Hello Agent

What this lesson demonstrates

This is the Agent primitive stripped to essentials. You wire a FoundryChatClient to a deployed model, hand it an instruction string and a name, and call it with a message. That single loop is what tools, memory, and workflows all build on. The lesson runs the same agent two ways: agent.run(...) collects the whole answer, and stream=True yields chunks as they arrive.

The code

Construction is factored out of main so it can be built and tested on its own:

client = FoundryChatClient(
    project_endpoint=os.environ["FOUNDRY_PROJECT_ENDPOINT"],
    model=os.environ["FOUNDRY_MODEL"],
    credential=AzureCliCredential(),
)
instructions = (
    "You are a terse, knowledgeable travel guide. "
    "Answer in at most two sentences and never use bullet points."
)
return Agent(client=client, name="HelloAgent", instructions=instructions)

Then main runs it two ways off the same object:

result = await agent.run("What is the capital of France?")
async for chunk in agent.run("Tell me a fun fact.", stream=True):
    if chunk.text:
        print(chunk.text, end="", flush=True)

What to notice

• The instructions ARE the agent. That single string shapes persona, tone, and guardrails. Tweak it, re-run, and both answers shift.
• One method, two call styles. agent.run(...) returns the finished answer; the same call with stream=True becomes an async iterator of chunks — guard each with if chunk.text since some carry no text.
• The gotcha: the client reads FOUNDRY_PROJECT_ENDPOINT and FOUNDRY_MODEL from the environment (via load_dotenv()), and auth comes from your az login session through AzureCliCredential. Miss any of those three and construction fails before a single token.

How it maps to Azure AI Foundry

FoundryChatClient binds the agent to your Foundry project endpoint, keyed by the deployed model name. AzureCliCredential reuses the token from az login, so there are no secrets in code. instructions is the system prompt; name is how the agent identifies itself.

Run it

uv run tutorial/01-get-started/01_hello_agent.py

Expected: a non-streaming answer, then the same agent streaming a second reply. The live call needs Foundry creds and az login.

Next: Add Tools