Bootstrapping your Microsoft Foundry environment for Agents

bootstrap-agent-environment.md

Bootstrapping your Microsoft Foundry environment for Agents

These are the simplest instructions for creating a project and running an AI Agent.

Phase 1: Create the Project (Foundry + Azure resources needed)

All paths start here. This sets up the cloud resources.

1. Click the "Deploy to Azure" button above
2. Sign in with your Azure credentials
3. Configure:
   • Region: Select East US 2 (or your preferred region)
   • Environment Name: Enter vibe-agent-offsite (or your preferred name)
   • principalID: enter the result of the following CLI command so your User identity gets RBAC roles/permissions needed (portal limitation)

   az ad signed-in-user show --query id -o tsv
   

4. Deploy: Click Review + Create, then Create
   • Note: Provisioning takes 2-3 minutes

What gets created:

• Resource Group: rg-<your-environment-name> (e.g., rg-vibe-agent-offsite)
• Model Deployment: Named chat using GPT-4.1-mini
• Azure AI Foundry Project: Named <your-environment-name><unique-id> (e.g. vibe-agent-12345)
• Go to Foundry Portal and look for your project starting with vibe (view All Resources if needed)

Alternative: Each sample repository below contains Bicep infrastructure you can deploy using azd provision or azd up.

---

Phase 2: Build the Agent

Choose your path - with code focus or the portal

Path A: The Code (Developer Experience using Agent Framework)

Best for: Understanding SDK implementation and running locally.

1. Gather Your Credentials

Before running code, you need two values from your Azure/Foundry settings:

• Azure OpenAI Endpoint: (e.g., https://<your-resource>.openai.azure.com/)
• Deployment Name: (The name of the model you deployed, e.g., chat)

The endpoint we want for Agent Framework is the Azure OpenAI one.
It's the foundry project resource, overview (old portal): Click on the Azure OpenAI tab. Get this endpoint. It should not have a route.

E.g. a valid one looks like:

https://agent-ai-servicesou6taoycpkkwc.openai.azure.com/

2. Configure Environment Variables

Copy the block matching your OS into your terminal to set the variables for the current session.

Windows PowerShell

$env:AZURE_OPENAI_ENDPOINT = "https://agent-ai-NNNNN.openai.azure.com/"
$env:AZURE_OPENAI_DEPLOYMENT_NAME = "chat"

macOS / Linux

export AZURE_OPENAI_ENDPOINT="https://agent-ai-NNNNN.openai.azure.com/"
export AZURE_OPENAI_DEPLOYMENT_NAME="chat"

Note: Replace https://agent-ai-NNNNN... with your actual endpoint URL.

3. Run the Agent

Option 1: Python

• Repo: https://github.com/paulyuk/simple-agent-af-python

1. Clone the repository.
2. Open your terminal to the repo folder.
3. Run the environment variable commands from Step 2.
4. Follow the README.md to install dependencies in a venv and run the agent script (python main.py).

Option 2: C# / .NET

• Repo: https://github.com/paulyuk/simple-agent-af

1. Clone the repository.
2. Open your terminal to the repo folder.
3. Run the environment variable commands from Step 2.
4. Follow the README.md to run the app (dotnet run).

---

Path B: The Portal UX (No-Code using Agent Service)

Best for: Rapid prototyping, testing prompts, and visual verification.

1. Enter Agent Builder: In your new project, look at the left-hand navigation menu and click Agents.
2. Create: Click + Create Agent.
3. Setup:
   • Name: Give your agent a name.
   • Model: Select gpt-4.1-mini from the dropdown. If no model is listed, click "Connect & deploy" to add one named gpt-4.1-mini.
   • Instructions: Type a simple system prompt (e.g., "You are a helpful project management assistant.").
4. Test: Use the Playground chat window on the right to type a message and verify the response.