02 First Model Call

Goal

Confirm your Foundry endpoint, authentication, and deployed model all work from Python.

Estimated time

10 minutes.

Official references

• Microsoft Foundry get-started code quickstart
• Microsoft Foundry SDKs and Endpoints
• Azure SDK for Python responses samples

Exercise

Run the example:

python examples/01-model-call/model_call.py

This example is using the Responses API. The script first connects to your Foundry project with AIProjectClient, then asks that project client for an OpenAI-compatible client, and finally sends a responses.create(...) call to your deployed model.

Example file

• Open model_call.py on GitHub

examples/01-model-call/model_call.py

import os

from azure.ai.projects import AIProjectClient
from azure.identity import DefaultAzureCredential
from dotenv import load_dotenv


def get_env(name: str, fallback: str | None = None) -> str:
    value = os.getenv(name) or (os.getenv(fallback) if fallback else None)
    if not value:
        missing = f"{name}"
        if fallback:
            missing = f"{name} or {fallback}"
        raise ValueError(f"Missing required environment variable: {missing}")
    return value


def main() -> None:
    load_dotenv()

    project_endpoint = get_env("AZURE_AI_PROJECT_ENDPOINT", "PROJECT_ENDPOINT")
    model_deployment_name = get_env(
        "AZURE_AI_MODEL_DEPLOYMENT_NAME", "MODEL_DEPLOYMENT_NAME"
    )

    project = AIProjectClient(
        endpoint=project_endpoint,
        credential=DefaultAzureCredential(),
    )
    openai = project.get_openai_client()

    response = openai.responses.create(
        model=model_deployment_name,
        input=(
            "List three common use cases for PCR in life sciences in one sentence each."
        ),
    )

    print("Model response:")
    print(response.output_text)


if __name__ == "__main__":
    main()

What this lab demonstrates

1. Create an AIProjectClient with DefaultAzureCredential.
2. Get the OpenAI-compatible client from the project.
3. Send a responses.create(...) request.
4. Print the generated answer.

What is happening under the hood

• AZURE_AI_PROJECT_ENDPOINT points to your Foundry project endpoint, not directly to a raw model endpoint.
• AIProjectClient handles the Foundry-side connection and authentication using DefaultAzureCredential.
• project.get_openai_client() returns an OpenAI-compatible client that can call model features exposed through your Foundry project.
• responses.create(...) sends your input prompt to the deployed model named in AZURE_AI_MODEL_DEPLOYMENT_NAME.
• response.output_text prints the final text returned by the model.

This is a useful first lab because it proves the full path is working: local Python environment, Azure authentication, Foundry project endpoint, deployed model, and the OpenAI-compatible Responses API surface.

Expected result

You should see a short response describing common PCR use cases in life sciences.

Verification

• The script runs without authentication errors.
• The script returns model output.
• The project endpoint in .env is valid.

Common issues

• DefaultAzureCredential failed to retrieve a token: sign in again with az login.
• 404 Not Found: check your project endpoint format.
• SDK mismatch errors: verify azure-ai-projects>=2.0.0.