available Foundry Tools Select a service from the table to learn how it can help you meet your development goals. Supported services Service documentation Description Reference documentation Azure AI Search Bring AI-powered cloud search to your mobile and web apps Azure AI Search API Azure OpenAI Perform a wide variety of natural language tasks Azure OpenAI APIs • resource creation & deployment • completions & embeddings • fine-tuning Content Safety A Foundry Tool that detects unwanted contents Content Safety API Custom Vision Customize image recognition for your business applications. Custom Vision APIs • prediction • training Document Intelligence Turn documents into intelligent data-driven solutions Document Intelligence API Face Detect and identify people and emotions in images Face API Language Build apps with industry-leading natural language understanding capabilities REST API Speech Speech to text, text to speech, translation, and speaker recognition Speech APIs • speech to text • text to speech Translator Translate more than 100 in-use, at-risk, and endangered languages and dialects Translator APIs • text translation • document translation Vision Analyze content in images and videos Vision API Deprecated services Service documentation Description Reference documentation Anomaly Detector (deprecated 2023) Identify potential problems early on Anomaly Detector API Content Moderator (deprecated 2024) Detect potentially offensive or unwanted content Content Moderat --- ## Azure Resource Manager/Bicep/Terraform ### Azure Resource Manager/Bicep/Terraform URL: https://learn.microsoft.com/azure/templates/microsoft.cognitiveservices/accounts?pivots=deployment-language-bicep?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Microsoft.CognitiveServices accounts Feedback Summarize this article for me In this article Latest 2025-10-01-preview 2025-09-01 2025-07-01-preview 2025-06-01 2025-04-01-preview 2024-10-01 2024-06-01-preview 2024-04-01-preview 2023-10-01-preview 2023-05-01 2022-12-01 2022-10-01 2022-03-01 2021-10-01 2021-04-30 2017-04-18 2016-02-01-preview Bicep resource definition The accounts resource type can be deployed with operations that target: Resource groups - See resource group deployment commands For a list of changed properties in each API version, see change log . Resource format To create a Microsoft.CognitiveServices/accounts resource, add the following Bicep to your template. resource symbolicname 'Microsoft.CognitiveServices/accounts@2025-10-01-preview' = { identity: { type: 'string' userAssignedIdentities: { {customized property}: {} } } kind: 'string' location: 'string' name: 'string' properties: { allowedFqdnList: [ 'string' ] allowProjectManagement: bool amlWorkspace: { identityClientId: 'string' resourceId: 'string' } apiProperties: { aadClientId: 'string' aadTenantId: 'string' eventHubConnectionString: 'string' qnaAzureSearchEndpointId: 'string' qnaAzureSearchEndpointKey: 'string' qnaRuntimeEndpoint: 'string' statisticsEnabled: bool storageAccountConnectionString: 'string' superUser: 'string' websiteName: 'string' } associatedProjects: [ 'string' ] customSubDomainName: 'string' defaultProject: 'string' disableLocalAuth: bool dynamicThrottlingEnabled: bool encryption: { keySource: 'string' keyVaultProperties: { identityClientId: 'string' keyName: 'string' keyVaultUri: 'string' keyVersion: 'string' } } locations: { regions: --- ## Azure CLI ### Azure CLI URL: https://learn.microsoft.com/cli/azure/cognitiveservices?view=azure-cli-latest&preserve-view=true Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Summarize this article for me az cognitiveservices Manage Azure Cognitive Services accounts. This article lists the Azure CLI commands for Azure Cognitive Services account and subscription management only. Refer to the documentation at https://learn.microsoft.com/azure/cognitive-services/ for individual services to learn how to use the APIs and supported SDKs. Commands Name Description Type Status az cognitiveservices account Manage Azure Cognitive Services accounts. Core GA az cognitiveservices account commitment-plan Manage commitment plans for Azure Cognitive Services accounts. Core GA az cognitiveservices account commitment-plan create Create a commitment plan for Azure Cognitive Services account. Core GA az cognitiveservices account commitment-plan delete Delete a commitment plan from Azure Cognitive Services account. Core GA az cognitiveservices account commitment-plan list Show all commitment plans from Azure Cognitive Services account. Core GA az cognitiveservices account commitment-plan show Show a commitment plan from Azure Cognitive Services account. Core GA az cognitiveservices account connection Manage Azure Cognitive Services connection and its more specific derivatives. Core GA az cognitiveservices account connection create Create a connection. Core GA az cognitiveservices account connection delete Delete a connection. Core GA az cognitiveservices account connection list List all connections. Core GA az cognitiveservices account connection show Show details of a connection. Core GA az cognitiveservices account connection update Update a connection. Core GA az cognitiveservices account create Manage Azure Cognitive Serv --- ## How to configure Guardrails and controls ### How to configure Guardrails and controls URL: https://learn.microsoft.com/en-us/azure/ai-foundry/guardrails/how-to-create-guardrails?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . How to configure guardrails and controls in Microsoft Foundry Feedback Summarize this article for me In this article This comprehensive guide walks you through every aspect of creating, configuring, and managing guardrails and controls in Microsoft Foundry. From basic setup to advanced features, this article covers both UI instructions and API configuration methods. Prerequisites An Azure account. If you don't have one, you can create one for free . An Azure AI resource . Create a guardrail in Foundry Go to Foundry and navigate to your project. Select Build in the top right menu. Select the Guardrails page from the left navigation. Select Create Guardrail in the top right. Add controls to a guardrail Default controls are displayed in the right pane when you create a new guardrail. Select a risk from the dropdown menu. Choose intervention points and actions : Recommended intervention points and actions for that risk is shown. Select one or many intervention points and one action to configure your control. Note Some intervention points will not be available for a risk if that is inapplicable at that intervention point. For example, by definition, user input attacks are malicious content added to the user input. So, that risk can be scanned only at that intervention point. Select Add control . The control is added to the table on the right. Delete controls from a guardrail To delete a control: Select the control you want to remove. Select Delete . Note Some controls can only be deleted by Managed Customers who are approved for modified content filtering. Learn more about modified content filtering . Edit controls in a guardrail The --- ## Harm categories ### Harm categories URL: https://learn.microsoft.com/en-us/azure/ai-foundry/guardrails/severity-levels?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Harm categories and severity levels in Microsoft Foundry Feedback Summarize this article for me In this article The content safety system integrated in Microsoft Foundry contains neural multiclass classification models aimed at detecting and filtering harmful content. The models cover four categories (hate, sexual, violence, and self-harm) across four severity levels (safe, low, medium, and high). Content detected at the 'safe' severity level is labeled in annotations but isn't subject to filtering and isn't configurable. The system also includes other optional classification models aimed at detecting jailbreak risk and known content for text and code. These models are binary classifiers that flag whether user or model behavior qualifies as a jailbreak attack or match to known text or source code. The use of these models is optional, but use of protected material code model may be required for Customer Copyright Commitment coverage. Note The text content safety models for the hate, sexual, violence, and self-harm categories are specifically trained and tested on the following languages: English, German, Japanese, Spanish, French, Italian, Portuguese, and Chinese. However, the service can work in many other languages, but the quality might vary. In all cases, you should do your own testing to ensure that it works for your application. Harm category descriptions The following table summarizes the harm categories supported by Foundry guardrails: Category Description Hate and Fairness Hate and fairness-related harms refer to any content that attacks or uses discriminatory language with reference to a person or identity group based o --- ## Prompt shields ### Prompt shields URL: https://learn.microsoft.com/en-us/azure/ai-foundry/openai/concepts/content-filter-prompt-shields?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Prompt shields in Foundry Feedback Summarize this article for me In this article Prompt shields are a feature of the Microsoft Foundry Guardrails and controls system that helps detect and mitigate user prompt attacks. These attacks occur when a user attempts to manipulate the model's behavior by embedding harmful or inappropriate content within their input. Prompt shields analyzes LLM inputs and detects adversarial user input attacks. Types of input attacks The types of input attacks that Prompt Shields detects are described in this table. Type Attacker Entry point Method Objective/impact Resulting behavior User Prompt attacks User User prompts Ignoring system prompts/RLHF training Altering intended LLM behavior Performing restricted actions against training Document attacks Third party Third-party content (documents, emails) Misinterpreting third-party content Gaining unauthorized access or control Executing unintended commands or actions Prompt Shields for user prompts Previously called Jailbreak risk detection , this shield targets User Prompt injection attacks, where users deliberately exploit system vulnerabilities to elicit unauthorized behavior from the LLM. This could lead to inappropriate content generation or violations of system-imposed restrictions. Examples Classification Description Example No prompt attack Requests from users aligned with the system’s intended use, without attempting to circumvent system rules. User : What are the top conclusions from yesterday’s meeting? Prompt attack User attempts to circumvent system rules by: Changing system rules Deceiving the model by including false conversational content R --- ## Sensitive data leakage ### Sensitive data leakage URL: https://learn.microsoft.com/en-us/azure/ai-foundry/openai/concepts/content-filter-personal-information?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Personally identifiable information (PII) filter Feedback Summarize this article for me In this article Personally identifiable information (PII) refers to any information that can be used to identify a particular individual, such as a name, address, phone number, email address, social security number, driver's license number, passport number, or similar information. PII detection is used to help prevent PII from being exposed or shared, protecting users from identity theft, financial fraud, or other types of privacy violations. In the context of large language models (LLMs), PII detection involves analyzing text content in LLM completions. When PII has been identified, it can be flagged for further review, or the output can be blocked. The PII filter scans the output of LLMs to identify and flag known personal information. It's designed to help organizations prevent the generation of content that closely matches sensitive personal information. PII types There are many different types of PII, and you can specify which types you want to filter. The set of PII types that can be detected by the filter matches the set that's defined in the Azure Language in Foundry Tools docs . Filtering modes The PII filter can be configured to operate in two modes. Annotate mode flags PII that's returned in the model output. Annotate and Block mode blocks the entire output if PII is detected. The filtering mode can be set for each PII category individually. Feedback Was this page helpful? Yes No No Need help with this topic? Want to try using Ask Learn to clarify or guide you through this topic? Ask Learn Ask Learn Suggest a fix? Additional resour --- ## Groundedness detection ### Groundedness detection URL: https://learn.microsoft.com/en-us/azure/ai-foundry/openai/concepts/content-filter-groundedness?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Groundedness detection filter Feedback Summarize this article for me In this article Groundedness detection in Azure AI Content Safety helps you ensure that large language model (LLM) responses are based on your provided source material, reducing the risk of non-factual or fabricated outputs. Ungroundedness refers to instances where the LLMs produce information that is non-factual or inaccurate from what was present in the source materials. Groundedness detection requires document embedding and formatting . Key terms Retrieval Augmented Generation (RAG) : RAG is a technique for augmenting LLM knowledge with other data. LLMs can reason about wide-ranging topics, but their knowledge is limited to the public data that was available at the time they were trained. If you want to build AI applications that can reason about private data or data introduced after a model’s cutoff date, you need to provide the model with that specific information. The process of bringing the appropriate information and inserting it into the model prompt is known as Retrieval Augmented Generation (RAG). For more information, see Retrieval-augmented generation (RAG) . Groundedness and Ungroundedness in LLMs : This refers to the extent to which the model's outputs are based on provided information or reflect reliable sources accurately. A grounded response adheres closely to the given information, avoiding speculation or fabrication. In groundedness measurements, source information is crucial and serves as the grounding source. User scenarios Groundedness detection supports text-based Summarization and QnA tasks to ensure that the generated summaries or answ --- ## Protected material for code ### Protected material for code URL: https://learn.microsoft.com/en-us/azure/ai-services/content-safety/quickstart-protected-material-code?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Quickstart: Protected material detection for code (preview) Feedback Summarize this article for me In this article The Protected Material for Code feature provides a comprehensive solution for identifying AI outputs that match code from existing GitHub repositories. This feature allows you to use code generation models confidently, in a way that enhances transparency to end users and promotes compliance with organizational policies. Caution The Content Safety service's code scanner/indexer is only current through April 6, 2023. Code that was added to GitHub after this date won't be detected. Use your own discretion when using Protected Material for Code to detect recent bodies of code. The key objectives of the Protected Material Detection for Code feature for AI-generated code are: To detect and prevent the display of protected code generated by AI models. To enable organizations to manage risks associated with AI-generated code. To ensure that AI-generated code complies with legal, ethical, and organizational policy standards. For more information about protected material detection, see the Protected material detection concept page . For API input limits, see the Input requirements section of the Overview. Prerequisites An Azure account. If you don't have one, you can create one for free . An Azure AI resource . Setup Follow these steps to use the Content Safety try it out page: Go to Azure AI Foundry and navigate to your project/hub. Then select the Guardrails + controls tab on the left nav and select the Try it out tab. On the Try it out page, you can experiment with various Guardrails & controls features such as text and im --- ## Protected material for text ### Protected material for text URL: https://learn.microsoft.com/en-us/azure/ai-foundry/openai/concepts/content-filter-protected-material?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Protected material detection filter Feedback Summarize this article for me In this article The protected material detection filter scans the output of large language models (LLMs) to identify and flag known protected material. It helps organizations prevent the generation of content that closely matches copyrighted text or code. The protected material text filter flags known text content that large language models might output, such as song lyrics, articles, recipes, and selected web content. The protected material code filter flags protected code content that large language models might output. This is content found in known GitHub repositories and includes software libraries, source code, algorithms, and other proprietary programming content. Important The Guardrails and controls models for protected material detection, groundedness detection, and custom categories (standard) work with English only. Other content filtering models are specifically trained and tested on the following languages: Chinese, English, French, German, Spanish, Italian, Japanese, Portuguese. However, these features can work in many other languages, but the quality might vary. In all cases, you should do your own testing to ensure that it works for your application. Important The content filtering models for protected material detection, groundedness detection, and custom categories (standard) work with English only. Other content filtering models are specifically trained and tested on the following languages: Chinese, English, French, German, Spanish, Italian, Japanese, Portuguese. However, these features can work in many other languages, but the qualit --- ## Block lists ### Block lists URL: https://learn.microsoft.com/en-us/azure/ai-services/content-safety/quickstart-blocklist?