A text chat is one of the smallest useful Foundry applications: send a user message to a deployed model and stream the answer back as it is generated. The same pattern works with current and future chat-capable model deployments because the script reads the deployment name from the environment.

What you need

1. A Foundry project with a chat-capable model deployment. The value sent as MODEL must be the exact deployment name shown in the project's Deployed models page; it may differ from the underlying model name.
2. Azure CLI login (az login) with permission to use the project — or another identity supported by DefaultAzureCredential.
3. The Python packages listed in code/requirements.txt:

pip install -r code/requirements.txt

The core idea

The sample uses four steps:

1. Load configuration from .env, including the Foundry project endpoint and deployment name.
2. Authenticate with Entra ID using DefaultAzureCredential; no API key is stored in the sample.
3. Create the project OpenAI client with AIProjectClient.get_openai_client().
4. Stream a chat completion with chat.completions.create.

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

with (
    DefaultAzureCredential() as credential,
    AIProjectClient(endpoint=PROJECT_ENDPOINT, credential=credential) as project_client,
    project_client.get_openai_client() as openai_client,
):
    stream = openai_client.chat.completions.create(
        model=MODEL,
        messages=[{"role": "user", "content": PROMPT}],
        stream=True,
    )

Streaming makes the application feel responsive: tokens are printed as they arrive instead of waiting for the complete answer.

Configure and run it

Copy code/.env.example to code/.env, then set PROJECT_ENDPOINT and MODEL. Set MODEL to the exact deployment name shown in the Foundry project's Deployed models page. For example, if you created a deployment named my-gpt-4o-mini, use MODEL=my-gpt-4o-mini; do not assume the base model ID gpt-4o-mini is the deployment name.

PROJECT_ENDPOINT should be the project endpoint shown in Foundry, for example:

https://<resource-name>.services.ai.azure.com/api/projects/<project-name>

Set PROMPT for a single response. If PROMPT is omitted, the full script starts an interactive terminal chat and keeps the conversation history between turns. SYSTEM_PROMPT is optional.

Run the sample from the article's code directory:

python text_chat.py

Example output:

Assistant (gpt-6-luna): Microsoft Foundry is Microsoft’s platform for building, testing, deploying, and managing AI applications and agents. It brings together access to AI models, development tools, data connections, evaluation and safety features, and deployment options, so teams can create AI solutions and operate them with enterprise security and governance on Microsoft’s cloud.

The attached text_chat.py is the complete runnable sample and is available in the code modal for viewing or download.

Microsoft Learn resources

• Azure AI Projects client library for Python — authenticate with AIProjectClient and obtain an OpenAI-compatible client.
• Use reasoning models with Microsoft Foundry Models — configure the OpenAI client and send chat completions through a Foundry project endpoint.
• Chat completions with Azure OpenAI — understand messages, deployments, and streaming responses.