Azure AI Inference Client Library

Warnings

• breaking This library is currently in a beta state (`1.0.0b9`). Breaking changes are expected in future releases until it reaches a stable (1.0.0) version. Public API surface and underlying models may change without prior notice.
  Fix: Refer to the latest documentation and release notes for API changes when upgrading to new beta versions.
• gotcha Authentication is required and often done via `DefaultAzureCredential`. This credential provider attempts various authentication methods (environment variables, Azure CLI, managed identity). If not configured correctly, it will fail to connect.
  Fix: Ensure you are logged into Azure CLI (`az login`), or have environment variables like `AZURE_CLIENT_ID`, `AZURE_TENANT_ID`, and `AZURE_CLIENT_SECRET` set for service principal authentication.
• gotcha The Azure AI Inference service endpoint must be provided to the client. This is typically done via the `AZURE_AI_INFERENCE_ENDPOINT` environment variable or passed directly to the `AzureAIInferenceClient` constructor. Forgetting this will lead to connection errors.
  Fix: Set the `AZURE_AI_INFERENCE_ENDPOINT` environment variable to your specific Azure AI Inference service URL (e.g., `https://<your-service-name>.inference.ai.azure.com/`).

Install

• pip install azure-ai-inference azure-identity Install core library and authentication

Imports

• AzureAIInferenceClient

  from azure.ai.inference import AzureAIInferenceClient

• ChatCompletionsOptions

  from azure.ai.inference.models import ChatCompletionsOptions

• DefaultAzureCredential

  from azure.identity import DefaultAzureCredential

Quickstart

This quickstart demonstrates how to initialize the `AzureAIInferenceClient`, authenticate using `DefaultAzureCredential`, and perform a basic chat completion request. Ensure your `AZURE_AI_INFERENCE_ENDPOINT` environment variable is set and you are logged into Azure CLI or have appropriate service principal environment variables configured.

import os
from azure.ai.inference import AzureAIInferenceClient
from azure.ai.inference.models import ChatCompletionsOptions
from azure.identity import DefaultAzureCredential

# Set your Azure AI Inference service endpoint as an environment variable:
# export AZURE_AI_INFERENCE_ENDPOINT="https://<your-service-name>.inference.ai.azure.com/"
endpoint = os.environ.get("AZURE_AI_INFERENCE_ENDPOINT", "")

if not endpoint:
    raise ValueError("Please set the AZURE_AI_INFERENCE_ENDPOINT environment variable.")

# Authenticate using DefaultAzureCredential (checks environment variables, Azure CLI, etc.)
credential = DefaultAzureCredential()

try:
    # Initialize the client
    client = AzureAIInferenceClient(endpoint=endpoint, credential=credential)

    # Example: Chat Completion
    messages = [
        {"role": "system", "content": "You are a helpful AI assistant."},
        {"role": "user", "content": "What is the capital of France?"},
    ]
    chat_options = ChatCompletionsOptions(
        messages=messages,
        model="gpt-35-turbo",  # Replace with your deployed model name
        max_tokens=128
    )

    print(f"Sending chat completion request to model: {chat_options.model}...")
    response = client.chat.completions(chat_options)

    for choice in response.choices:
        print(f"Assistant: {choice.message.content}")

except Exception as e:
    print(f"An error occurred: {e}")
    print("Ensure AZURE_AI_INFERENCE_ENDPOINT is set and you are authenticated to Azure (e.g., via `az login`).")