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . QuickStart: Use a text blocklist Feedback Summarize this article for me In this article Get started using Azure AI Content Safety to create a custom text blocklist and use it to detect harmful content in text. Caution The sample data and code may contain offensive content. User discretion is advised. Prerequisites An Azure account. If you don't have one, you can create one for free . An Azure AI resource . Setup Follow these steps to use the Content Safety try it out page: Go to Azure AI Foundry and navigate to your project/hub. Then select the Guardrails + controls tab on the left nav and select the Try it out tab. On the Try it out page, you can experiment with various Guardrails & controls features such as text and image content, using adjustable thresholds to filter for inappropriate or harmful content. Use a blocklist The Use blocklist tab lets you create, edit, and add a blocklist to the moderation workflow. If you have a blocklist enabled when you run the test, you get a Blocklist detection panel under Results . It reports any matches with the blocklist. Prerequisites An Azure subscription - Create one for free Once you have your Azure subscription, create a Content Safety resource in the Azure portal to get your key and endpoint. Enter a unique name for your resource, select your subscription, and select a resource group, supported region (see Region availability ), and supported pricing tier. Then select Create . The resource takes a few minutes to deploy. After it finishes, Select go to resource . In the left pane, under Resource Management , select Subscription Key and Endpoint . The endpoint and either of the keys ar --- ## Custom categories ### Custom categories URL: https://learn.microsoft.com/en-us/azure/ai-services/content-safety/concepts/custom-categories?context=/azure/ai-foundry/context/context&view=foundry&preserve-view=true Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Custom categories (preview) Feedback Summarize this article for me In this article Azure AI Content Safety lets you create and manage your own content categories for enhanced moderation and filtering that matches your specific policies or use cases. Types of customization You can define and use custom categories through multiple methods. This section details and compares these methods. API Functionality Custom categories (standard) API Use a customizable machine learning model to create, get, query, and delete a customized category. Or, list all your customized categories for further annotation tasks. Custom categories (rapid) API Use a large language model (LLM) to quickly learn specific content patterns in emerging content incidents. Custom categories (standard) API The Custom categories (standard) API enables you to define categories specific to your needs, provide sample data, train a custom machine learning model, and use it to classify new content according to the learned categories. This API provides the standard workflow for customization with machine learning models. Depending on the training data quality, it can reach very good performance levels, but it can take up to several hours to train the model. This implementation works on text content, not image content. Custom categories (rapid) API The Custom categories (rapid) API is quicker and more flexible than the standard method. Use it to identify, analyze, contain, eradicate, and recover from cyber incidents that involve inappropriate or harmful content on online platforms. An incident might involve a set of emerging content patterns (text, image, or other modalities --- ## Intervention points ### Intervention points URL: https://learn.microsoft.com/en-us/azure/ai-foundry/guardrails/intervention-points?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Intervention points Feedback Summarize this article for me In this article Agentic AI expands both capability and attack surface. As soon as an agent can call external tools, write to databases, or trigger downstream processes, malfunctions or malicious attacks can lead to steering it off course, leaking sensitive data, or executing harmful actions. Relying solely on guardrails applied to models can leave these vectors exposed. To close this gap Microsoft Foundry allows guardrails to be applied directly to agents and allows the individual controls within those guardrails to be applied to four different intervention points: Intervention Point Description Example Control at this Intervention Point User input A query sent from a user to a model or agent. Sometimes referred to as "prompt." Some controls at this intervention point require the inclusion of document embedding by the user to take effect. Risk: User input attacks Action: Annotate and block When this control is specified in an agent's or model's guardrail, the user's input is scanned by a classification model that detects jailbreak attacks. If an attack is detected, the user's input is blocked from being sent to the model, halting the model. Tool call (Preview) The next action the agent is proposing to take, as generated by its underlying model. The tool call consists of which tool is called and the arguments it's called with, including data being sent to the tool. Risk: Hate (High) Action: Annotate and block When this control is specified, every time the agent is about to execute a tool call, the proposed content being sent to the tool is scanned for hateful content. If --- ## Default safety policies ### Default safety policies URL: https://learn.microsoft.com/en-us/azure/ai-foundry/openai/concepts/default-safety-policies?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Default Guidelines & controls policies Feedback Summarize this article for me In this article Note This document refers to the Microsoft Foundry (classic) portal. 🔄 Switch to the Microsoft Foundry (new) documentation if you're using the new portal. Note This document refers to the Microsoft Foundry (new) portal. Azure OpenAI in Microsoft Foundry Models includes default safety policies applied to all models (excluding Azure OpenAI Whisper). These configurations provide you with a responsible experience by default, including content filtering models , blocklists, prompt transformation, content credentials , and other features. Default safety aims to mitigate risks in different categories such as hate and fairness, sexual, violence, self-harm, protected material content, and user prompt injection attacks. To learn more about guardrail and controls, visit our documentation describing categories and severity levels . All safety policies are configurable. To learn more about configurability, see the documentation on configuring guardrails . Azure OpenAI in Foundry Models includes default safety policies applied to all models (excluding Azure OpenAI Whisper). These configurations provide you with a responsible experience by default, including content filtering models , blocklists, prompt transformation, content credentials , and other features. Default safety aims to mitigate risks in different categories such as hate and fairness, sexual, violence, self-harm, protected material content, and user prompt injection attacks. To learn more about content filtering, visit our documentation describing categories and severity levels . All safe --- ## Content streaming ### Content streaming URL: https://learn.microsoft.com/en-us/azure/ai-foundry/openai/concepts/content-streaming?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Content streaming Feedback Summarize this article for me In this article This guide describes the Azure OpenAI content streaming experience and options. Customers can receive content from the API when it's generated, instead of waiting for chunks of content that have been verified to pass the content filters. Default filtering behavior The content guardrails system is integrated and enabled by default for all customers. In the default streaming scenario, completion content buffers, the content guardrail system runs on the buffered content, and – depending on the guardrail configuration – content is returned to the user if it doesn't violate the guardrail policy (Microsoft's default or a custom user configuration), or it is immediately blocked and a guardrail error is returned instead. This process repeats until the end of the stream. Content is fully vetted according to the guardrail policy before it's returned to the user. Content isn't returned token-by-token in this case, but in "content chunks" of the respective buffer size. The content filtering system is integrated and enabled by default for all customers. In the default streaming scenario, completion content buffers, the content filtering system runs on the buffered content, and – depending on the content filtering configuration – content is returned to the user if it doesn't violate the content filtering policy (Microsoft's default or a custom user configuration), or it is immediately blocked and a content filtering error is returned instead. This process repeats until the end of the stream. Content is fully vetted according to the content filtering policy before it's re --- ## Abuse monitoring ### Abuse monitoring URL: https://learn.microsoft.com/en-us/azure/ai-foundry/openai/concepts/abuse-monitoring?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Abuse Monitoring Feedback Summarize this article for me In this article Azure Direct Models detect and mitigate instances of recurring content and/or behaviors that suggest use of the service in a manner that might violate the Code of Conduct . Details on how data is handled can be found on the Data, Privacy, and Security page. Components of abuse monitoring There are several components to abuse monitoring: Content Classification : Classifier models detect harmful text and/or images in user prompts (inputs) and completions (outputs). The system looks for categories of harms as defined in the Content Requirements , and assigns severity levels as described in more detail on the Content Filtering page. The content classification signals contribute to pattern detection as described below. Abuse Pattern Capture : The abuse monitoring system for Azure Direct Models looks at customer usage patterns and employs algorithms and heuristics to detect and score indicators of potential abuse. Detected patterns consider, for example, the frequency and severity at which harmful content is detected (as indicated in content classifier signals) in a customer’s prompts and completions, as well as the intentionality of the behavior. The trends and urgency of the detected pattern will also affect scoring of potential abuse severity. For example, a higher volume of harmful content classified as higher severity, or recurring conduct indicating intentionality (such as recurring jailbreak attempts) are both more likely to receive a high score indicating potential abuse. Review and Decision : Prompts and completions that are flagged through content classi --- ## Responsible AI overview ### Responsible AI overview URL: https://learn.microsoft.com/en-us/azure/ai-foundry/responsible-use-of-ai-overview?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Responsible AI for Microsoft Foundry Feedback Summarize this article for me In this article Note This document refers to the Microsoft Foundry (classic) portal. 🔄 Switch to the Microsoft Foundry (new) documentation if you're using the new portal. Note This document refers to the Microsoft Foundry (new) portal. This article provides an overview of the resources for building and deploying trustworthy AI agents. This includes end-to end security, observability, and governance with controls and checkpoints at all stages of the agent lifecycle. Our recommended essential development steps are grounded in the Microsoft Responsible AI Standard , which sets policy requirements that our own engineering teams follow. Much of the content of the Standard follows a pattern, asking teams to Discover, Protect, and Govern potential content risks. At Microsoft, our approach is guided by a governance framework rooted in AI principles, which establish product requirements and serve as our "north star." When we identify a business use case for generative AI, we first discover and assess the potential risks of the AI system to pinpoint critical focus areas. Once we identify these risks, we evaluate their prevalence within the AI system through systematic measurement, helping us prioritize areas that need attention. We then apply appropriate protection at the model and agent level against those risks. Finally, we examine strategies for managing risks in production, including deployment and operational readiness and setting up monitoring to support ongoing governance to ensure compliance and surface new risks after the application is live. In alignment --- ## Limited access ### Limited access URL: https://learn.microsoft.com/en-us/azure/ai-services/cognitive-services-limited-access?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Limited Access features for Foundry Tools Feedback Summarize this article for me In this article Our vision is to empower developers and organizations to use AI to transform society in positive ways. We encourage responsible AI practices to protect the rights and safety of individuals. To help achieve this goal, Microsoft has implemented a Limited Access policy grounded in our AI Principles to support responsible deployment of Azure services. What is Limited Access? Limited Access services require registration, and only customers managed by Microsoft—meaning anyone who works directly with Microsoft account teams—is eligible for access. The use of these services is limited to the use case selected at the time of registration. Customers must acknowledge that they've reviewed and agree to the terms of service. Microsoft might require customers to reverify this information. Limited Access services are made available to customers under the terms governing their subscription to Microsoft Azure Services (including the terms and conditions ). Review these terms carefully as they contain important conditions and obligations governing your use of Limited Access services. List of Limited Access services The following services are Limited Access: Custom Neural Voice : Pro features Custom text to speech avatar : All features Speaker Recognition : All features Face : Identify and Verify features, face ID property Vision : Celebrity Recognition feature Azure OpenAI : Azure OpenAI in Foundry Models, modified abuse monitoring, and modified content filters Features of these services that aren't listed are available without registration. FAQ about --- ### Limited access URL: https://learn.microsoft.com/en-us/azure/ai-foundry/responsible-ai/openai/limited-access?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Limited access for Azure Direct Models Feedback Summarize this article for me In this article Important Non-English translations are provided for convenience only. Please consult the EN-US version of this document for the definitive version. As part of Microsoft's commitment to responsible AI, we have designed and operate Azure Direct Models (as defined in the Product Terms ) with the intention of respecting the rights of individuals and society and fostering transparent human-computer interaction. For this reason, certain Azure Direct Models (or versions of them) are designated as Limited Access Services, and access and use are subject to eligibility criteria determined by Microsoft. Unless otherwise indicated in the service, all Azure customers are eligible for access to Azure Direct Models, and all uses consistent with the Product Terms and Code of Conduct are permitted, so customers are not required to submit a registration form unless they are: (a) accessing an Azure Direct Model designated as a Limited Access Service, or (b) requesting approval to modify Guardrails (previously content filters) and/or abuse monitoring for an Azure Direct Model. Azure Direct Models are made available to customers under the terms governing their subscription to Microsoft Azure Services, including Product Terms such as the Universal License Terms applicable to Microsoft Generative AI Services and the product offering terms for the Azure Direct Model. Please review these terms carefully as they contain important conditions and obligations governing your use. Azure OpenAI Service is made available to customers under the terms governing their sub --- ## Transparency note ### Transparency note URL: https://learn.microsoft.com/en-us/azure/ai-foundry/responsible-ai/openai/transparency-note?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Transparency note for Azure OpenAI Feedback Summarize this article for me In this article Important Non-English translations are provided for convenience only. Please consult the EN-US version of this document for the definitive version. What is a transparency note? An AI system includes not only the technology, but also the people who use it, the people who are affected by it, and the environment in which it's deployed. Creating a system that is fit for its intended purpose requires an understanding of how the technology works, what its capabilities and limitations are, and how to achieve the best performance. Microsoft's Transparency Notes are intended to help you understand how our AI technology works, the choices system owners can make that influence system performance and behavior, and the importance of thinking about the whole system, including the technology, the people, and the environment. You can use Transparency Notes when developing or deploying your own system, or share them with the people who will use or be affected by your system. Microsoft's Transparency Notes are part of a broader effort at Microsoft to put our AI Principles into practice. To find out more, see the Microsoft's AI principles . The basics of the Azure OpenAI Models Azure OpenAI provides customers with a fully managed Foundry Tool that lets developers and data scientists apply OpenAI's powerful models including models that can generate natural language, code, and images. Within the Azure OpenAI Service, the OpenAI models are integrated with Microsoft-developed Guardrails (previously content filters) and abuse detection models. Learn more about Gua --- ### Transparency note URL: https://learn.microsoft.com/en-us/azure/ai-foundry/responsible-ai/agents/transparency-note?