-black?logo=apple&style=flat-square){kind=icon}
demo.mp4
Voca is a Claude Code plugin that adds an ambient voice interface to your coding sessions. It listens to your microphone, transcribes speech in real-time, and speaks back—all running locally on Apple Silicon. No cloud APIs, no latency, no data leaving your machine.
- 🗣️ Natural Voice Interface: Talk to Claude Code naturally as you work.
- ⚡️ Real-time Processing: Sub-second transcription ensures fluid conversations.
- 🔒 Privacy First: 100% local inference. Your audio never leaves your machine.
- 🎙️ Advanced Audio Routing: Seamless text-to-speech and voice activity detection built directly in.
- 🔁 Instant Focus Handoff: Starting voice mode in a second session interrupts the active poll and transfers ownership immediately.
All inference runs on-device using MLX.
⚠️ Requirement: macOS with Apple Silicon (M1+) is required to run Voca.
Add the marketplace and install the plugin:
/plugin marketplace add patelnav/voca
/plugin install voca@patelnav-voca-
Configure Devices: Run the setup command to configure your audio input/output devices and verify that everything is working:
/voca:setup
-
Start a Voice Session: Once setup is complete, start the ambient voice interface:
/voca:start
Voca uses token-based HTTP polling under the hood. The first poll claims focus; every later poll in that session reuses the returned token. Starting voice mode in another session claims a new token and hands off listening immediately.
Claude will now listen in the background and respond seamlessly to your voice.
Voca is built on cutting-edge local AI technologies to ensure maximum performance and privacy:
| Component | Technology | Description |
|---|---|---|
| Speech-to-Text | Parakeet MLX v2 | Blazing fast transcription (~0.1s on M-series) |
| Text-to-Speech | Kokoro via mlx-audio | High-quality, natural voice generation |
| Voice Activity | Silero VAD | Enterprise-grade voice activity detection |
| Runtime | Python MCP | Model Context Protocol server over stdio |