RAPP — Install

🧠

The Brainstem

Local agent server — the survival basics

Start here ▶

The biological brainstem keeps you alive — breathing, heartbeat, reflexes. Your AI brainstem does the same: it runs the core agent loop, calls tools, and responds. Everything else builds on top of this.

1. Authenticate with GitHub

Your GitHub Copilot subscription IS the AI engine. No OpenAI keys needed.

gh auth login

2. Start the brainstem

The server starts on port 7071 with a chat UI, health endpoint, and agent auto-discovery.

brainstem

3. Talk to it

Open localhost:7071 in your browser, or hit the API directly.

curl -s localhost:7071/health | python3 -m json.tool

4. Make it yours

Write a soul.md to define its personality. Drop *_agent.py files into the agents folder to give it skills. Point SOUL_PATH and AGENTS_PATH in .env to your private files.

☁️

The Spinal Cord

Deploy RAPP to Azure — give it a cloud body

Next step ▶

The spinal cord connects your brainstem to the rest of your body. Deploy RAPP to Azure — your agent is now always-on, reachable from anywhere, with persistent storage and monitoring.

1. Deploy Azure resources

One script creates everything: Function App, Azure OpenAI, Storage Account, Application Insights. All using Entra ID auth — no API keys.

curl -fsSL https://raw.githubusercontent.com/microsoft/aibast-agents-library/main/deploy.sh | bash

Or click to deploy directly from the Azure Portal:

2. Deploy your code

Push the RAPP function code to your new Azure Function App. The app name comes from the deploy output.

func azure functionapp publish YOUR_APP_NAME --build remote

3. Test it

Hit your deployed endpoint to verify the agent is live in the cloud.

curl -s https://YOUR_APP_NAME.azurewebsites.net/api/businessinsightbot_function -X POST -H "Content-Type: application/json" -d '{"user_input":"hello"}' | python3 -m json.tool

What you learn: ARM templates, Azure Functions (Python), managed identity, RBAC, Azure OpenAI.

🤖

The Nervous System

Reach into Teams & M365 — two paths

Advanced ▶

Your nervous system extends your brain's reach everywhere. Choose your deployment path:

Option A: Azure Function → Copilot Studio

Wire your deployed Azure Function to Copilot Studio. Import the Power Platform solution, point it at your endpoint, publish. Same RAPP agents, now in Teams and M365 Copilot.

Requires: Tier 2 (Spinal Cord) completed

curl -fsSL -o MSFTAIBASMultiAgentCopilot.zip https://raw.githubusercontent.com/microsoft/aibast-agents-library/main/MSFTAIBASMultiAgentCopilot_1_0_0_5.zip

Import at make.powerapps.com → Solutions → Import. Update the HTTP action URL to your Azure Function. Publish.

Option B: Native Copilot Studio (No Azure needed)

Skip the Azure Function entirely. Use Skills for Copilot Studio — the CAT team's open-source plugin for Claude Code & GitHub Copilot CLI. Author agents as YAML from your terminal, up to 20x faster.

Requires: Claude Code or GitHub Copilot CLI + VS Code Copilot Studio Extension

/plugin marketplace add microsoft/skills-for-copilot-studio

@copilot-studio:author — create topics, actions, knowledge sources
@copilot-studio:test — run test suites against published agents
@copilot-studio:troubleshoot — debug routing and validation issues

What you learn: Copilot Studio, declarative agents, Power Platform solutions, Teams integration, YAML-as-code agent authoring.

Current Limitations

⚠️ IMPORTANT: This is an experimental project managed by a v-team from the Artificial Intelligence Business Applications Specialist Team (AIBAST), not an officially supported Microsoft product.

This project is still in beta. Even if we're optimizing it for best-practice adherence, by using it you might sometimes experience unwanted patterns, errors, or simply bad architectures. By filing a GitHub issue you will help us improve these tools.

Agent templates are starting points, not turnkey solutions. Each template provides the structure, system prompts, and logic scaffold for a specific business function. Customize before production use.
AI-generated output may contain errors or unsupported patterns. Human review remains important.
The project is evolving quickly. We're actively improving it based on feedback.

This project has adopted the Microsoft Open Source Code of Conduct. This project may contain trademarks or logos for projects, products, or services. Authorized use of Microsoft trademarks or logos is subject to and must follow Microsoft's Trademark & Brand Guidelines. MIT License — Copyright (c) 2026 Microsoft.