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Transparency Note for Azure Agent Service Feedback Summarize this article for me In this article Important Non-English translations are provided for convenience only. Please consult the EN-US version of this document for the definitive version. What is a Transparency Note? An AI system includes not only the technology, but also the people who will use it, the people who will be affected by it, and the environment in which it is deployed. Creating a system that is fit for its intended purpose requires an understanding of how the technology works, what its capabilities and limitations are, and how to achieve the best performance. Microsoft’s Transparency Notes are intended to help you understand how our AI technology works, the choices system owners can make that influence system performance and behavior, and the importance of thinking about the whole system, including the technology, the people, and the environment. You can use Transparency Notes when developing or deploying your own system or share them with the people who will use or be affected by your system. Microsoft’s Transparency Notes are part of a broader effort at Microsoft to put our AI Principles into practice. To find out more, see the Microsoft AI principles . The basics of Azure AI Agent Service Introduction Azure AI Agent Service is a fully managed service designed to empower developers to securely build, deploy, and scale high-quality and extensible AI agents without needing to manage the underlying compute and storage resources. Azure AI Agent Service provides integrated access to models, tools, and technology and enables you to extend the functionality of agen --- ## Code of conduct ### Code of conduct URL: https://learn.microsoft.com/legal/ai-code-of-conduct?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Microsoft Enterprise AI Services Code of Conduct Feedback Summarize this article for me In this article This Microsoft Enterprise AI Services Code of Conduct (“Code of Conduct”) defines the requirements that all customers of Microsoft AI Services (as defined in the Product Terms) must adhere to in good faith. This Code of Conduct unifies and replaces the previous codes for Microsoft Generative AI Services, Azure AI Vision Face API, and Azure AI Speech text to speech. This Code of Conduct differs from earlier codes of conduct in structure and phrasing, and it is designed to better align with emerging AI regulations (e.g., EU AI Act). This Code of Conduct applies in addition to the Microsoft Product Terms , including the Acceptable Use Policy. Any capitalized terms not otherwise defined in this Code of Conduct will have the meaning given to them in the applicable Microsoft agreement(s) under which customers purchase the Online Service. Responsible AI requirements Customers must ensure that all of their applications built with Microsoft AI Services, including applications that make decisions, or take actions, autonomously or with varying levels of human intervention: Implement technical and operational measures to detect fraudulent user behavior in account creation and during use. Implement strong technical controls on inputs and outputs, including decisions made and actions taken by their applications, to reduce the likelihood of misuse beyond the application's intended purpose. Disclose when the output, decisions, or actions are generated by AI, including the synthetic nature of generated voices, images, and/or videos, such that --- ### Code of conduct URL: https://learn.microsoft.com/legal/ai-code-of-conduct?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Microsoft Enterprise AI Services Code of Conduct Feedback Summarize this article for me In this article This Microsoft Enterprise AI Services Code of Conduct (“Code of Conduct”) defines the requirements that all customers of Microsoft AI Services (as defined in the Product Terms) must adhere to in good faith. This Code of Conduct unifies and replaces the previous codes for Microsoft Generative AI Services, Azure AI Vision Face API, and Azure AI Speech text to speech. This Code of Conduct differs from earlier codes of conduct in structure and phrasing, and it is designed to better align with emerging AI regulations (e.g., EU AI Act). This Code of Conduct applies in addition to the Microsoft Product Terms , including the Acceptable Use Policy. Any capitalized terms not otherwise defined in this Code of Conduct will have the meaning given to them in the applicable Microsoft agreement(s) under which customers purchase the Online Service. Responsible AI requirements Customers must ensure that all of their applications built with Microsoft AI Services, including applications that make decisions, or take actions, autonomously or with varying levels of human intervention: Implement technical and operational measures to detect fraudulent user behavior in account creation and during use. Implement strong technical controls on inputs and outputs, including decisions made and actions taken by their applications, to reduce the likelihood of misuse beyond the application's intended purpose. Disclose when the output, decisions, or actions are generated by AI, including the synthetic nature of generated voices, images, and/or videos, such that --- ## Data, privacy, and security ### Data, privacy, and security URL: https://learn.microsoft.com/en-us/azure/ai-foundry/responsible-ai/openai/data-privacy?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Data, privacy, and security for Azure Direct Models in Microsoft Foundry Feedback Summarize this article for me In this article Important Non-English translations are provided for convenience only. Please consult the EN-US version of this document for the definitive version. This article provides details regarding how data provided by you to Azure Direct Models in Microsoft Foundry are processed, used, and stored. Azure Direct Model means an AI model designated and deployed as an “Azure Direct Model” in Foundry, and includes Azure OpenAI models. Azure Direct Models store and process data to provide the service and to monitor for uses that violate the applicable product terms. Please also see Microsoft Products and Services Data Protection Addendum , which governs data processing by Azure Direct Models. Foundry is an Azure service; learn more about applicable Azure compliance offerings. Important Your prompts (inputs) and completions (outputs), your embeddings, and your training data: are NOT available to other customers. are NOT available to OpenAI or other Azure Direct Model providers. are NOT used by Azure Direct Model providers to improve their models or services. are NOT used to train any generative AI foundation models without your permission or instruction. Customer Data, Prompts, and Completions are NOT used to improve Microsoft or third-party products or services without your explicit permission or instruction. Your fine-tuned Azure Direct Models are available exclusively for your use. Foundry is an Azure service; Microsoft hosts the Azure Direct Models in Microsoft's Azure environment and Azure Direct Models do NOT inte --- ## Customer Copyright Commitment ### Customer Copyright Commitment URL: https://learn.microsoft.com/en-us/azure/ai-foundry/responsible-ai/openai/customer-copyright-commitment?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Customer Copyright Commitment Required Mitigations Feedback Summarize this article for me In this article Important Non-English translations are provided for convenience only. Please consult the EN-US version of this document for the definitive version. Note The requirements described below apply only to customers using Azure OpenAI in Microsoft Foundry Models ("Azure OpenAI") and other Covered Products with configurable Metaprompts or other safety systems ("Configurable GAI Services"). They do not apply to customers using other Covered Products including Copilots with safety systems that are fixed. The only Configurable GAI Services are Microsoft Copilot Studio and GitHub Copilot; the Universal Required Mitigations do not apply to these offerings, but service-specific mitigations apply instead. The Customer Copyright Commitment ("CCC") is a provision in the Microsoft Product Terms that describes Microsoft's obligation to defend customers against certain third-party intellectual property claims relating to Output Content. For Azure OpenAI and any Configurable GAI Service, Customer also must have implemented all mitigations required by the Azure OpenAI documentation in the offering that delivered the Output Content that is the subject of the claim. The required mitigations to maintain CCC coverage are set forth below. This page describes only the required mitigations necessary to maintain CCC coverage for Azure OpenAI and Configurable GAI Services. It is not an exhaustive list of requirements or mitigations required to use Azure OpenAI (or Configurable GAI Services) responsibly in compliance with the documentation. Azure OpenAI c --- ## Data, privacy, and security for agents ### Data, privacy, and security for agents URL: https://learn.microsoft.com/en-us/azure/ai-foundry/responsible-ai/agents/data-privacy-security?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Data, privacy, and security for Azure AI Agent Service Feedback Summarize this article for me In this article Important Non-English translations are provided for convenience only. Please consult the EN-US version of this document for the definitive version. Azure AI Agent Service is a fully managed service designed to empower developers to securely build, deploy, and scale high-quality, and extensible AI agents without needing to manage the underlying compute and storage resources. Azure AI Agent Service integrates models, tools and technology and enables you to extend agents with knowledge from connected sources (such as Bing Search, SharePoint, Fabric, Azure Blob storage, and licensed data) and actions using tools such as Azure Logic Apps, Azure Functions, OpenAPI 3.0 specified tools and Code Interpreter Note This article provides details regarding how data provided by you to the Azure AI Agent service is processed, used, and stored. Please also see the Microsoft Products and Services Data Protection Addendum , which governs data processing by the Azure AI Agent Service (but may not necessarily apply to external tools or services with which Azure AI Agent Service interacts, which are subject to their own data processing terms). Important Your prompts (inputs) and completions (outputs) and your data: are NOT available to other customers. are NOT available to OpenAI, Meta, Cohere, or Mistral. are NOT used to improve OpenAI, Meta, Cohere, or Mistral models. When you use Azure AI Agent Service with tools that retrieve data from external sources or services (such as the Grounding with Bing Search tool), the terms (including data pr --- ## Get started with DeepSeek-R1 reasoning model ### Get started with DeepSeek-R1 reasoning model URL: https://learn.microsoft.com/en-us/azure/ai-foundry/foundry-models/tutorials/get-started-deepseek-r1?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Tutorial: Get started with DeepSeek-R1 reasoning model in Microsoft Foundry Models Feedback Summarize this article for me In this article Note This document refers to the Microsoft Foundry (classic) portal. 🔄 Switch to the Microsoft Foundry (new) documentation if you're using the new portal. Note This document refers to the Microsoft Foundry (new) portal. In this tutorial, you learn how to deploy and use a DeepSeek reasoning model in Microsoft Foundry. This tutorial uses DeepSeek-R1 for illustration. However, the content also applies to the newer DeepSeek-R1-0528 reasoning model. What you'll accomplish: In this tutorial, you'll deploy the DeepSeek-R1 reasoning model, send inference requests programmatically using code, and parse the reasoning output to understand how the model arrives at its answers. The steps you perform in this tutorial are: Create and configure the Azure resources to use DeepSeek-R1 in Foundry Models. Configure the model deployment. Use DeepSeek-R1 with the next generation v1 Azure OpenAI APIs to consume the model in code. Prerequisites To complete this article, you need: An Azure subscription with a valid payment method. If you don't have an Azure subscription, create a paid Azure account to begin. If you're using GitHub Models , you can upgrade from GitHub Models to Microsoft Foundry Models and create an Azure subscription in the process. Access to Microsoft Foundry with appropriate permissions to create and manage resources. Typically requires Contributor or Owner role on the resource group for creating resources and deploying models. Install the Azure OpenAI SDK for your programming language: Python : pip --- ## Prompt engineering techniques ### Prompt engineering techniques URL: https://learn.microsoft.com/en-us/azure/ai-foundry/openai/concepts/prompt-engineering?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Prompt engineering techniques Feedback Summarize this article for me In this article These techniques aren't recommended for reasoning models like gpt-5 and o-series models. Prompt construction can be difficult. In practice, the prompt acts assist the model complete the desired task, but it's more of an art than a science, often requiring experience and intuition to craft a successful prompt. The goal of this article is to help get you started with this learning process. It attempts to capture general concepts and patterns that apply to all GPT models. However it's important to understand that each model behaves differently, so the learnings might not apply equally to all models. Basics This section covers the basic concepts and elements of GPT prompts. Text prompts are how users interact with GPT models. As with all generative language models, GPT models attempt to produce the next series of words that are most likely to follow from the previous text. It's as if we're saying What is the first thing that comes to your mind when I say ? The examples below demonstrate this behavior. Given the first words of famous content, the model is able to accurately continue the text. Prompt Completion Four score and seven years ago our fathers brought forth on this continent, a new nation, conceived in Liberty, and dedicated to the proposition that all men are created equal. […] "Call me Ishmael." "Some years ago—never mind how long precisely—having little or no money in my purse, and nothing particular to interest me on shore, I thought I would sail about a little and see the watery part of the world." […] As you develop more compl --- ## Performance & latency ### Performance & latency URL: https://learn.microsoft.com/en-us/azure/ai-foundry/openai/how-to/latency?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Performance and latency Feedback Summarize this article for me In this article This article provides you with background around how latency and throughput works with Azure OpenAI and how to optimize your environment to improve performance. Understanding throughput vs latency There are two key concepts to think about when sizing an application: (1) System level throughput measured in tokens per minute (TPM) and (2) Per-call response times (also known as latency). System level throughput This looks at the overall capacity of your deployment – how many requests per minute and total tokens that can be processed. For a standard deployment, the quota assigned to your deployment partially determines the amount of throughput you can achieve. However, quota only determines the admission logic for calls to the deployment and isn't directly enforcing throughput. Due to per-call latency variations, you might not be able to achieve throughput as high as your quota. In a provisioned deployment, a set amount of model processing capacity is allocated to your endpoint. The amount of throughput that you can achieve on the endpoint is a factor of the workload shape including input token amount, output amount, call rate and cache match rate. The number of concurrent calls and total tokens processed can vary based on these values. For all deployment types, understanding system level throughput is a key component of optimizing performance. It is important to consider system level throughput for a given model, version, and workload combination as the throughput will vary across these factors. Estimating system level throughput Estimating TPM with Azur --- ## Red teaming large language models (LLMs) ### Red teaming large language models (LLMs) URL: https://learn.microsoft.com/en-us/azure/ai-foundry/openai/concepts/red-teaming?