Prerequisites & Environment Setup

Work through this checklist with your coach before Day 1. The RVAS runs on the customer's own environment — your Azure subscription(s) plus a Microsoft Fabric capacity. Getting access and quota sorted up front is the single biggest factor in a smooth event.

⏱️ Lead time: Azure AI Foundry model quota and Fabric tenant enablement can take from minutes to a few business days depending on your tenant's governance. Start at least 3–5 business days early.

1. Azure subscription & permissions#

You need at least one Azure subscription where your team can create resources.

🏢 Multiple subscriptions? That's fine and realistic. A common pattern: the agent workload and landing zone live in one subscription, while cost exports are scoped at the billing account / management group. Note your subscription IDs in Challenge 0.

2. Azure AI Foundry / Azure OpenAI quota#

The agent needs a chat model. Confirm before the event:

• Access to Azure AI Foundry (Azure OpenAI) in your tenant. If your subscription has never used it, request access / confirm it is enabled.
• Model quota for gpt-4o or gpt-4o-mini (the agent defaults to gpt-4o; gpt-4o-mini is cheaper and recommended for the RVAS) in your chosen region.
• A region that offers both the model and Microsoft Fabric. Good defaults: East US 2, West US 3, Sweden Central. Verify model availability for your region in the Foundry portal.

🔎 Check quota in the Azure AI Foundry portal → Quotas, or with the azure-quotas tooling. 30K–50K TPM of gpt-4o-mini is plenty for a team.

3. Microsoft Fabric capacity#

The Control Tower runs on Microsoft Fabric. You have two options:

Option A — Fabric free trial (recommended for the RVAS)#

• Provides free access for 60 days with a trial capacity (provisioned as an F64 or F4; eligibility for F64 may vary by tenant). It includes a Power BI individual trial.
• The person who starts the trial becomes the Capacity administrator and can share it by assigning workspaces — one trial capacity can back every team's workspace.
• Start it from app.fabric.microsoft.com → Account manager → Start trial. See Try Microsoft Fabric for free.

Option B — Paid Fabric capacity#

• Any F2 or higher capacity works. F2 handles the RVAS data volumes comfortably; the reference Q&A notes F64+ for production scale.
• Create it in the Azure portal (Microsoft.Fabric/capacities) and pause it outside working hours to control cost.

Tenant settings (admin, one-time)#

A Fabric / Power BI tenant administrator must ensure:

• Microsoft Fabric is enabled for the tenant (or for the relevant security group). See Enable Microsoft Fabric.
• Users can create workspaces.
• For Challenge 3, the Mirroring and service principal / managed identity workspace settings are permitted (used by the setup script and Cosmos DB Mirroring).

⚠️ On trial capacity, Private Link is disabled and a few enterprise features aren't supported — none of which block this RVAS. Mirroring, shortcuts, Spark, Direct Lake, and Data Activator all work.

4. Power BI#

• Each report author/consumer needs a Power BI Pro license, Premium Per User, or the Power BI individual trial that accompanies the Fabric trial.
• Power BI Desktop (Windows) is optional but handy for editing the semantic model offline. Authoring in the Fabric/Power BI service is sufficient for all challenges.

5. Identity & networking#

• A Microsoft Entra ID account in the same tenant as the Azure subscription and Fabric capacity. Using the same tenant end-to-end avoids cross-tenant friction with managed identity and Mirroring.
• Outbound HTTPS to Azure, Fabric (*.fabric.microsoft.com), and package registries (PyPI, npm, Docker Hub / MCR). If the customer network restricts egress, flag it to your coach early.

6. Local tooling#

Install on each team member's machine (or use a shared Azure Cloud Shell / dev box):

Verify:

Copyaz version
azd version
python --version
node --version
docker --version
git --version

7. Cost expectations & hygiene#

This RVAS is intentionally low-cost, but clean up when you're done.

• Fabric — free on trial; pause paid capacity outside hours. OneLake storage is a few cents/GB.
• Agent workload — serverless Cosmos DB, Container Apps scale-to-low, and gpt-4o-mini keep spend to a few dollars/day. APIM Developer tier (used here) is the main fixed cost.
• Landing zone — ADLS Gen2 + Log Analytics are inexpensive at RVAS volumes.
• Tear down with azd down --purge in each workload directory when the event ends (Challenge 0 cleanup notes), and delete the Fabric workspace / end the trial.

8. Get the code#

Copygit clone https://github.com/microsoft/frontier-fabric-agentops-rvas.git
cd frontier-fabric-agentops-rvas

✅ Pre-flight checklist#

Bring this to the kickoff. You're ready when every box is ticked:

• Each member can az login to the target subscription
• Roles confirmed: Contributor (+ User Access Administrator), Cost Management Reader
• Resource providers registered: Microsoft.App, Microsoft.DocumentDB, Microsoft.ApiManagement, Microsoft.CognitiveServices, Microsoft.OperationalInsights, Microsoft.Storage
• Azure OpenAI / Foundry access confirmed with quota for gpt-4o or gpt-4o-mini in your region
• Region chosen that supports both the model and Fabric (e.g., East US 2)
• Fabric trial started (or paid F2+ capacity created) and you know your capacity ID / name
• Fabric tenant settings allow workspace creation + Mirroring
• Everyone has a Power BI license (Pro / PPU / trial)
• Local tooling installed and verified (az, azd, python, docker, git)
• Repo cloned

Stuck on any item? That's exactly what Challenge 0 and your coach are for.