Voice agent · settings

Conversation model (Gemini, hears audio directly)

TTS engine

Voice Language Greeting (spoken first, no model call) Scenario / system prompt Follow-up after silence

s, then hang up after tries

The agent can call set_wait (e.g. "let me check my records" → 30 s) and hang_up. Changing any setting applies mid-call.

Turn taking

Silence to end turn 500 ms

Let me interrupt the bot (barge-in; use headphones)

Silero VAD (in-browser) → 16 kHz WAV → Gemini (audio in) → streamed text → TTS → speaker. No separate speech-to-text step. The silence slider applies at the next start.

Settings are remembered in this browser; the address bar holds a shareable preset. Reset to defaults