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Planning red teaming for large language models (LLMs) and their applications Feedback Summarize this article for me In this article This guide offers some potential strategies for planning how to set up and manage red teaming for responsible AI (RAI) risks throughout the large language model (LLM) product life cycle. What is red teaming? The term red teaming has historically described systematic adversarial attacks for testing security vulnerabilities. With the rise of LLMs, the term has extended beyond traditional cybersecurity and evolved in common usage to describe many kinds of probing, testing, and attacking of AI systems. With LLMs, both benign and adversarial usage can produce potentially harmful outputs, which can take many forms, including harmful content such as hate speech, incitement or glorification of violence, or sexual content. Why is RAI red teaming an important practice? Red teaming is a best practice in the responsible development of systems and features using LLMs. While not a replacement for systematic measurement and mitigation work, red teamers help to uncover and identify harms and, in turn, enable measurement strategies to validate the effectiveness of mitigations. While Microsoft has conducted red teaming exercises and implemented safety systems (including content filters and other mitigation strategies ) for its Azure OpenAI in Microsoft Foundry Models (see this Overview of responsible AI practices ), the context of each LLM application will be unique and you also should conduct red teaming to: Test the LLM base model and determine whether there are gaps in the existing safety systems, given the contex --- ## Plan rollout ### Plan rollout URL: https://learn.microsoft.com/en-us/azure/ai-foundry/concepts/planning?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Microsoft Foundry rollout across my organization Feedback Summarize this article for me In this article Note This document refers to the Microsoft Foundry (classic) portal. 🔄 Switch to the Microsoft Foundry (new) documentation if you're using the new portal. Note This document refers to the Microsoft Foundry (new) portal. This guide outlines key decisions for rolling out Microsoft Foundry, including environment setup, data isolation, integration with other Azure services, capacity management, and monitoring. Use this guide as a starting point and adapt it to your needs. For implementation details, see the linked articles for further guidance. Example organization Contoso is a global enterprise exploring GenAI adoption across five business groups, each with distinct needs and technical maturity. To accelerate adoption while maintaining oversight, Contoso Enterprise IT aims to enable a model with common shared resources including networking and centralized data management, while enabling self-serve access to Foundry for each team within a governed, secure environment to manage their use cases. Rollout considerations The Foundry resource defines the scope for configuring, securing, and monitoring your team's environment. It's available in the Foundry portal and through Azure APIs. Projects are like folders to organize your work within this resource context. Projects also control access and permissions to Foundry developer APIs and tools. To ensure consistency, scalability, and governance across teams, consider the following environment setup practices when rolling out Foundry: Establish distinct environments for development, testin --- ## Create your first resource ### Create your first resource URL: https://learn.microsoft.com/en-us/azure/ai-services/multi-service-resource?context=/azure/ai-foundry/context/context?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Quickstart: Set up your first Foundry resource Feedback Summarize this article for me In this article In this quickstart, you create a Microsoft Foundry resource and verify access. Learn how to create and manage a Foundry resource. It's the primary Azure resource type for building, deploying, and managing generative AI models and applications including agents in Azure. An Azure resource is required to use and manage services in Azure. It defines the scope for configuring access, security such as networking, billing, and monitoring. Foundry resource is the next version and renaming of former "Foundry Tools". It provides the application environment for hosting your agents, model deployments, evaluations, and more. A Foundry resource can organize the work for multiple use cases, and is typically shared between a team of developers that work on use cases in a similar business or data domain. Projects act as folders to group related work. Note Only the default project is available in the Foundry (new) portal. Use the Foundry (classic) portal to interact with all other projects on a Foundry resource. Looking to configure Foundry with advanced security settings? See advanced Foundry creation options Create your first resource To create your first resource, with basic Azure settings, follow the below steps using either Azure portal, Azure CLI, or PowerShell. Prerequisites A valid Azure subscription - Create one for free . Azure RBAC role to create resources. You need one of the following roles assigned on your Azure subscription or resource group: Contributor Owner Custom role with Microsoft.CognitiveServices/accounts/write permission C --- ## Create resources using Bicep template ### Create resources using Bicep template URL: https://learn.microsoft.com/en-us/azure/ai-foundry/how-to/create-resource-template?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Quickstart: Create a Microsoft Foundry resource using a Bicep file Feedback Summarize this article for me In this article Use a Microsoft Bicep file (template) to create a Microsoft Foundry resource. A template makes it easy to create resources as a single, coordinated operation. A Bicep file is a text document that defines the resources that are needed for a deployment. It might also specify deployment parameters. You use parameters to provide input values when deploying resources by using the file. Prerequisites An Azure account with an active subscription. If you don't have one, create a free Azure account, which includes a free trial subscription . A copy of the files from the GitHub repo. To clone the GitHub repo to your local machine, you can use Git . Use the following command to clone the quickstart repository to your local machine and navigate to the aifoundry-basics directory. Azure CLI Azure PowerShell git clone https://github.com/Azure-AI-Foundry/foundry-samples cd foundry-samples/infrastructure/infrastructure-setup-bicep/00-basic git clone https://github.com/Azure-AI-Foundry/foundry-samples cd foundry-samples/infrastructure/infrastructure-setup-bicep/00-basic The Bicep command-line tools. To install the Bicep CLI, see Install the Bicep CLI . Access to a role that allows you to complete role assignments, such as Owner . For more information about permissions, see Role-based access control for Microsoft Foundry . Deploy the Bicep file Deploy the Bicep file by using either Azure CLI or Azure PowerShell. Azure CLI Azure PowerShell az group create --name exampleRG --location eastus az deployment group create --resource-g --- ## Manage resources using Terraform ### Manage resources using Terraform URL: https://learn.microsoft.com/en-us/azure/ai-foundry/how-to/create-resource-terraform?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Use Terraform to manage Microsoft Foundry resources Feedback Summarize this article for me In this article Use Terraform to automate the creation of Microsoft Foundry resources, projects, deployments, and connections. You can use either the Terraform AzAPI Provider or AzureRM Provider to manage Foundry resources. The AzAPI provider lets you access all Foundry control plane configurations including preview features. The AzureRM variant is limited to core management capabilities. The following table shows which actions each provider supports: Action AzAPI Provider AzureRM Provider Create a resource group ✅ ✅ Create a Foundry resource ✅ ✅ Configure deployments ✅ ✅ Configure projects ✅ ✅ Configure a connection to knowledge and tools ✅ - Configure a capability host (for advanced tool configurations like Agent standard setup ) ✅ - Terraform enables the definition, preview, and deployment of cloud infrastructure. Using Terraform, you create configuration files using HCL syntax . The HCL syntax allows you to specify the cloud provider - such as Azure - and the elements that make up your cloud infrastructure. After you create your configuration files, you create an execution plan that allows you to preview your infrastructure changes before they're deployed. Once you verify the changes, you apply the execution plan to deploy the infrastructure. Prerequisites An Azure account with an active subscription. If you don't have one, create a free Azure account, which includes a free trial subscription . Access to a role that allows you to create a Foundry resource, such as Azure Account AI Owner or Azure AI Owner on the subscription or resource --- ## Recover or purge deleted resources ### Recover or purge deleted resources URL: https://learn.microsoft.com/en-us/azure/ai-services/recover-purge-resources?context=/azure/ai-foundry/context/context?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Recover or purge deleted Microsoft Foundry resources Feedback Summarize this article for me In this article This article provides instructions on how to recover or purge a Foundry resource that is already deleted. Once you delete a resource, you can't create another one with the same name for 48 hours. To create a resource with the same name, you need to purge the deleted resource. Note The instructions in this article are applicable to both a multi-service resource and a single-service resource. A multi-service resource enables access to multiple Foundry Tools using a single key and endpoint. On the other hand, a single-service resource enables access to just that specific Foundry Tools for which the resource was created. Charges for provisioned deployments on a deleted resource continue until the resource is purged. To prevent unnecessary charges, delete a resource's deployment before deleting the resource. Recover a deleted resource The following prerequisites must be met before you can recover a deleted resource: The resource to be recovered must be deleted within the past 48 hours. The resource to be recovered must not be purged already. A purged resource can't be recovered. Before you attempt to recover a deleted resource, make sure that the resource group for that account exists. If the resource group was deleted, you must recreate it. Recovering a resource group isn't possible. For more information, see Manage resource groups . If the deleted resource used customer-managed keys with Azure Key Vault and the key vault is also deleted, then you must restore the key vault before you restore the Foundry resource. For more inf --- ## Upgrade from Azure OpenAI Service ### Upgrade from Azure OpenAI Service URL: https://learn.microsoft.com/en-us/azure/ai-foundry/how-to/upgrade-azure-openai?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Upgrade from Azure OpenAI to Microsoft Foundry Feedback Summarize this article for me In this article Note This document refers to the Microsoft Foundry (classic) portal. 🔄 Switch to the Microsoft Foundry (new) documentation if you're using the new portal. Note This document refers to the Microsoft Foundry (new) portal. The Microsoft Foundry resource type provides a superset of capabilities compared to the Azure OpenAI resource type. It gives you access to a broader model catalog, agents service, and evaluation capabilities. You can upgrade your Azure OpenAI resource to a Foundry resource. You keep your existing Azure OpenAI API endpoint, state of work, and security configurations, and don't need to create a new Foundry resource. Benefits of upgrading When you upgrade your Azure OpenAI resource to a Foundry resource, you get access to the following capabilities. Feature Azure OpenAI Foundry Models sold directly by Azure Azure OpenAI only Azure OpenAI, Black Forest Labs, DeepSeek, Meta, xAI, Mistral, Microsoft Partner and community models sold through Marketplace - Stability, Cohere, and others ✅ Azure OpenAI API - batch, stored completions, fine-tuning, evaluation, and more ✅ ✅ Agent service ✅ Azure Foundry API ✅ Foundry Tools - Speech, Vision, Language, Content Understanding ✅ Your existing resource configurations and state stay the same, including: Resource name Azure resource tags Network configurations Access and identity configurations API endpoint and API key Custom Domain Name Existing state including fine-tuning jobs, batch, stored completions, and more Limitations Foundry model and feature availability differs by region --- ## Create and manage projects ### Create and manage projects URL: https://learn.microsoft.com/en-us/azure/ai-foundry/how-to/create-projects?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Create a project for Microsoft Foundry Feedback Summarize this article for me In this article Note This document refers to the Microsoft Foundry (classic) portal. 🔄 Switch to the Microsoft Foundry (new) documentation if you're using the new portal. Note This document refers to the Microsoft Foundry (new) portal. This article describes how to create a Foundry project in Microsoft Foundry . Projects let you organize your work—such as agents, evaluations, and files—as you build stateful apps and explore new ideas. A Foundry project is managed under a Microsoft Foundry resource. It's a container for access management, data upload and integration, and monitoring. This lets you keep your work separated between use cases without needing to create extra Azure resources. If you need access to open-source models or PromptFlow, create a hub project type instead. For more information about the different project types, see Types of projects . If your organization requires customized Azure configurations like alternative names, security controls, or cost tags, you might need to use the Azure portal or template options to comply with your organization's Azure Policy requirements. Prerequisites An Azure account with an active subscription. If you don't have one, create a free Azure account, which includes a free trial subscription . Access to a role that allows you to create a Foundry resource, such as Azure Account AI Owner or Azure AI Owner on the subscription or resource group. For more information about permissions, see Role-based access control for Microsoft Foundry . Access to a role that allows you to create a Foundry resource, such as A --- ## Set up your agent resources ### Set up your agent resources URL: https://learn.microsoft.com/en-us/azure/ai-foundry/agents/environment-setup?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Set up your environment Feedback Summarize this article for me In this article Note This document refers to the Microsoft Foundry (classic) portal. 🔄 Switch to the Microsoft Foundry (new) documentation if you're using the new portal. Note This document refers to the Microsoft Foundry (new) portal. Creating your first agent with Foundry Agent Service is a two-step process: Set up your agent environment. Create and configure your agent using either the SDK of your choice or the Azure Foundry portal. Use this article to learn more about setting up your agent environment. Required permissions Action Required Role Create an account and project Azure AI Account Owner standard setup Only: Assign RBAC for required resources (Cosmos DB, Search, Storage, etc.) Role Based Access Control Administrator Create and edit agents Azure AI User Set up your agent environment To get started, you need a Microsoft Foundry resource and a Foundry project. Agents are created within a specific project, and each project acts as an isolated workspace. This means: All agents in the same project share access to the same file storage, thread storage (conversation history), and search indexes. Data is isolated between projects. Agents in one project cannot access resources from another. Projects are currently the unit of sharing and isolation in Foundry. See the what is AI foundry article for more information on Foundry projects. Prerequisites An Azure subscription - Create one for free . Ensure that the individual creating the account and project has the Azure AI Account Owner role at the subscription scope If configuring a standard setup , the same individual --- ## Standard agent setup ### Standard agent setup URL: https://learn.microsoft.com/en-us/azure/ai-foundry/agents/concepts/standard-agent-setup?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Built-in enterprise readiness with standard agent setup Feedback Summarize this article for me In this article Note This document refers to the Microsoft Foundry (classic) portal. 🔄 Switch to the Microsoft Foundry (new) documentation if you're using the new portal. Note This document refers to the Microsoft Foundry (new) portal. Standard Agent Setup uses customer-managed, single-tenant Azure resources to store agent state and keep all agent data under your control. In this setup: Agent states (conversations, responses) are stored in your own Azure resources. Leveraging your own resources for storing customer data Both standard setup configurations are designed to give you complete control over sensitive data by requiring the use of your own Azure resources. The required Bring Your Own (BYO) resources include: BYO File Storage : All files uploaded by developers (during agent configuration) or end-users (during interactions) are stored directly in the customer’s Azure Storage account. BYO Search : All vector stores created by the agent leverage the customer’s Azure AI Search resource. BYO Thread Storage : All customer messages and conversation history will be stored in the customer’s own Azure Cosmos DB account. By bundling these BYO features (file storage, search, and thread storage), the standard setup guarantees that your deployment is secure by default. All data processed by Foundry Agent Service is automatically stored at rest in your own Azure resources, helping you meet internal policies, compliance requirements, and enterprise security standards. Azure Cosmos DB for NoSQL Your existing Azure Cosmos DB for NoSQL Account use --- ## Use your own Azure resources ### Use your own Azure resources URL: https://learn.microsoft.com/en-us/azure/ai-foundry/agents/how-to/use-your-own-resources?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Use your own resources Feedback Summarize this article for me In this article Note This document refers to the Microsoft Foundry (classic) portal. 