- Voice: the full feature description - Voice Setup: connecting LiveKit and a speech provider

Voice FAQ

Whether Voice is phone calling, what it needs to work, and where your audio goes.

Is Voice a phone call feature?

No. Voice is a real-time conversation with an agent in your browser, using WebRTC through LiveKit. There's no telephony, no dialing a phone number, and no outbound calling involved.

Do I need my own accounts to use it?

Yes. Voice needs a LiveKit project for the audio connection and at least one speech provider account (Deepgram or ElevenLabs) for transcription, connected in Settings → Voice. See Voice Setup.

Is my audio recorded or stored?

No. Only the resulting text transcript is kept, the same as any other message in the conversation. Raw audio isn't persisted anywhere.

Can I use voice without the agent speaking back?

Yes. Connecting a transcription provider without a text-to-speech provider is a supported setup: you speak, the agent replies in text. See Voice.

Why can't my agent speak a certain language?

Vendor coverage varies. Deepgram, for example, transcribes Mandarin but doesn't speak it, while ElevenLabs covers both directions for every language this platform ships in. If a language has no vendor for a given direction, the settings page states that directly instead of letting a session start and answer in the wrong language.

Why does my spoken message show up as editable text before it sends?

That's expected, and the point: every spoken message appears as a draft transcript first, so a misheard word can be corrected before the agent reads it.

Voice: the full feature description
Voice Setup: connecting LiveKit and a speech provider