Python: ChatCompletionAgent constructor and invocation improvements

[moonbox3]

Supporting the simplified agent scenario:

# necessary imports

async def main():
    agent = ChatCompletionAgent(
        service=OpenAIChatCompletion(),
        plugins=[DatePlugin(), WeatherPlugin()], #optional
        instructions="you are a helpful assistant that can answer questions about the world", #optional
    )

    print(await agent.invoke("What is the capital of France?"))

if __name__ == "__main__":
    asyncio.run(main())

as well as make updates for invoke: from AsyncIterable -> couroutine.

[eavanvalkenburg]

Wouldn't you need Kernel as well in order to do the function calling and have things like Filters available as well?

# necessary imports

async def main():
    kernel.add_plugins([DatePlugin(), WeatherPlugin()])
    agent = ChatCompletionAgent(
        service=OpenAIChatCompletion(),
        kernel=kernel
        instructions="you are a helpful assistant that can answer questions about the world", #optional
    )

    print(await agent.invoke("What is the capital of France?"))

if __name__ == "__main__":
    asyncio.run(main())

[moonbox3]

Today, if a kernel is not provided, we create one. So we'll use a model_validator after the model is formed, and add the plugins to the kernel for the caller. This is considered the "easiest" of easy paths where someone wants to write the most minimal amount of code to get started. We will still support that path where one can configure the kernel and pass it in, which you could say is the "medium/intermediate" path. Passing in plugins via the constructor will take precedence over any plugins added previously.

[eavanvalkenburg]

gotcha, thanks for clarifying, that sounds good, less is more!