🔄 Switch to the Microsoft Foundry (new) documentation if you're using the new portal. Note This document refers to the Microsoft Foundry (new) portal. Use this article if you want to set up your Foundry project with your own resources. Limitations There are some limitations you should be aware of when you plan to use existing resources with the Foundry Agent Service. If you are using a hub-based project or Azure OpenAI Assistants At this time, there is no direct upgrade path to migrate existing agents or their associated data assets such as files, conversations, or vector stores from a hub-based project to a Microsoft Foundry project. There is also no upgrade path to convert existing Azure OpenAI Assistants into Foundry Agents, nor a way to automatically migrate Assistants' files, conversations, or vector stores. You can reuse your existing model deployments and quota from Foundry Tools or Azure OpenAI resources within a Foundry project. SDK usage with hub-based projects Starting in May 2025, the Azure AI Agent Service uses an endpoint for Foundry projects instead of the connection string that was used for hub-based projects before this time. Connection strings are no longer supported in current versions of the SDKs and REST API. We recommend creating a new foundry project. If you want to continue using your hub-based project and connection string, you will need to: Use the connection string for your project located under Connection string in the overview of your project. Use one of t --- ## Migrate agents ### Migrate agents URL: https://learn.microsoft.com/en-us/azure/ai-foundry/agents/how-to/migrate?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Upgrading to the new agents developer experience Feedback Summarize this article for me In this article Tip You can use the available migration tool to migrate from the Assistants API to Agents. The new agents offer an upgraded experience which enables developers and enterprises to design intelligent agents that are easy to build, version, operate, and observe. It introduces a modernized API and SDK, new enterprise-grade capabilities, and preserves the identity, governance, and observability features customers rely on today. Key benefits The new agents provide the following benefits: New Agent Types : Create prompt-based, workflow-based, or container-based agents. More models : Generate responses using any Azure Foundry model either in your agent or directly as a response generation call. Enterprise Readiness : Use single tenant storage, with the added option to bring your own Cosmos DB to store state and keep your data secure. Single or Multi-Agent Workflows : Easily chain agents for complex workflows. More features : Web Search, File Search, Code Interpreter, MCP tool calling, image generation, and reasoning summaries. Stateful Context : Automatically retains context across calls unless opted out using store: false. Deployable Agents : agents can be exposed as individual endpoints. Enhanced Security : Control who can run or modify agent definitions. Separation of Duties : Define agents once; execute with various inputs. Superset of Responses API : builds on Responses API and adds more capabilities. Improved State Management : Uses conversations instead of threads/messages. Modern API Primitive : built on Responses API instead --- ## Virtual networks ### Virtual networks URL: https://learn.microsoft.com/en-us/azure/ai-foundry/agents/how-to/virtual-networks?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Create a new network-secured environment with user-managed identity Feedback Summarize this article for me In this article Note This document refers to the Microsoft Foundry (classic) portal. 🔄 Switch to the Microsoft Foundry (new) documentation if you're using the new portal. Note This document refers to the Microsoft Foundry (new) portal. Foundry Agent Service offers Standard Setup with private networking environment setup, allowing you to bring your own (BYO) private virtual network. This setup creates an isolated network environment that lets you securely access data and perform actions while maintaining full control over your network infrastructure. This guide provides a step-by-step walkthrough of the setup process and outlines all necessary requirements. Tip See the FAQ article for common questions when working with Virtual Networks. Note End-to-end network isolation is not supported in the new Foundry portal experience. Please use the classic Foundry portal experience or the SDK or CLI to securely access your Foundry projects when network isolation is enabled. Security features By default, the Standard Setup with Private Network Isolation ensures: No public egress : foundational infrastructure ensures the right authentication and security for your agents and tools, without you having to do trusted service bypass. Container injection : allows the platform network to host APIs and inject a subnet into your network, enabling local communication of your Azure resources within the same virtual network. Private resource access : If your resources are marked as private and nondiscoverable from the internet, the platform network --- ## Use an AI Gateway ### Use an AI Gateway URL: https://learn.microsoft.com/en-us/azure/ai-foundry/agents/how-to/ai-gateway?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Bring your own AI gateway to Azure AI Agent Service (preview) Feedback Summarize this article for me In this article The Azure AI Agent Service allows you to connect and use models hosted behind your enterprise AI gateways such as Azure API Management or other non-Azure hosted AI model gateways . This capability allows you to maintain control over your model endpoints while leveraging the power of Foundry's agent capabilities. Note This feature is currently in preview. Consider the preview conditions before enabling this feature. This capability enables organizations to: Maintain control over their model endpoints. Keep your model endpoints secure behind your existing enterprise infrastructure. Integrate securely with enterprise gateways. Leverage your existing gateway investments and security policies. Build agents that leverage models without exposing them publicly. Apply your organization's compliance and governance requirements to AI model access. View the diagram to understand the potential flows from the Agent service to your gateway and models behind it: Prerequisites An Azure subscription with access to Microsoft Foundry. Create a Foundry resource in your subscription. Installed Azure CLI and Agent SDK . Access credentials for your enterprise AI gateway (for example API Management or another non-Azure AI model gateway). GitHub access for Foundry samples Connections for AI gateway Depending on the AI gateway you would like to use, there are two different connections you can create to your resource from Microsoft Foundry. For more details on these connections, see the samples on GitHub API Management (APIM) connection APIM --- ## Create a connection ### Create a connection URL: https://learn.microsoft.com/en-us/azure/ai-foundry/how-to/connections-add?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Add a new connection to your project Feedback Summarize this article for me In this article Note This document refers to the Microsoft Foundry (classic) portal. 🔄 Switch to the Microsoft Foundry (new) documentation if you're using the new portal. Note This document refers to the Microsoft Foundry (new) portal. Tip An alternate hub-scoped connections article is available: Create and manage connections (Hubs) . Important Items marked (preview) in this article are currently in public preview. This preview is provided without a service-level agreement, and we don't recommend it for production workloads. Certain features might not be supported or might have constrained capabilities. For more information, see Supplemental Terms of Use for Microsoft Azure Previews . In this article, you learn how to add a new connection in Microsoft Foundry portal . Connections are a way to authenticate and consume both Microsoft and other resources within your Foundry projects. They're required for scenarios such as building Standard Agents or building with Agent knowledge tools. Certain connections can be created in the Foundry UI while others require deployment through code in Bicep template. See our foundry-samples on GitHub . Read the table descriptions below to learn more. Prerequisites If you don't have one, create a project . Connection types Service connection type Preview Description Azure AI Search Azure AI Search is an Azure resource that supports information retrieval over your vector and textual data stored in search indexes. Required for Standard Agent deployment. Azure Storage Azure Storage is a cloud storage solution for storing unstru --- ## Connect to your own storage in Foundry ### Connect to your own storage in Foundry URL: https://learn.microsoft.com/en-us/azure/ai-foundry/how-to/bring-your-own-azure-storage-foundry?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Connect to your own storage Feedback Summarize this article for me In this article Note This document refers to the Microsoft Foundry (classic) portal. 🔄 Switch to the Microsoft Foundry (new) documentation if you're using the new portal. Note This document refers to the Microsoft Foundry (new) portal. Important Items marked (preview) in this article are currently in public preview. This preview is provided without a service-level agreement, and we don't recommend it for production workloads. Certain features might not be supported or might have constrained capabilities. For more information, see Supplemental Terms of Use for Microsoft Azure Previews . Microsoft Foundry brings Agents, Azure OpenAI, Speech, and Language services together under one unified resource type. Bring-your-own-storage (BYOS) lets you route data produced by these capabilities to an Azure Storage account that you own and govern. The configuration patterns align with (and provide backwards compatibility to) earlier standalone Speech and Language resource types. This article shows you how to connect your storage to Foundry by using two overarching approaches: Connections: recommended baseline for most features. Connections provide the shared data pointer. Capability hosts: optionally override/explicitly bind a specific feature (for example, Agents standard setup) to one connection among several. userOwnedStorage field: a resource-level binding used only by Speech and Language. Prerequisites Before connecting your storage, ensure you have: An Azure account with an active subscription. If you don't have one, create a free Azure account, which includes a free tri --- ## Connect to your own storage for Speech/Language ### Connect to your own storage for Speech/Language URL: https://learn.microsoft.com/en-us/azure/ai-foundry/how-to/bring-your-own-azure-storage-speech-language-services?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Connect your own storage for Speech and Language services (Preview) Feedback Summarize this article for me In this article Note This document refers to the Microsoft Foundry (classic) portal. 🔄 Switch to the Microsoft Foundry (new) documentation if you're using the new portal. Note This document refers to the Microsoft Foundry (new) portal. Important Items marked (preview) in this article are currently in public preview. This preview is provided without a service-level agreement, and we don't recommend it for production workloads. Certain features might not be supported or might have constrained capabilities. For more information, see Supplemental Terms of Use for Microsoft Azure Previews . Configure bring-your-own-storage (BYOS) for Speech and Language capabilities in a Foundry resource by setting the userOwnedStorage binding at creation time. This binding routes Speech and Language data to your Azure Storage account while maintaining backward compatibility with earlier standalone resource patterns. Tip Use this article when you specifically need Speech and Language data to land in storage you own. For the broader approaches (connections, capability hosts), see Connect to your own storage . Prerequisites An Azure account with an active subscription. If you don't have one, create a free Azure account, which includes a free trial subscription . An Azure Storage account (Blob) in a region supported by your Foundry resource. Resource group permissions : Owner or Contributor role on the resource group containing the Foundry resource. Storage account permissions : Storage Blob Data Contributor role on the storage account (assigned to --- ## Manage Grounding with Bing ### Manage Grounding with Bing URL: https://learn.microsoft.com/en-us/azure/ai-foundry/agents/how-to/manage-grounding-with-bing?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Manage Grounding With Bing in Microsoft Foundry and Azure Feedback Summarize this article for me In this article Grounding with Bing enables agents to retrieve and incorporate real-time public web data into model-generated responses. It supports summarization, question answering, conversational assistance, and other scenarios by using Grounding with Bing Search or Grounding with Bing Custom Search to fill knowledge gaps. Grounding is available across features in Foundry Agent Service and Azure AI Search. You might need to disable access to these features to meet compliance, privacy, or data governance requirements. As an admin, you can manage access to Grounding with Bing in the following ways: Disable Grounding with Bing Search tools in Foundry Agent Service. Disable web search tool in Foundry Agent Service. Disable web knowledge in Azure AI Search. Disable Grounding with Bing Search tools You can disable Grounding with Bing Search and/or Grounding with Bing Custom Search at the subscription or resource group level. For more information, see Disable use of Grounding with Bing Search and Grounding with Bing Custom Search . Disable web search tool You can disable the web search tool for all accounts in a subscription. For more information, see Disable Bing Web Search . Disable web knowledge You can disable Web Knowledge Source access for all search services in a subscription. For more information, see Disable use of Web Knowledge Source . Related content Grounding with Bing Search tools for agents Web search tool (preview) Create a Web Knowledge Source resource Feedback Was this page helpful? Yes No No Need help with this topic? --- ## Manage quotas for Foundry resources ### Manage quotas for Foundry resources URL: https://learn.microsoft.com/en-us/azure/ai-foundry/how-to/quota?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Manage and increase quotas for resources with Microsoft Foundry (Foundry projects) Feedback Summarize this article for me In this article Note This document refers to the Microsoft Foundry (classic) portal. 🔄 Switch to the Microsoft Foundry (new) documentation if you're using the new portal. Note This document refers to the Microsoft Foundry (new) portal. Tip An alternate hub-focused quota article is available: Manage and increase quotas for hub resources . Quota provides the flexibility to actively manage the allocation of rate limits across the deployments within your subscription. This article walks through the process of managing quota for your Microsoft Foundry Models (Foundry projects). Quota provides the flexibility to actively manage the allocation of rate limits across the deployments within your subscription. This article walks through the process of managing quota for your Foundry Models (Foundry projects). Azure uses limits and quotas to prevent budget overruns due to fraud and to honor Azure capacity constraints. It's also a good way for admins to control costs. Consider these limits as you scale for production workloads. In this article, you learn about: Viewing your quotas and limits Requesting quota and limit increases Foundry shared quota Foundry provides a pool of shared quota that different users across various regions can use concurrently. Depending on availability, users can temporarily access quota from the shared pool and use the quota to perform testing for a limited amount of time. The specific time duration depends on the use case. By temporarily using quota from the quota pool, you no longer need to fi --- ## Foundry Models quotas and limits ### Foundry Models quotas and limits URL: https://learn.microsoft.com/en-us/azure/ai-foundry/foundry-models/quotas-limits?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Microsoft Foundry Models quotas and limits Feedback Summarize this article for me In this article Note This document refers to the Microsoft Foundry (classic) portal. 🔄 Switch to the Microsoft Foundry (new) documentation if you're using the new portal. Note This document refers to the Microsoft Foundry (new) portal. This article provides a quick reference and detailed description of the quotas and limits for Foundry Models sold directly by Azure . For quotas and limits specific to the Azure OpenAI in Foundry Models, see Quota and limits in Azure OpenAI . Quotas and limits reference Azure uses quotas and limits to prevent budget overruns due to fraud and to honor Azure capacity constraints. Consider these limits as you scale for production workloads. The following sections provide a quick guide to the default quotas and limits that apply to Azure AI model inference service in Foundry: Resource limits (per Azure subscription, per region) Limit name Limit value Foundry resources per region per Azure subscription 100 Max projects per resource 250 Max deployments per resource (model deployments within a Foundry resource) 32 Rate limits The following table lists limits for Foundry Models for the following rates: Tokens per minute Requests per minute Concurrent request Models Tokens per minute Requests per minute Concurrent requests Azure OpenAI models Varies per model and SKU. See limits for Azure OpenAI . Varies per model and SKU. See limits for Azure OpenAI . not applicable - DeepSeek-R1 - DeepSeek-V3-0324 5,000,000 5,000 300 - Llama 3.3 70B Instruct - Llama-4-Maverick-17B-128E-Instruct-FP8 - Grok 3 - Grok 3 mini 400,000 1,000 300 - --- ## Azure OpenAI quotas and limits ### Azure OpenAI quotas and limits URL: https://learn.microsoft.com/en-us/azure/ai-foundry/openai/quotas-limits?