Screen face
Start a conversation, then type. Escape stops a reply.
Connection
Checking API configuration…
Local prototype only. Keys belong in the backend’s .env file. Camera and wake word are not active.
Plays a short tone locally to check sound. No API calls.
Microphone testing stays on this device.
Conversation tuning
Higher values reject more background sound. This is basic voice activity detection, not speaker recognition.
- Reasoning route
- —
- Latest input → first playback
- —
- Interruption → stop request
- —
- Voice latency median / p95
- —
- Input audio
- 0 s
- Model tokens in / out
- 0 / 0
- Speech characters requested
- 0
Usage counters are partial on failed/cancelled calls. Use actual provider billing for the cost projection below. The session has duration and turn limits.
Conversation check
Try a ten-minute voice conversation: pause mid-thought, interrupt a long reply, correct a name, mute/unmute, then repeat with speaker and game audio. Mark problems as you hear them.
Start a conversation to measure it. Reports contain timings and counters, never transcripts or recordings.
After ending the session, enter its total cost from all provider dashboards to project a month of similar use.