Clarify how to filter final vs. intermediate generation events in streaming API

[paulsengh]

Streaming sends multiple generation events where output[].type == "message". Some of those events also include a "function_call" entry (tool step). I only want to show the final assistant reply, but right now there’s no obvious way to tell final vs. intermediate messages.

Is there a flag or stable pattern to identify the final generation event (i.e. ones without any tool calls)? If not, what’s the recommended workaround to filter out intermediate messages that accompany tool calls without waiting for the full message generation, as that defeats the purpose of streaming?

[paulsengh]

Two ideas we've had:

• Prompt the model to tag when it's doing a tool call
• Try to detect if there's a type function_call in the output array

I don't think these are 100% reliable, so hoping there's a better solution cc @rm-openai

[seratch]

Thanks for asking the question! Checking the following resources might be helpful to you:

• https://github.com/openai/openai-agents-python/blob/main/examples/basic/stream_items.py
• https://platform.openai.com/docs/api-reference/responses-streaming