Small models for voice agents that run on the device: in the browser (WebAssembly), in Python and on Android, with plugins for LiveKit Agents and Pipecat. No server, no key. Code and weights under Apache-2.0. Early preview (0.x).
| Model | What it does | Size |
|---|---|---|
| Focus | Keeps the closest voice, or one person's voice from their print, and removes noise, music and other voices | 1.1 MB |
| Cue | Tells when someone speaks, every 30 ms | 1.1 MB |
| Spot | Hears any word you type, in 7 languages, with no training | 2.8 MB |
| Lex | Turns a written word into its likely pronunciations | 1.8 MB |
| Timbre | Makes voiceprints and compares them | 0.74 MB |