Questions about end_strategy, is it being used? How does it work?

[bgiori]

Question

Hi, question about end_strategy. Searching through the code base it looks like it's not actually used anywhere?

I am trying to understand how exactly it works, and based on some testing I am not seeing a significant difference in the number of tool calls being used when setting early vs exhaustive. Wondering if anyone can provide more guidance on if the field is being used, and if so, how it is expected to work with a little more detail.

Thanks!

Additional Context

• Pydantic AI version: 0.6.0
• Python version: 3.11

[DouweM]

end_strategy is used here:

pydantic-ai/pydantic_ai_slim/pydantic_ai/_agent_graph.py

Lines 635 to 648 in d6d7f37

	calls_to_run: list[_messages.ToolCallPart] = []
	if final_result and ctx.deps.end_strategy == 'early':
	output_parts.extend(
	[
	_messages.ToolReturnPart(
	tool_name=call.tool_name,
	content='Tool not executed - a final result was already processed.',
	tool_call_id=call.tool_call_id,
	)
	for call in tool_calls_by_kind['function']
	]
	)
	else:
	calls_to_run.extend(tool_calls_by_kind['function'])

early means that if the model had already called an output tool (which will set final_result in that code), any parallel tools in the model response will not be executed.

[DouweM]

I'll close this issue assuming that clarifies it, but let me know if not and I'll reopen.