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Azure OpenAI in Microsoft Foundry Models quotas and limits Feedback Summarize this article for me In this article This article contains a quick reference and a detailed description of the quotas and limits for Azure OpenAI. Scope of quota Quotas and limits aren't enforced at the tenant level. Instead, the highest level of quota restrictions is scoped at the Azure subscription level. Regional quota allocation Tokens per minute (TPM) and requests per minute (RPM) limits are defined per region , per subscription , and per model or deployment type . For example, if the gpt-4.1 Global Standard model is listed with a quota of 5 million TPM and 5,000 RPM , then each region where that model or deployment type is available has its own dedicated quota pool of that amount for each of your Azure subscriptions. Within a single Azure subscription, it's possible to use a larger quantity of total TPM and RPM quota for a given model and deployment type, as long as you have resources and model deployments spread across multiple regions. Quotas and limits reference The following section provides you with a quick guide to the default quotas and limits that apply to Azure OpenAI: Limit name Limit value Azure OpenAI resources per region, per Azure subscription 30. Default DALL-E 2 quota limits 2 concurrent requests. Default DALL-E 3 quota limits 6 requests per minute Default GPT-image-1 quota limits 9 requests per minute Default GPT-image-1-mini quota limits 12 requests per minute Default GPT-image-1.5 quota limits 9 requests per minute Default Sora quota limits 60 requests per minute. Default Sora 2 quota limits 2 job requests 1 per minute Default s --- ## Authentication and authorization ### Authentication and authorization URL: https://learn.microsoft.com/en-us/azure/ai-foundry/concepts/authentication-authorization-foundry?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Authentication and authorization in Microsoft Foundry Feedback Summarize this article for me In this article Authentication and authorization in Microsoft Foundry define how principals prove identity and gain permission to perform control plane and data plane operations. Foundry supports API key and Microsoft Entra ID token-based authentication. Microsoft Entra ID enables conditional access, managed identities, granular role-based access control (RBAC) actions, and least privilege scenarios. API keys remain available for rapid prototyping and legacy integration but lack per-user traceability. This article explains the control plane and data plane model, compares API key and Microsoft Entra ID authentication, maps identities to roles, and describes common least privilege scenarios. Important Use Microsoft Entra ID for production workloads to enable conditional access, managed identities, and least privilege RBAC. API keys are convenient for quick evaluation but provide coarse-grained access. Prerequisites An Azure subscription. If you don't have one, create a free account . A Microsoft Foundry resource with a custom subdomain configured. Understanding of Azure RBAC concepts . To assign roles, you need the Owner role or User Access Administrator role at the appropriate scope. (Optional) The Azure CLI or Azure SDK for Python installed for programmatic authentication. Control plane and data plane Azure operations divide into two categories: control plane and data plane. Azure separates resource management (control plane) from operational runtime (data plane). Therefore, you use the control plane to manage resources in your subscript --- ## Disable preview features with role-based access control ### Disable preview features with role-based access control URL: https://learn.microsoft.com/en-us/azure/ai-foundry/concepts/disable-preview-features-with-rbac?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Disable preview features in Microsoft Foundry by using role-based access control Feedback Summarize this article for me In this article Note This document refers to the Microsoft Foundry (classic) portal. 🔍 View the Microsoft Foundry (new) documentation to learn about the new portal. In Microsoft Foundry projects, some features are in preview. You can block access to these features by excluding specific data actions from a custom role, and then assigning that role to users. This article lists the data actions for each preview feature so you can block features individually. Because you can't modify built-in roles in Foundry projects, you need to create a custom role. Prerequisites A Microsoft Foundry project. Permissions to create custom roles at the scope where you want the role to be assignable (for example, Owner or User Access Administrator). Permissions to assign roles at the scope where you assign access (for example, Role Based Access Control Administrator or User Access Administrator). Example: Create a custom role that blocks a preview feature This example shows the JSON shape for a custom role definition and where to put the preview feature data actions. If you clone an existing role or use wildcard permissions, add the preview feature data actions to notDataActions so the role excludes them. { "properties": { "roleName": "Foundry custom role (preview features blocked)", "description": "Custom role that excludes specific Foundry preview features.", "assignableScopes": [ "/subscriptions/" ], "permissions": [ { "actions": [], "notActions": [], "dataActions": [], "notDataActions": [ "Microsoft.CognitiveSer --- ## Role-based access control ### Role-based access control URL: https://learn.microsoft.com/en-us/azure/ai-foundry/concepts/rbac-foundry?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Role-based access control for Microsoft Foundry Feedback Summarize this article for me In this article Note This document refers to the Microsoft Foundry (classic) portal. 🔄 Switch to the Microsoft Foundry (new) documentation if you're using the new portal. Note This document refers to the Microsoft Foundry (new) portal. Tip An alternate hub-focused RBAC article is available: Role-based access control for Microsoft Foundry (Hubs and Projects) . In this article, you learn about role-based access control (RBAC) in your Microsoft Foundry resource and how to assign roles that control access to resources. Tip RBAC roles apply when you authenticate using Microsoft Entra ID. If you use key-based authentication instead, the key grants full access without role restrictions. Microsoft recommends using Entra ID authentication for improved security and granular access control. In this article, you learn how to manage access to Microsoft Foundry resources using role-based access control (RBAC). Tip RBAC roles apply when you authenticate using Microsoft Entra ID. If you use key-based authentication instead, the key grants full access without role restrictions. Microsoft recommends using Entra ID authentication for improved security and granular access control. For more information about authentication and authorization in Microsoft Foundry, see Authentication and Authorization . This article mentions terminology explained in the previous article. Getting started For new users to Azure and Microsoft Foundry, use the following check-list to ensure all the correct roles are assigned to your user principal and your project's managed identity to g --- ## Configure keyless authentication ### Configure keyless authentication URL: https://learn.microsoft.com/en-us/azure/ai-foundry/foundry-models/how-to/configure-entra-id?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Configure keyless authentication with Microsoft Entra ID Feedback Summarize this article for me In this article Note This document refers to the Microsoft Foundry (classic) portal. 🔄 Switch to the Microsoft Foundry (new) documentation if you're using the new portal. Note This document refers to the Microsoft Foundry (new) portal. This article explains how to configure keyless authentication with Microsoft Entra ID for Microsoft Foundry Models. Keyless authentication enhances security by eliminating the need for API keys, simplifies the user experience with role-based access control (RBAC), and reduces operational complexity while providing robust compliance support. Prerequisites To complete this article, you need: An Azure subscription. If you're using GitHub Models , you can upgrade your experience and create an Azure subscription in the process. Read Upgrade from GitHub Models to Microsoft Foundry Models if that's your case. A Foundry project. This kind of project is managed under a Foundry resource. If you don't have a Foundry project, see Create a project for Foundry (Foundry projects) The endpoint's URL. An account with Microsoft.Authorization/roleAssignments/write and Microsoft.Authorization/roleAssignments/delete permissions, such as the Administrator role-based access control. See the next section on Required Azure roles and permissions for more details. Required Azure roles and permissions Microsoft Entra ID uses role-based access control (RBAC) to manage access to Azure resources. You need different roles, depending on whether you're setting up authentication (administrator) or using it to make API calls (developer). --- ## Configure private link ### Configure private link URL: https://learn.microsoft.com/en-us/azure/ai-foundry/how-to/configure-private-link?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . How to configure a private link for Microsoft Foundry (Foundry projects) Feedback Summarize this article for me In this article Tip An alternate hub-focused version of this article is available: How to configure a private link for a Microsoft Foundry hub . When you use a Foundry project, you can use a private link to secure communication with your project. This article describes how to establish a private connection to your project using a private link. Note End-to-end network isolation isn't supported in the new Foundry portal experience. Use the classic Foundry portal experience or the SDK or CLI to securely access your Foundry projects when network isolation is enabled. For more on limitations with private networking in Foundry, see limitations . Prerequisites An existing Azure virtual network and subnet to create the private endpoint in. Azure permissions to create and approve private endpoint connections: On the virtual network: Network Contributor (or equivalent) to create the private endpoint. On the Foundry project resource: Contributor (or Owner ) to create private endpoint connections. If you don't have approval permissions, the private endpoint connection stays in a Pending state until the resource owner approves it. If you manage private DNS zones: Private DNS Zone Contributor (or equivalent) for the private DNS zone that you link to the virtual network. Important Don't use the 172.17.0.0/16 IP address range for your virtual network. This range is the default subnet range used by the Docker bridge network on-premises. Securely connect to Foundry To connect to Foundry secured by a virtual network, use one of these met --- ## Managed virtual network ### Managed virtual network URL: https://learn.microsoft.com/en-us/azure/ai-foundry/how-to/managed-virtual-network?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Configure managed virtual network for Microsoft Foundry projects Feedback Summarize this article for me In this article Important Items marked (preview) in this article are currently in public preview. This preview is provided without a service-level agreement, and we don't recommend it for production workloads. Certain features might not be supported or might have constrained capabilities. For more information, see Supplemental Terms of Use for Microsoft Azure Previews . This article explains how to set up a managed virtual network for your Foundry resource. Managed virtual network streamlines and automates network isolation for your Foundry resource by provisioning a Microsoft‑managed virtual network that secures the Agents service underlying compute within your Foundry projects. When enabled, Agents outbound network traffic is secured by this managed network boundary, and the isolation mode you choose governs all the traffic. You can create the required private endpoints to dependent Azure services and apply the necessary network rules, giving you a secure default without requiring you to build or maintain your own virtual network. This managed network restricts what your Agents can access, helping prevent data exfiltration while still allowing connectivity to approved Azure resources. Before continuing, consider the limitations of the offering and review the prerequisites. This feature is currently in public preview, so consider preview conditions before enabling this network isolation method. If you're not allowed to use preview features in your enterprise, use the existing GA supported custom virtual network support for Ag --- ## Network security perimeter ### Network security perimeter URL: https://learn.microsoft.com/en-us/azure/ai-foundry/how-to/add-foundry-to-network-security-perimeter?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Add Microsoft Foundry to a network security perimeter Feedback Summarize this article for me In this article Use a network security perimeter (NSP) to restrict data-plane access to your Microsoft Foundry resource and group it with other protected PaaS resources. An NSP lets you: Enforce inbound and outbound access rules instead of broad public exposure. Reduce data exfiltration risk by containing traffic within a logical boundary. Centrally log network access decisions across associated resources. This article gives only the Foundry-specific pointers you need. All procedural detail for creating perimeters, defining access rules, enabling logging, and using APIs lives in existing Azure networking documentation. Follow the links in each section for the authoritative steps. Note You must use a Foundry project for this feature. A hub-based project isn't supported. See How do I know which type of project I have? and Create a Foundry project . To migrate your hub-based project to a Foundry project, see Migrate from hub-based to Foundry projects . Prerequisites An Azure subscription where you can create and manage network security perimeter resources. At minimum, use an account with the Owner , Contributor , or Network Contributor role (or a custom role with equivalent permissions). A Foundry resource. A network security perimeter (NSP) and profile. If you use Azure CLI automation, Azure CLI 2.75.0 or later. If you want to query access logs, a Log Analytics workspace. For the full set of required actions and permissions (profiles, associations, access rules, diagnostic settings), see Azure RBAC permissions required for network security --- ## Configure customer-managed keys ### Configure customer-managed keys URL: https://learn.microsoft.com/en-us/azure/ai-foundry/concepts/encryption-keys-portal?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Customer-managed keys (CMK) for Microsoft Foundry Feedback Summarize this article for me In this article Note This document refers to the Microsoft Foundry (classic) portal. 🔄 Switch to the Microsoft Foundry (new) documentation if you're using the new portal. Note This document refers to the Microsoft Foundry (new) portal. Tip An alternate hub-focused CMK article is available: Customer-managed keys for hub projects . Customer-managed key (CMK) encryption in Microsoft Foundry gives you control over encryption of your data. Use CMKs to add an extra protection layer and help meet compliance requirements with Azure Key Vault integration. Customer-managed key (CMK) encryption in Microsoft Foundry gives you control over encryption of your data. Use CMKs to add an extra protection layer and help meet compliance requirements with Azure Key Vault integration. Microsoft Foundry provides robust encryption capabilities, including the ability to use customer-managed keys (CMKs) stored in Azure Key Vault to secure your sensitive data. This article explains the concept of encryption with CMKs and provides step-by-step guidance for configuring CMK using Azure Key Vault. It also discusses encryption models and access control methods like Azure Role-Based Access Control (RBAC) and Vault Access Policies and ensuring compatibility with system-assigned managed identities and user-assigned managed identities (UAI) . Why use customer-managed keys? With CMK, you gain full control over encryption keys, providing enhanced protection for sensitive data and helping meet compliance requirements. The key benefits of using CMKs include: Using your own keys to --- ## Store secrets in your Azure Key Vault ### Store secrets in your Azure Key Vault URL: https://learn.microsoft.com/en-us/azure/ai-foundry/how-to/set-up-key-vault-connection?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Set up an Azure Key Vault connection in Microsoft Foundry Feedback Summarize this article for me In this article Note This document refers to the Microsoft Foundry (classic) portal. 🔄 Switch to the Microsoft Foundry (new) documentation if you're using the new portal. Note This document refers to the Microsoft Foundry (new) portal. If you don't set up a Key Vault connection, Microsoft Foundry stores connection details in a Microsoft-managed Key Vault outside your subscription. To manage your own secrets, connect your Azure Key Vault to Foundry. Before you begin, review the limitations for Key Vault connections. Azure Key Vault is a cloud service for securely storing and accessing secrets. A secret is anything that you want to tightly control access to, such as API keys, passwords, certificates, or cryptographic keys. For more information, see About Azure Key Vault . Prerequisites An Azure subscription. A Foundry resource with no existing connections at the resource or project level. An Azure Key Vault in your subscription, or permissions to create one. One of the following Azure RBAC roles on your Key Vault: Key Vault Secrets Officer (minimal permissions) Key Vault Contributor Key Vault Administrator Limitations Create Azure Key Vault connections only when you need them. If you bring your own Azure Key Vault, review these limitations: Limit Azure Key Vault connections to one per Foundry resource. Delete an Azure Key Vault connection only if no other connections exist at the Foundry resource or project level. Foundry doesn't support secret migration. Remove and recreate connections yourself. Deleting the underlying Azure Key Vault --- ## Rotate API access keys ### Rotate API access keys URL: https://learn.microsoft.com/azure/ai-services/rotate-keys?context=/azure/ai-foundry/context/context?