Vocal is an on-device text-to-speech engine. It runs entirely on your Mac's GPU — select any text, right-click, and your Mac reads it aloud. No cloud, no internet, nothing leaves your machine.
Right-click any text → Services → Speak with Vocal
Same engine, three surfaces — pick whichever fits the moment.
Select any text anywhere on your Mac, right-click, and choose Speak with Vocal. Built into the system, works everywhere.
→ the default, zero-app pathA desktop app to load models, switch backends, test voices, and watch the worker logs in real time.
→ for setup & tuningVocal exposes a speak tool over MCP, so AI agents like Claude Code and Cursor can talk out loud.
Inference runs on your Mac's Metal GPU. No audio or text is ever sent to a server.
English and Hindi on the native engine today; the pluggable Qwen3 backend covers more languages.
Audio streams as it's generated, so playback starts fast. The model frees itself the instant it goes idle.
A pure-Swift Chatterbox engine by default — no Python — with Qwen3-TTS and mlx-audio backends a setting away.