Simple Agent QuickStart (Python Copilot SDK)

A simple AI agent built with the GitHub Copilot SDK, running as an Azure Function.

Prerequisites

• Python 3.11+ (via uv)
• Azure Functions Core Tools
• Azure Developer CLI (azd) (only needed for deploying Microsoft Foundry resources)
• Access to an AI model via one of:
  • GitHub Copilot subscription — models are available automatically
  • Bring Your Own Key (BYOK) — use an API key from Microsoft Foundry (see BYOK docs)

Deploy Microsoft Foundry Resources (if needed)

If you're using BYOK and don't already have a Microsoft Foundry project with a model deployed:

azd auth login
azd up

This provisions all resources and configures local development automatically.

What Gets Deployed

• Microsoft Foundry project with GPT-5-mini model
• Azure Functions app (Python, Flex Consumption plan)
• Storage, monitoring, and all necessary RBAC role assignments
• Optional: Search for vector store (disabled by default)
• Optional: Cosmos DB for agent thread storage (disabled by default)

Quickstart

1. Clone the repository

2. Install dependencies:

   uv venv
   source .venv/bin/activate  # macOS/Linux
   # .venv\Scripts\activate   # Windows
   uv pip install -r requirements.txt

3. Run the function locally:

   func start

4. Test the agent (in a new terminal):

   # Interactive chat client
   uv run chat.py
   
   # Or use curl directly
   curl -X POST http://localhost:7071/api/ask -d "what are the laws"

   Set AGENT_URL to point to a deployed instance:

   AGENT_URL=https://<your-function-app>.azurewebsites.net uv run chat.py

Source Code

The agent logic is in function_app.py. It creates a CopilotClient, configures a session with a system message (Asimov's Three Laws of Robotics), and exposes an HTTP endpoint (/api/ask) that accepts a prompt and returns the agent's response.

chat.py is a lightweight console client that POSTs messages to the function in a loop, giving you an interactive chat experience. It defaults to http://localhost:7071 but can be pointed at a deployed instance via the AGENT_URL environment variable.

Using Microsoft Foundry (BYOK)

By default the agent uses GitHub Copilot's models. To use your own model from Microsoft Foundry instead, set these environment variables:

export AZURE_OPENAI_ENDPOINT="https://<your-ai-services>.openai.azure.com/"
export AZURE_OPENAI_API_KEY="<your-api-key>"
export AZURE_OPENAI_MODEL="gpt-5-mini"  # optional, defaults to gpt-5-mini

Getting these values:

• If you ran azd up, the endpoint is already in your environment — run azd env get-values | grep AZURE_OPENAI_ENDPOINT
• For the API key, go to Azure Portal → your AI Services resource → Keys and Endpoint → select the Azure OpenAI tab
• Or find both in the Microsoft Foundry portal under your project settings

See the BYOK docs for details.

Learn More

• GitHub Copilot SDK
• Copilot SDK Python docs
• BYOK (Bring Your Own Key)
• Azure Developer CLI