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Rotate API keys Feedback Summarize this article for me In this article Each resource has two API keys to enable secret rotation. This is a security precaution that lets you regularly change the keys that can be used to access your service, protecting the privacy of your resource if a key gets leaked. How to rotate keys You can rotate keys using the following procedure: If you're using both keys in production, change your code so that only one key is in use. In this guide, assume it's key 1. This is a necessary step because once a key is regenerated, the older version of that key stops working immediately. This would cause clients using the older key to get 401 access denied errors. Once you have only key 1 in use, you can regenerate key 2. Go to your resource's page on the Azure portal, select the Keys and Endpoint tab, and select the Regenerate Key 2 button at the top of the page. Next, update your code to use the newly generated key 2. It helps to have logs or availability to check that users of the key have successfully swapped from using key 1 to key 2 before you proceed. Now you can regenerate key 1 using the same process. Finally, update your code to use the new key 1. See also Configure key-less authentication Feedback Was this page helpful? Yes No No Need help with this topic? Want to try using Ask Learn to clarify or guide you through this topic? Ask Learn Ask Learn Suggest a fix? Additional resources Last updated on 2025-10-02 --- ## Built-in policy definitions ### Built-in policy definitions URL: https://learn.microsoft.com/en-us/azure/ai-services/policy-reference?context=/azure/ai-foundry/context/context?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Azure Policy built-in policy definitions for Foundry Tools Feedback Summarize this article for me In this article This page is an index of Azure Policy built-in policy definitions for Foundry Tools. For additional Azure Policy built-ins for other services, see Azure Policy built-in definitions . The name of each built-in policy definition links to the policy definition in the Azure portal. Use the link in the Version column to view the source on the Azure Policy GitHub repo . Foundry Tools Name (Azure portal) Description Effect(s) Version (GitHub) [Preview]: Cognitive Services Deployments should only use allowed completion content filtering Mandate minimum levels of content filtering for completion content for model deployments within your organization. Audit, Disabled 1.0.0-preview [Preview]: Cognitive Services Deployments should only use allowed control Mandate minimum levels of multi-severity content filtering for harmful content for model deployments within your organization. Audit, Disabled 1.0.0-preview [Preview]: Cognitive Services Deployments Should Only Use Allowed Control Mode Mandate content filtering mode for model deployments within your organization. Audit, Disabled 1.0.0-preview [Preview]: Cognitive Services Deployments should only use allowed prompt content filtering Mandate minimum levels of content filtering for prompt content for model deployments within your organization. Audit, Disabled 1.0.0-preview Azure AI Services resources should encrypt data at rest with a customer-managed key (CMK) Using customer-managed keys to encrypt data at rest provides more control over the key lifecycle, including rotation and --- ## Built-in policy for model deployment ### Built-in policy for model deployment URL: https://learn.microsoft.com/en-us/azure/ai-foundry/how-to/model-deployment-policy?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Built-in policy for model deployment in Microsoft Foundry portal Feedback Summarize this article for me In this article Note This document refers to the Microsoft Foundry (classic) portal. 🔄 Switch to the Microsoft Foundry (new) documentation if you're using the new portal. Note This document refers to the Microsoft Foundry (new) portal. Azure Policy provides built-in policy definitions that help you govern the deployment of AI models in Microsoft Foundry portal. You can use these policies to control what models your developers can deploy in the Foundry portal. Prerequisites An Azure account with an active subscription. If you don't have one, create a free Azure account . Your Azure account lets you access the Foundry portal. Permissions to create and assign policies. To create and assign policies, you must be an Owner or Resource Policy Contributor at the Azure subscription or resource group level. Familiarity with Azure Policy. To learn more, see What is Azure Policy? . Assign the policy Azure CLI Azure portal Use Azure CLI to find the built-in policy definition and assign it at a scope. Sign in and select the subscription you want to work in: az login az account set --subscription "" Find the policy definition ID for the built-in definition: az policy definition list \ --query "[?displayName=='Cognitive Services Deployments should only use approved Registry Models'].{name:name, id:id}" \ --output table Expected result: a row that includes the policy id . Create a parameters file (example): { "effect": { "value": "Deny" }, "allowedModelsPublishers": { "value": ["OpenAI"] }, "allowedAssetIds": { "value": [ "azu --- ## Create custom policy definitions ### Create custom policy definitions URL: https://learn.microsoft.com/en-us/azure/ai-foundry/how-to/custom-policy-definition?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Create custom policies for Microsoft Foundry Feedback Summarize this article for me In this article Note This document refers to the Microsoft Foundry (classic) portal. 🔄 Switch to the Microsoft Foundry (new) documentation if you're using the new portal. Note This document refers to the Microsoft Foundry (new) portal. Learn how to use custom Azure policies to enable teams to self-manage Microsoft Foundry resources. Apply guardrails and constraints on allowed configurations so you can provide flexibility while meeting security and compliance requirements. By using custom policies, you can: Enforce governance : Prevent unauthorized creation of Foundry hubs, projects, connections, or capability hosts. Control resource behavior : Ensure security configurations, enforce tagging, or allow only approved integrations. Ensure compliance : Apply enterprise security and operational standards consistently across environments. Prerequisites An Azure account with an active subscription. If you don't have one, create a free Azure account, which includes a free trial subscription . Access to a role that allows you to complete role assignments, such as Owner . For more information about permissions, see Role-based access control for Microsoft Foundry . For more information, see What is Azure Policy? Steps to create a custom policy Open policy in the Azure portal Go to Azure portal . Search for Policy and select it. Define a new policy In the Authoring section, select Definitions > + Policy definition . Provide: Definition location : Subscription or management group. Name : A unique name (for example, Deny-Unapproved-Connections ). Description : --- ## High availability and resiliency ### High availability and resiliency URL: https://learn.microsoft.com/en-us/azure/ai-foundry/how-to/high-availability-resiliency?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . High availability and resiliency for Microsoft Foundry projects and Agent Services Feedback Summarize this article for me In this article Note This document refers to the Microsoft Foundry (classic) portal. 🔄 Switch to the Microsoft Foundry (new) documentation if you're using the new portal. Note This document refers to the Microsoft Foundry (new) portal. Important Items marked (preview) in this article are currently in public preview. This preview is provided without a service-level agreement, and we don't recommend it for production workloads. Certain features might not be supported or might have constrained capabilities. For more information, see Supplemental Terms of Use for Microsoft Azure Previews . Plan ahead to maintain business continuity and prepare for disaster recovery with Microsoft Foundry . Microsoft strives to ensure that Azure services are always available. However, unplanned service outages might occur. Create a disaster recovery plan to handle regional service outages. In this article, you learn how to: Plan a multi-region deployment of Foundry and associated resources. Maximize your chances to recover logs, notebooks, Docker images, and other metadata. Design your solution for high availability. Fail over to another region. Important Foundry itself doesn't provide automatic failover or disaster recovery. :::moniker-range="foundry-classic" Note The information in this article applies only to Foundry project . For disaster recovery for hub-based project , see Disaster recovery for Foundry hubs . :::moniker-end Prerequisites An Azure subscription. If you don't have one, create a free account . A Microsoft Foundr --- ## Disaster recovery for agent services ### Disaster recovery for agent services URL: https://learn.microsoft.com/en-us/azure/ai-foundry/how-to/agent-service-disaster-recovery?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Foundry Agent Service disaster recovery Feedback Summarize this article for me In this article Note This document refers to the Microsoft Foundry (classic) portal. 🔄 Switch to the Microsoft Foundry (new) documentation if you're using the new portal. Note This document refers to the Microsoft Foundry (new) portal. Use this article as the starting point for disaster recovery (DR) planning for Foundry Agent Service in the Standard deployment mode . It explains what you can and can't recover, what to prepare before an incident, and where to find recovery procedures for platform outages and resource or data loss. Important This is the overview article in a three-part series. You're here: Understand limitations, prevention controls, and required baseline configuration. For prolonged regional outages and platform incidents, see Agent Service platform outage recovery strategies . For human-caused or automation-caused deletions and localized data loss, see Agent Service resource and data loss recovery strategies . To reduce the likelihood of recovery events, see High availability and resiliency for Foundry projects and agent services . Scope and definitions This series focuses on DR for Foundry projects that use Agent Service in Standard deployment mode. Blast radius boundary : In most workloads, a single Foundry project is the recovery unit. State : Agent definitions, conversation threads (including user-uploaded files), and any file-based knowledge stored in the capability host dependencies. Data plane APIs : APIs used to create, update, and invoke agents and threads. For details, see AI Agents REST API operation groups . For general r --- ## Disaster recovery from a platform outage ### Disaster recovery from a platform outage URL: https://learn.microsoft.com/en-us/azure/ai-foundry/how-to/agent-service-platform-disaster-recovery?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Foundry Agent Service platform outage recovery Feedback Summarize this article for me In this article Note This document refers to the Microsoft Foundry (classic) portal. 🔄 Switch to the Microsoft Foundry (new) documentation if you're using the new portal. Note This document refers to the Microsoft Foundry (new) portal. This article covers recovery from Azure platform incidents that take an entire region or a regional dependency offline. Examples of these incidents include prolonged regional outages or loss of a stateful dependency. The recommended approach to design for recovery in your workload is a warm standby, failover, and failback plan for mass outages combined with per-service recovery capabilities for localized outages. Important This article is part of a three-part series. To understand platform limitations, prevention controls, and required baseline configuration, see the overview guide. For prerequisites and context, see Foundry Agent Service disaster recovery . This article explains why some losses are unrecoverable and why recovery often means reconstruction rather than restoration. For recommendations on recovering from human-caused or automation-caused deletions and localized data loss, see Resource and data loss recovery . Prerequisites Before you implement the disaster recovery procedures in this article, ensure you have: An Azure subscription with an active Microsoft Foundry account using Agent Service in Standard deployment mode . One of the following Azure RBAC roles at the subscription or resource group scope: Contributor or Owner for creating and managing Foundry accounts, projects, and dependencies Azure --- ## Disaster recovery from resource and data loss ### Disaster recovery from resource and data loss URL: https://learn.microsoft.com/en-us/azure/ai-foundry/how-to/agent-service-operator-disaster-recovery?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Foundry Agent Service resource and data loss recovery Feedback Summarize this article for me In this article Note This document refers to the Microsoft Foundry (classic) portal. 🔄 Switch to the Microsoft Foundry (new) documentation if you're using the new portal. Note This document refers to the Microsoft Foundry (new) portal. This article describes how to recover from human or automation errors that cause Azure resource or data loss for Foundry Agent Service projects that use the Standard deployment mode . Incidents include accidental deletion of Microsoft Foundry accounts or projects, deletion of agents or threads, and loss or corruption of state in Azure Cosmos DB, Azure AI Search, or Azure Storage that supports the capability host. Important This article is part of a three-part series. To understand platform limitations, prevention controls, and baseline configuration, see the Agent Service disaster recovery overview . That article explains why some losses are unrecoverable and why recovery often means reconstruction rather than restoration. To learn how to architect your solution for high availability and resiliency to prevent these scenarios, see High availability and resiliency . If you're looking for recommendations on how to recover from platform or regional outages, see Platform outage recovery for warm standby, regional failover, and failback. Prerequisites Azure subscription with access to the affected Foundry account Familiarity with the Agent Service disaster recovery overview Infrastructure as code (IaC) assets for your Foundry projects, capability hosts, and dependencies Required RBAC roles: Owner or Contributor --- ## Plan and manage costs ### Plan and manage costs URL: https://learn.microsoft.com/en-us/azure/ai-foundry/concepts/manage-costs?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Plan and manage costs for Microsoft Foundry Feedback Summarize this article for me In this article Note This document refers to the Microsoft Foundry (classic) portal. 🔄 Switch to the Microsoft Foundry (new) documentation if you're using the new portal. Note This document refers to the Microsoft Foundry (new) portal. This article shows you how to estimate expenses before deployment, track spending in real time, and set up alerts to avoid budget surprises. Prerequisites Before you begin, ensure you have: Azure subscription: An active Azure subscription with the resources you want to monitor. Role-based access control (RBAC): One or both of the following roles at the subscription or resource group scope: Cost Management Reader – View costs and usage data. AI User – View Foundry resource data and costs. Supported Azure account type: One of the supported account types for Cost Management . If you need to grant these roles to team members, see Assign access to Cost Management data and Foundry RBAC roles . Note Foundry doesn't have a dedicated page in the Azure pricing calculator because Foundry is composed of several optional Azure services. This article shows how to use the calculator to estimate costs for these services. Estimate costs before using Foundry Use the Azure pricing calculator to estimate costs before you add Foundry resources. Go to the Azure pricing calculator . Search for and select a product, such as Azure Speech in Foundry or Azure Language in Foundry. Select additional products to estimate costs for multiple services. For example, add Azure AI Search to include potential search costs. As you add resources to your --- ## Security baseline ### Security baseline URL: https://learn.microsoft.com/security/benchmark/azure/baselines/azure-ai-foundry-security-baseline?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Azure security baseline for Azure AI Foundry Feedback Summarize this article for me In this article This security baseline applies guidance from the Microsoft cloud security benchmark version 1.0 to Azure AI Foundry. The Microsoft cloud security benchmark provides recommendations on how you can secure your cloud solutions on Azure. The content is grouped by the security controls defined by the Microsoft cloud security benchmark and the related guidance applicable to Azure AI Foundry. You can monitor this security baseline and its recommendations using Microsoft Defender for Cloud. Azure Policy definitions will be listed in the Regulatory Compliance section of the Microsoft Defender for Cloud portal page. When a feature has relevant Azure Policy Definitions, they are listed in this baseline to help you measure compliance with the Microsoft cloud security benchmark controls and recommendations. Some recommendations may require a paid Microsoft Defender plan to enable certain security scenarios. Note Features not applicable to Azure AI Foundry have been excluded.