Loading benchmark data…
EyesInAI·Loading live benchmark data
Client bot · voice
Voice is a per-assistant setting, not a rebuild. Leave it offand it’s the normal chat. Turn it up and the assistant can read its replies aloud, or take a spoken question, transcribe it, answer, and speak back — while typing always still works as a fallback. Shown in the neutral indigo default.
type ↔ read
The standard chat — typing in, text back. The default; nothing changes unless a client turns voice on.
type ↔ read
Still typed, but served on a fast text model so replies stream back quicker. A routing hint, no audio.
type → hear
Every reply is spoken back as well as shown. Good for hands-busy or accessibility, keyboard still drives it.
speak ↔ hear
Tap the mic, speak, and hear the answer: speech-to-text → reply → spoken back. Typing always still works.
How a spoken turn flows
The mic captures the question in the widget.
Transcribed with Whisper, then treated like any typed question.
The same grounded, routed answer path as a typed turn.
The reply is spoken back and shown on screen together.
Where the keys live
Voice adds two hosted services, and both are wired the safe way: speech-to-text runs on our server’s key, and text-to-speech is proxied through the bench where the provider key lives. Your assistant’s config carries only the mode and a voice name — never a credential. Same as the rest of the platform: the bot config is data, the secrets stay server-side.