ag_ui module not working with structured `output_type`

[lukekh]

Initial Checks

• I confirm that I'm using the latest version of Pydantic AI
• I confirm that I searched for my issue in https://github.com/pydantic/pydantic-ai/issues before opening this issue

Description

Function calling seems to be working in general, however when there is a structured output_type the AG-UI client doesn't seem to return the corresponding tool_call_response while using the handle_ag_ui_request function.

Using logfire I could see that the tool_call_response was generated in the response, but when the request is being built the tool_call_response message part is missing.

Example Code

from enum import Enum
from pydantic_ai import Agent
from pydantic_ai.ag_ui import handle_ag_ui_request
from starlette.requests import Request

class output_type(Enum):
    A = "a"
    B = "b"

agent = Agent(
    output_type=output_type
)

async def agent_endpoint(request: Request):
    return await handle_ag_ui_request(agent, request)

Python, Pydantic AI & LLM client version

python 3.11, pydantic-ai-slim==1.0.10, LLM client is OpenAI (Azure) with openai==1.109.1

[DouweM]

@lukekh We don't currently yield a tool call result event for output tools; see my comments in #2640 for more context.

A workaround would be to use the memory stream workarounds from #2382 to send a custom event from a handle_ag_ui_request on_complete handler. We intend to make emitting custom events easier in the future.

Another option would be to use a regular tool, rather than output_type, and prompt the model to always call it, which they're reasonably good at following. #2799 will also let you set tool_choice=required for non-output tools.