Make intermediate results available when MaxTurnExceededException is thrown

[chiehmin-wei]

Describe the feature

I want to access intermediate results when MaxTurnExceededException is thrown.

I have a workflow broken into 2 phases.
Phase 1 is investigation.
Phase 2 is writing up a report based on the investigation results.

I'm feeding the results of phase 1 into phase 2 like so:

result = Runner.run(investigation_agent, max_turns=5)
report = Runner.run(report_agent, result.to_input_list())

I want to limit the number of turns in Phase 1.
If a MaxTurnExceededException is thrown in Phase 1, I still want to make use of the available information in Phase 2.
Currently this doesn't seem possible to do. I have no way to access the intermediate results (tool calls, etc.) if I reach max turns in Phase 1.

[rm-openai]

Yes we should indeed do this. Will try to get to it soon - as always, PR welcome from anyone

[DanieleMorotti]

I can have a look if @chiehmin-wei is not already working on it. How would you return / save the partial results?
In the Runner object?

[rm-openai]

I think partial results are already built up in Runner, as you described; when the exception is raised, you can add a RunErrorDetails object with all that data

[DanieleMorotti]

Okay, so I have to create a new class RunErrorDetails and then pass the data in the MaxTurnsExceeded exception? Obviously adding this optional attribute to the exception

[rm-openai]

yup!