Skip to content

Conversation

@JonathanLennox
Copy link
Owner

No description provided.

rpurdel and others added 30 commits November 26, 2025 11:00
Add source maps for emscripten-generated code.
Make turn_detection configurable with an env var.
* Add metrics

* Debug logs

* Remove metric debug log
I'm not sure what Cloudflare does if you don't do this, but it may relate to
how we saw a callback stop firing.
Previously we just checked if they were truthy, but "false" is truthy...
Enable logpush feature for the worker
rpurdel and others added 19 commits December 9, 2025 10:58
* Fix PLC logic.

* Add metrics for various Opus events.

* Run 'prettier'.
* Add some metrics for normal correct behavior.

* Only log happy-path metrics if env.DEBUG is true.
* fix: rm var shadowing

* fix: avoid race condition between reset triggered and OpenAI response

* fix: memory leaks on error, don't close already closed connection

* fix: check if shift returns undefined

* fix: if undefined set to now on interim
* chore: log unhandled openai events

* chore: run prettier
* feat: add pong to ping

* feat: extend cpu limit to 5 minutes
* feat: transcription flush

* fix: add missing comma
This can happen if our manual commit happens to hit right after
the VAD detected that speech stopped.
* Use OPENAI_TURN_DETECTION as JSON.

* Adjust default turn detection.

* Make sure the turn detection is an object and has a type field.
* Mirror the "id" field when responding to ping.

* Add an "event" field to transcription messages, so they parse as MediaJson.
* fix: log the actual error instead of [object ErrorEvent]

* fix: get reason and code from close event
Aggregate happy-path metrics so as to not overwhelm the metrics system.
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

4 participants