Openai Endpoints

What it demonstrates

Agent Framework speaks the OpenAI wire protocols — the Chat Completions API and the newer, stateful Responses API — on both sides. On the HOSTING side you expose your agent behind /v1/chat/completions and /v1/responses so any OpenAI SDK client can call it. On the CONSUMING side you point the framework at any OpenAI-compatible endpoint (Ollama, vLLM, LM Studio) via base_url. This lesson builds a Foundry-backed “backend brain” agent and drives it with the exact JSON body an OpenAI client would POST.

One real excerpt

# The exact JSON an OpenAI SDK client would POST to /pirate/v1/chat/completions.
chat_completions_request = {
    "model": "pirate",
    "stream": False,
    "messages": [{"role": "user", "content": "Hey mate, how be the weather?"}],
}
user_text = chat_completions_request["messages"][-1]["content"]
response = await agent.run(user_text)          # same agent, whatever protocol fronts it
async for chunk in agent.run_stream("Tell me a very short sea shanty."):
    if chunk.text:
        print(chunk.text, end="", flush=True)   # a Responses endpoint relays chunks like these

The gotcha

The Python pivot only ships the CONSUMING side: OpenAIChatCompletionClient(base_url=..., api_key=..., model=...).as_agent(...) for Chat Completions, and OpenAIChatClient(base_url=...) for the stateful Responses API. HOSTING an agent as OpenAI endpoints is a .NET-only library (Microsoft.Agents.AI.Hosting.OpenAI) — there is no Python host API, so the hosting side here is illustrative, not a runnable server. The Responses API is the recommended default: it is stateful and adds a conversation parameter plus streaming event types.

Azure / MAF mapping

The backend agent is a plain Agent over FoundryChatClient (project_endpoint + model + AzureCliCredential). Because tools, options, and streaming are client-agnostic, the same Foundry agent works whichever protocol fronts it; to consume another server set OPENAI_BASE_URL (or pass base_url) and the generic OpenAI client picks it up.

Run it

uv run tutorial/04-hosting/04_openai_endpoints.py — needs Foundry creds (az login). Worked if it prints a simulated OpenAI request, a pirate reply, then a streamed reply.

Next: Foundry Hosted Agent