**. Security profile The security profile summarizes high-impact behaviors of Azure AI Foundry, which may result in increased security considerations. Service Behavior Attribute Value Product Category AI+ML Customer can access HOST / OS Full Access Service can be deployed into customer's virtual network True Stores customer content at rest True Network security For more information, see the Microsoft cloud security benchmark: Network security . NS-1: Establish network segmentation boundaries Features Virtual Network Integration Description : Service supports deployment into --- ## Service architecture ### Service architecture URL: https://learn.microsoft.com/en-us/azure/ai-foundry/concepts/architecture?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Microsoft Foundry architecture Feedback Summarize this article for me In this article Note This document refers to the Microsoft Foundry (classic) portal. 🔄 Switch to the Microsoft Foundry (new) documentation if you're using the new portal. Note This document refers to the Microsoft Foundry (new) portal. Microsoft Foundry provides a comprehensive set of tools to support development teams in building, customizing, evaluating, and operating AI Agents and its composing models and tools. This article provides IT operations and security teams with details on the Foundry resource and underlying Azure service architecture, its components, and its relation with other Azure resource types. Use this information to guide how to customize your Foundry deployment to your organization's requirements. For more information on how to roll out Foundry in your organization, see Foundry Rollout . Azure AI resource types and providers Within the Azure AI product family, you can use these Azure resource providers that support user needs at different layers in the stack. Resource provider Purpose Supports resource type kinds Microsoft.CognitiveServices Supports Agentic and GenAI application development composing and customizing prebuilt models. Foundry; Azure OpenAI service; Azure Speech; Azure Vision Microsoft.Search Support knowledge retrieval over your data Azure AI Search Microsoft.MachineLearningServices Train, deploy, and operate custom and open source machine learning models Azure AI Hub (and its projects); Azure Machine Learning Workspace Resource provider Purpose Supports resource type kinds Microsoft.CognitiveServices Supports Agentic and Ge --- ## Data, privacy, and security for Foundry Models sold directly by Azure ### Data, privacy, and security for Foundry Models sold directly by Azure URL: https://learn.microsoft.com/en-us/azure/ai-foundry/responsible-ai/openai/data-privacy?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Data, privacy, and security for Azure Direct Models in Microsoft Foundry Feedback Summarize this article for me In this article Important Non-English translations are provided for convenience only. Please consult the EN-US version of this document for the definitive version. This article provides details regarding how data provided by you to Azure Direct Models in Microsoft Foundry are processed, used, and stored. Azure Direct Model means an AI model designated and deployed as an “Azure Direct Model” in Foundry, and includes Azure OpenAI models. Azure Direct Models store and process data to provide the service and to monitor for uses that violate the applicable product terms. Please also see Microsoft Products and Services Data Protection Addendum , which governs data processing by Azure Direct Models. Foundry is an Azure service; learn more about applicable Azure compliance offerings. Important Your prompts (inputs) and completions (outputs), your embeddings, and your training data: are NOT available to other customers. are NOT available to OpenAI or other Azure Direct Model providers. are NOT used by Azure Direct Model providers to improve their models or services. are NOT used to train any generative AI foundation models without your permission or instruction. Customer Data, Prompts, and Completions are NOT used to improve Microsoft or third-party products or services without your explicit permission or instruction. Your fine-tuned Azure Direct Models are available exclusively for your use. Foundry is an Azure service; Microsoft hosts the Azure Direct Models in Microsoft's Azure environment and Azure Direct Models do NOT inte --- ## Data, privacy, and security for Claude models in Microsoft Foundry ### Data, privacy, and security for Claude models in Microsoft Foundry URL: https://learn.microsoft.com/en-us/azure/ai-foundry/responsible-ai/claude-models/data-privacy?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Data, privacy, and security for Claude models in Microsoft Foundry (preview) Feedback Summarize this article for me In this article Note This document refers to the Microsoft Foundry (classic) portal. 🔄 Switch to the Microsoft Foundry (new) documentation if you're using the new portal. Note This document refers to the Microsoft Foundry (new) portal. This article describes how the data that you provide is processed, used, and stored when you use Anthropic Claude models from the Microsoft Marketplace in Microsoft Foundry. Claude in Foundry processes data differently from models sold directly by Azure. When you transact for Claude in Foundry, you will agree to Anthropic's terms of use and Anthropic (not Microsoft) is the processor of the data. As detailed in the following sections, to the extent Microsoft plays a role in processing data in connection with Claude on Foundry, it will do so pursuant to its own terms. What data is processed by Anthropic for Anthropic Claude models in Microsoft Foundry? When you deploy an Anthropic Claude model from the model catalog in Foundry with pay-per-token offer for inferencing, an API is provisioned. The API gives you access to the model that Anthropic service hosts and manages. To learn more about Anthropic Claude models in Foundry, see Models from Partners and Community . The model processes your input prompts and generates outputs based on its functionality, as described in the model details. Your use of the model (along with Anthropic's accountability for the model and its outputs) is subject to the terms of use for the model provided by Anthropic. Anthropic acts as the data processor for pr --- ## What's new in Microsoft Foundry ### What's new in Microsoft Foundry URL: https://learn.microsoft.com/en-us/azure/ai-foundry/whats-new-foundry?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . What's new in Microsoft Foundry documentation? Feedback Summarize this article for me In this article Welcome! This article highlights key changes and updates in Microsoft Foundry documentation for December 2025. This month marks a significant update to our documentation structure. With the introduction of the new Microsoft Foundry portal, we now maintain two corresponding versions of the documentation to support each portal experience. This dual-version approach ensures that users can access accurate, version-specific guidance tailored to their portal environment. New articles Available in Foundry (new) only: Developer journey: Idea to prototype Publish agents in Microsoft Foundry Agent memory concepts Build your own MCP server Manage agent identities with Microsoft Entra ID Optimization model upgrade Cluster analysis Optimization dashboard Human evaluation Azure Language tools and agents Azure Language CLU Multi-turn conversations Available in both Foundry (new) and Foundry (classic): Install CLI SDK SDK overview High availability and resiliency Agent service disaster recovery Agent service operator disaster recovery Agent service platform disaster recovery Integrate with other apps Create a custom photo avatar Customize voice live Bring your own model Use the LLM-speech API Priority processing for Foundry Models Classification in Content Understanding Studio Foundry playgrounds Use Claude in Foundry Models Monitor and manage agents with Foundry control plane Updated articles All articles were updated in some way this month: Articles that apply to the new version were updated to add version-specific information. Articles that --- ## Status dashboard ### Status dashboard URL: https://learn.microsoft.com/en-us/azure/ai-foundry/foundry-status-dashboard-documentation?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Microsoft Foundry status dashboard Feedback Summarize this article for me In this article Note This document refers to the Microsoft Foundry (classic) portal. 🔄 Switch to the Microsoft Foundry (new) documentation if you're using the new portal. Note This document refers to the Microsoft Foundry (new) portal. The Microsoft Foundry Status Dashboard provides visibility into the health and availability of key Foundry services. It helps customers monitor service status, stay informed about ongoing incidents, and plan around scheduled maintenance windows. This dashboard is currently in preview , and it might not reflect all components or issues. Prerequisites A web browser. Check service status Open the Microsoft Foundry Status Dashboard . Review the overall status at the top of the page. Select a component to view details and recent status changes. Select Incidents to review incident history. Select Subscribe to updates to get notified about updates. Key features Live status indicators for core Foundry services. Incident reports with timelines, resolutions, and root cause summaries. Historical uptime to help assess service reliability over time. Frequently asked questions Q: Is this data real-time? The dashboard updates as incident and maintenance status changes are posted. Q: What does it mean that this dashboard is in “Preview”? During preview, service coverage is expanding and the experience is still being refined. Some services might not appear, and update timing might vary. Q: Can I subscribe to updates? Yes. Select Subscribe to updates in the dashboard. Q: Does the dashboard cover all regions and environments? The dashboard is --- ## Use Microsoft Foundry with a screen reader ### Use Microsoft Foundry with a screen reader URL: https://learn.microsoft.com/en-us/azure/ai-foundry/tutorials/screen-reader?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Use a screen reader with Microsoft Foundry Feedback Summarize this article for me In this article Note This document refers to the Microsoft Foundry (classic) portal. 🔄 Switch to the Microsoft Foundry (new) documentation if you're using the new portal. Note This document refers to the Microsoft Foundry (new) portal. This article is for people who use screen readers such as Microsoft's Narrator , JAWS, NVDA, or Apple's VoiceOver. In this quickstart, you learn about the basic structure of Microsoft Foundry and how to navigate around efficiently. Get oriented in Foundry portal Most Microsoft Foundry (classic) pages have the following landmark structure: Banner (has Foundry app title, settings, and profile information) Sometimes has a breadcrumb navigation element Navigation - Three different states: Outside a project - there's no left navigation until you're in a project. The page is divided into sections. Once you have projects, the top section is a list of recent projects. In a project - the left navigation is the same for all parts of a project until you move to the Management center . The Management center left navigation has two sections. In a Foundry project, the sections are for the resource the project is in, then a section for the project itself. In a hub-based project, the sections are for the hub the project is in, then a section for the project itself. Most Microsoft Foundry (new) pages have the following landmark structure: Banner has Foundry application title Project selector Search Ask AI toggle Main section navigation: Home, Discover, Build, Operate, Docs Settings Profile information Left pane has navigation for the --- ## Known issues ### Known issues URL: https://learn.microsoft.com/en-us/azure/ai-foundry/reference/foundry-known-issues?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Microsoft Foundry known issues Feedback Summarize this article for me In this article Note This document refers to the Microsoft Foundry (classic) portal. 🔄 Switch to the Microsoft Foundry (new) documentation if you're using the new portal. Note This document refers to the Microsoft Foundry (new) portal. Microsoft Foundry is updated regularly and the product team is continually improving and enhancing its features and capabilities. This article lists known issues related to Foundry and provides steps to resolve them. Before submitting a support request, review the following list to see if your problem is already being addressed and to find a possible solution. For more information about service-level outages, check the Azure status page . To set up outage notifications and alerts, visit the Azure Service Health Portal . General Foundry known issues Issue ID Category Title Description Workaround Issues publish date 0001 Foundry Portal Network isolation in new Foundry The new Foundry portal experience doesn't support end-to-end network isolation. When you configure network isolation (disable public network access, enable private endpoints, and use virtual network-injected Agents), you must use the classic Foundry portal experience, the SDK, or CLI to securely access your Foundry projects. December 5, 2025 0002 Foundry Portal Multiple projects per Foundry resource The new Foundry portal experience doesn't support multiple projects per Foundry resource. Each Foundry resource supports only one default project. None December 5, 2025 AI Speech The following tables describe the current known issues for the Speech services, including Spe --- ## Feature availability by region ### Feature availability by region URL: https://learn.microsoft.com/en-us/azure/ai-foundry/reference/region-support?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Microsoft Foundry feature availability across cloud regions Feedback Summarize this article for me In this article Note This document refers to the Microsoft Foundry (classic) portal. 🔄 Switch to the Microsoft Foundry (new) documentation if you're using the new portal. Note This document refers to the Microsoft Foundry (new) portal. Microsoft Foundry brings together various Azure AI capabilities that were previously only available as standalone Azure services. While Microsoft strives to make all features available in all regions where Microsoft Foundry is supported at the same time, feature availability might vary by region. In this article, you learn what Foundry features are available across cloud regions. Foundry projects Foundry is currently available in the following Azure regions. Microsoft Foundry brings together various Azure AI capabilities that were previously only available as standalone Azure services. While Microsoft strives to make all features available in all regions where Foundry is supported at the same time, feature availability might vary by region. In this article, you learn what Foundry features are available across cloud regions. Foundry projects Foundry is currently available in the following Azure regions. You can create either a Foundry project or hub-based project in Foundry in these regions. Australia East Brazil South Canada Central Canada East Central India East Asia East US East US 2 France Central Germany West Central Italy North Japan East Korea Central North Central US North Europe Norway East Qatar Central South Africa North South Central US South India Southeast Asia Spain Central Sweden Centr --- ## Model lifecycle and retirement ### Model lifecycle and retirement URL: https://learn.microsoft.com/en-us/azure/ai-foundry/concepts/model-lifecycle-retirement?view=foundry Table of contents Exit editor mode Ask Learn Ask Learn Focus mode Table of contents Read in English Add Add to plan Edit Share via Facebook x.com LinkedIn Email Print Note Access to this page requires authorization. You can try signing in or changing directories . Access to this page requires authorization. You can try changing directories . Model deprecation and retirement for Microsoft Foundry Models Feedback Summarize this article for me In this article Note This document refers to the Microsoft Foundry (classic) portal. 🔄 Switch to the Microsoft Foundry (new) documentation if you're using the new portal. Note This document refers to the Microsoft Foundry (new) portal. Microsoft Foundry Models are continually refreshed with newer and more capable models. As part of this process, model providers might deprecate and retire their older models, and you might need to update your applications to use a newer model. This document communicates information about the model lifecycle and deprecation timelines and explains how you're informed of model lifecycle stages. This article covers general deprecation and retirement information for Foundry Models. For details specific to Azure OpenAI in Foundry Models, see Azure OpenAI in Foundry Models model deprecations and retirements . Model lifecycle stages Models in the model catalog belong to one of these stages: Preview Generally available Legacy Deprecated Retired Preview Models labeled Preview are experimental in nature. A model's weights, runtime, and API schema can change while the model is in preview. Models in preview aren't guaranteed to become generally available. Models in preview have a Preview label next to their name in the model catalog. Generally available (GA) This stage is the default model stage. Models that don't include a lifecycle label next to their name are GA and suitable for use in production environments. In this stage, model weights and APIs are fixed. However, model containers or runtimes with vulnera ---