Azure AI Foundry Integration for Microsoft Agent Framework

1.0.0rc6 · active · verified Sun Apr 12

The `agent-framework-azure-ai` library provides an integration layer for various Azure AI services (e.g., Azure OpenAI, Azure AI Search, Azure Speech, Azure Vision) with the Microsoft Agent Framework. It enables developers to easily incorporate these powerful Azure capabilities into their agent-based applications. Currently at version `1.0.0rc6`, its release cadence is tied to the evolving Microsoft Agent Framework and Azure AI services, with new release candidates emerging as features stabilize.

Warnings

breaking This library is currently in Release Candidate (RC) status (`1.0.0rc6`). APIs, module paths, and behaviors are subject to significant change without backward compatibility guarantees until a stable `1.0.0` release. Users should expect frequent updates and potential breaking changes.
Fix: Regularly check the official GitHub repository and PyPI for updates. Pin dependency versions for production, but anticipate necessary code modifications during upgrades.
gotcha This library is an extension for the `agent-framework`. It *requires* `agent-framework` to be installed and understood. Core concepts like `Agent`, `LLM`, `Tool` come from the parent framework, not this library.
Fix: Familiarize yourself with the `agent-framework` documentation before using this integration library.
gotcha Proper configuration of Azure AI service credentials (endpoint, API key, deployment name, etc.) via environment variables or explicit parameters is critical. Misconfigurations often lead to authentication errors or incorrect service routing.
Fix: Ensure `AZURE_OPENAI_ENDPOINT`, `AZURE_OPENAI_API_KEY`, `AZURE_OPENAI_DEPLOYMENT_NAME` (and similar for other services) are correctly set. Double-check your Azure resource configurations.
breaking The library explicitly pins `pydantic<2` in its `install_requires`. Using `pydantic>=2` in your project alongside this library will cause dependency conflicts and likely runtime errors.
Fix: Ensure your project's `pydantic` version is less than 2. If other dependencies require `pydantic>=2`, you may face irreconcilable conflicts until this library updates its `pydantic` compatibility.

Install

pip install agent-framework-azure-ai Install core library

Imports

AzureOpenAIService

from agent_framework_azure_ai.azure_open_ai_service import AzureOpenAIService

AzureAISearchService

from agent_framework_azure_ai.ai_search_service import AzureAISearchService

AzureSpeechService

from agent_framework_azure_ai.azure_speech_service import AzureSpeechService

AzureVisionService

from agent_framework_azure_ai.azure_vision_service import AzureVisionService

Agent
```
from agent_framework.core.agent import Agent
```
Imported from the parent `agent-framework` library.

Quickstart

This quickstart demonstrates how to initialize the `AzureOpenAIService` provided by the library and integrate it as an LLM with the Microsoft Agent Framework. It requires proper Azure OpenAI service credentials set as environment variables.

import os
from agent_framework.core.agent import Agent
from agent_framework.core.model import ModelConfig
from agent_framework_azure_ai.azure_open_ai_service import AzureOpenAIService

# Ensure environment variables are set for Azure OpenAI
azure_openai_endpoint = os.environ.get("AZURE_OPENAI_ENDPOINT", "https://YOUR_AZURE_OPENAI_ENDPOINT.openai.azure.com/")
azure_openai_api_key = os.environ.get("AZURE_OPENAI_API_KEY", "YOUR_AZURE_OPENAI_API_KEY")
azure_openai_deployment_name = os.environ.get("AZURE_OPENAI_DEPLOYMENT_NAME", "YOUR_DEPLOYMENT_NAME")

if "YOUR_AZURE_OPENAI" in azure_openai_endpoint or "YOUR_AZURE_OPENAI" in azure_openai_api_key or "YOUR_DEPLOYMENT_NAME" in azure_openai_deployment_name:
    print("Please set AZURE_OPENAI_ENDPOINT, AZURE_OPENAI_API_KEY, and AZURE_OPENAI_DEPLOYMENT_NAME environment variables.")
    print("Example: export AZURE_OPENAI_ENDPOINT='https://your-resource.openai.azure.com/'")
    print("Example: export AZURE_OPENAI_API_KEY='...' ")
    print("Example: export AZURE_OPENAI_DEPLOYMENT_NAME='gpt-4o' ")
    exit(1)

# 1. Initialize the Azure OpenAI Service as an LLM
azure_llm_service = AzureOpenAIService(
    azure_endpoint=azure_openai_endpoint,
    api_key=azure_openai_api_key,
    deployment_name=azure_openai_deployment_name,
    model_config=ModelConfig(max_tokens=256) # Optional: Configure LLM behavior
)

# 2. Create an Agent and provide it the Azure LLM
agent = Agent(name="AzureAIAgent").with_llm(azure_llm_service)

# 3. Use the agent to get a response
print(f"Agent '{agent.name}' is ready to chat...")
response = agent.chat("What is the capital of France?")
print(f"Agent response: {response.output}")

# Example with a simple tool (if using other Azure AI services as tools)
# from agent_framework.core.tool import Tool
# class MyAzureAISearchTool(Tool):
#     def call(self, query: str) -> str:
#         # Placeholder for actual Azure AISearchService logic
#         return f"Searching Azure AI Search for: {query}"
# agent.with_tool(MyAzureAISearchTool())
# response_with_tool = agent.chat("Find documents about renewable energy.")

view raw JSON →