Manual or server VAD for Node.js Elevenlabs STT plugin?

Agents
, ,
1

In @livekit/agents-plugin-elevenlabs@1.3.2 JS STT, realtime Scribe appears to default to ElevenLabs provider VAD because serverVad undefined results in commit_strategy=vad, while serverVad: null switches to manual.

For a LiveKit AgentSession that already uses Silero VAD plus LiveKit turn detection, is the intended setup to leave ElevenLabs STT on provider VAD, or should callers disable provider VAD and manually flush/commit based on LiveKit user state?

Also, is there any supported public API for manual ElevenLabs STT commits in the JS plugin, or is SpeechStream.flush() only intended to flush buffered audio frames?

2

Has anyone had any experiences on this topic they can share, how did you implement this in your project, what about with other STT plugins?