to_cli_sync: when agent is provided with `output_type` of a function, the function is called multiple times with partial input parameters

[AdiBenTolila]

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

when using agent with output_type of a function and running it using agent.to_cli_sync() the parameters of the function is streamed and the function is called for each new token of its parameters.

Example Code

from pydantic_ai import Agent

def output_func(output: str) -> list[str]:
    print("CALLED WITH:", output)
    # parameter to the function could be for example an SQL query and result with error or execution of partial redundant query 
    return output.split()

model = OpenAIChatModel(
    MODEL_NAME,
    provider=AzureProvider(
        azure_endpoint=ENDPOINT,
        api_version=API_VERSION,
        api_key=API_KEY,
    ),
)

agent = Agent(model, system_prompt="You are a helpful Assistant", output_type=output_func)

agent.to_cli_sync()

Python, Pydantic AI & LLM client version

using gpt-4o-mini from AzureOpenAI, Python 3.13.5 and pydantic-ai==1.0.0b1

[DouweM]

@AdiBenTolila This is expected insofar that the CLI streams the agent's output as it comes in, and since you've specified an output function, that output is the result of that function called over each chunk as it came in.

Is this causing issues for you?

If so, see if changing this stream_output to stream_text gives you better results:

pydantic-ai/pydantic_ai_slim/pydantic_ai/_cli.py

Line 295 in 7e8ebec

async for content in handle_stream.stream_output(debounce_by=None):

We could consider making that the default.

[github-actions]

This issue is stale, and will be closed in 3 days if no reply is received.

[github-actions]

Closing this issue as it has been inactive for 10 days.