Question on realtime agent: behaviour related to conversation.item.truncate truncate event.

[nbondoux]

Hello,
In the realtime agent code (agents/realtime/openai_realtime.py), I can read the following comment in the input_audio_buffer.speech_started handler:
"
# Avoid sending conversation.item.truncate here; when GA is set to
# interrupt on VAD start, the server will handle truncation.
"

I wanted to double-check with you that it is true, because I see nowhere in the documentation of the realtime API that the conversation.item.truncate is automatically handled on server side.
Thanks

[seratch]

Thanks for asking this question. Indeed, the comment could be confusing, so I've updated the comment at #2000

[nbondoux]

Thanks for asking this question. Indeed, the comment could be confusing, so I've updated the comment at #2000

Thanks for you prompt answer, but however, my question was "is the SDK really aligned on this with the realtime API ?". Indeed, when Interrupt on VAD is set, the SDK will never generate any interrupt event, and I believe the LLM server won't either (how would it be possible, has it doesn't have any play back tracker).

[seratch]

This SDK's default settings go with the current one, and we recommend for most use cases. That being said, if it does not fit for your apps, you can set your own initialization options and send the events as necessary.

[nbondoux]

So should I send the truncate event myself ? I thought the openai stack was supposed to do that as long as I use a playback tracker.