Voice engines
The voice engine turns your companion’s words into speech, synchronized with her lips and face. Three engines are available in Settings → Voice; every character keeps their own distinct voice on every engine, so switching engines never makes Grace sound like someone else.
The engines
Section titled “The engines”ElevenLabs: cloud, maximum realism
Section titled “ElevenLabs: cloud, maximum realism”The most natural voices available, rendered in ElevenLabs’ cloud. Requires your own ElevenLabs API key, and speech costs their usual per-character rates. Choose this when voice fidelity matters more than keeping things local.
Supertonic-3: local, cloned voices, 31 languages
Section titled “Supertonic-3: local, cloned voices, 31 languages”A fast on-device engine that gives every character their own cloned voice, with broad language coverage if your conversations aren’t in English. No key, no account, no network. The recommended local default.
Kokoro: local, open-weight
Section titled “Kokoro: local, open-weight”A fully open-source engine. Choose Recommended and Unclaw installs and runs it for you, or point at your own Kokoro server if you already run one.
Choosing
Section titled “Choosing”| You want | Pick |
|---|---|
| Local, instant, character-true voices | Supertonic-3 |
| Conversations in other languages | Supertonic-3 |
| The most lifelike voice possible | ElevenLabs (your key) |
| Open-source everything / your own server | Kokoro |
- Voice switching is instant: no restart, the next reply uses the new engine.
- Loudness is matched across engines, so switching never blasts or whispers.
- Voice input (your microphone) is always processed locally regardless of the speech engine. See Privacy.