How can I access AI Foundry Agents in Python application with API Key

Question

I have created an agent for Azure Bing search with grounding. How do I access the agent created in Azure Foundry with API Key in a python app

Answer 1

Hi Venkateswaran,
AI Foundry Agents cannot be accessed by API keys.

You can use Entra ID authentication to access them in your python application.

Here is the supported documentation:
https://learn.microsoft.com/en-us/python/api/overview/azure/ai-agents-readme?view=azure-python

Below is a example code snippet for creating and using an agent with a Bing grounding resource:

import asyncio
import os
from azure.identity.aio import DefaultAzureCredential
from semantic_kernel.agents import AzureAIAgent, AzureAIAgentThread
from azure.ai.agents.models import BingGroundingTool
from azure.core.exceptions import ResourceNotFoundError
AGENT_NAME = "MyAgent"
ENDPOINT = "<YOUR_PROJECT_ENDPOINT>"
DEPLOYMENT_NAME = "<YOUR_MODEL_DEPLOYMENT_NAME>"
GROUNDING_CONNECTION_NAME = "<YOUR_GROUNDING_CONNECTION_NAME>"
os.environ["AZURE_AI_AGENT_ENDPOINT"] = ENDPOINT
os.environ["AZURE_AI_AGENT_MODEL_DEPLOYMENT_NAME"] = DEPLOYMENT_NAME
async def run_agent_by_name():
    try:
        async with (
            DefaultAzureCredential() as creds,
            AzureAIAgent.create_client(credential=creds) as client,
        ):
            print("Client initialized.")
            found_agent = None
            print(f"Looking for agent with name: {AGENT_NAME}")
            async for agent in client.agents.list_agents():
                if agent.name == AGENT_NAME:
                    found_agent = agent
                    break
            if not found_agent:
                print(f"No agent found with name '{AGENT_NAME}', creating a new one...")
                grounding_connection = await client.connections.get(
                    name=GROUNDING_CONNECTION_NAME
                )
                conn_id = grounding_connection.id
                bing_tool = BingGroundingTool(connection_id=conn_id)
                new_agent_def = {
                    "name": AGENT_NAME,
                    "model": DEPLOYMENT_NAME,
                    "instructions": "You are a helpful AI agent. Use Bing Search grounding to fetch real-time information.",
                    "tools": bing_tool.definitions,
                }
                found_agent = await client.agents.create_agent(new_agent_def)
                print(f"Created new agent '{AGENT_NAME}' with ID: {found_agent.id}")
            else:
                print(f"Found agent '{AGENT_NAME}' with ID: {found_agent.id}")
            agent_definition = await client.agents.get_agent(found_agent.id)
            agent = AzureAIAgent(client=client, definition=agent_definition)
            print("Azure AI Agent initialized successfully. New conversation thread created. Conversation started.")
            thread = AzureAIAgentThread(client=client)
            user_input = "What is the temperature and time in Seattle now?"
            print(f"Sending message: {user_input}")
            try:
                response = await asyncio.wait_for(
                    agent.get_response(messages=user_input, thread=thread),
                    timeout=60,
                )
                print("\nAgent response raw:")
                print(response)
                if hasattr(response, "message"):
                    print("\nAgent response content:")
                    print(response.message.content)
            except asyncio.TimeoutError:
                print("Error: The request to the agent timed out.")
    except Exception as e:
        import traceback
        print("An unexpected error occurred:", str(e))
        print("Message:", str(e))
        traceback.print_exc()
    finally:
        pass
if __name__ == "__main__":
    asyncio.run(run_agent_by_name())

Documentation to use Grounding with Bing Search:

https://learn.microsoft.com/en-us/azure/ai-foundry/agents/how-to/tools/bing-code-samples?pivots=python

Feel free to accept this as an answer.

Thank you for reaching out to the Microsoft Q&A Portal.