Monitoring

Track health of the voice pipeline end-to-end.

What to watch

• WebSocket/WebRTC connection counts and reconnect rates
• STT latency, LLM latency, TTS latency per provider
• Token usage and cost per session
• Error rates by provider/model

Endpoints

• /health for liveness; add custom checks for provider reachability if needed.

Dashboards

• Time-to-first-transcript
• Time-to-first-audio (TTS)
• Barge-in frequency