Loading a foundry agent as workflow executor

[pjirsa]

The current samples seem to be out of sync with the latest foundry SDKs for loading or creating agents as workflow executors. The agent-framework agent examples demonstrate the latest code which leverages the AzureAIProjectAgentProvider class to create or load agents. However, probably due to my inexperience with Python, I cannot figure out how to use these async functions when trying to declare my agent (or workflow) in the global namespace. If i use asyncio.run, i get an error that I can't do that from an already running loop. If I skip that, I get an error that my entity type is not correct.

# Agent instance following Agent Framework conventions
provider = AzureAIProjectAgentProvider(
    project_endpoint=os.environ.get("AZURE_AI_PROJECT_ENDPOINT"),
    model=os.environ.get("FOUNDRY_MODEL_DEPLOYMENT_NAME"),
    credential=AzureCliCredential()
)

agent = provider.create_agent(
    name="FoundryWeatherAgent",
    instructions="""
    You are a weather assistant using Azure AI Foundry models. You can provide
    current weather information and forecasts for any location. Always be helpful
    and provide detailed weather information when asked.
    """,
    tools=[get_weather, get_forecast],
)

Error is: Error: Entity loading failed: No valid entity found

OR with asyncio

provider = AzureAIProjectAgentProvider(
    project_endpoint=os.environ.get("AZURE_AI_PROJECT_ENDPOINT"),
    model=os.environ.get("FOUNDRY_MODEL_DEPLOYMENT_NAME"),
    credential=AzureCliCredential()
)

agent = asyncio.run(provider.create_agent(
    name="FoundryWeatherAgent",
    instructions="""
    You are a weather assistant using Azure AI Foundry models. You can provide
    current weather information and forecasts for any location. Always be helpful
    and provide detailed weather information when asked.
    """,
    tools=[get_weather, get_forecast],
))

Error is: Entity loading failed: Failed to load entity 'foundry_agent': asyncio.run() cannot be called from a running event loop

[moonbox3]

@pjirsa the following call needs an await as it is async:

agent = await provider.create_agent(
    name="FoundryWeatherAgent",
    instructions="""
    You are a weather assistant using Azure AI Foundry models. You can provide
    current weather information and forecasts for any location. Always be helpful
    and provide detailed weather information when asked.
    """,
    tools=[get_weather, get_forecast],
)

[moonbox3]

Also, this is the proper pattern since the provider should be within an async context manager:

async with (
    AzureCliCredential() as credential,
    AzureAIProjectAgentProvider(credential=credential) as provider,
):
    agent = await provider.create_agent(
        name="BasicWeatherAgent",
        instructions="You are a helpful weather agent.",
        tools=get_weather,
    )

[pjirsa]

Hey @moonbox3, thanks for the response. I have some additional information to share. When I serve the devui from the workflow script file itself, i am able to use the asyncio.run() function to wrap the async call. However, when running the devui from either the CLI or standalone process and then trying to load that workflow entity from its folder, nothing seems to work. That's when I get the errors described above. The fix you suggested also gives me the error "await" allowed only within async function