Skip to content
| Marketplace
Sign in
Visual Studio Code>Machine Learning>claude_purr — "Hey Claude" voice for Claude CodeNew to Visual Studio Code? Get it now.
claude_purr — "Hey Claude" voice for Claude Code

claude_purr — "Hey Claude" voice for Claude Code

DKS-Analytics GmbH

|
2 installs
| (0) | Free
Local voice control ("Hey Claude") for Claude Code: spoken commands land directly in the terminal prompt. macOS only (Apple Silicon).
Installation
Launch VS Code Quick Open (Ctrl+P), paste the following command, and press enter.
Copied to clipboard
More Info

claude_purr — "Hey Claude" voice for Claude Code

Hands-free voice control for Claude Code on macOS. Say "Hey Claude", speak your prompt, and it lands directly in the Claude Code terminal — with @file mentions auto-tagged. Fully on-device (Apple Neural Engine via WhisperKit); nothing leaves your Mac.

macOS only (Apple Silicon). Works in VS Code and Cursor.

How it works

The extension manages a small local voice daemon: wake word → on-device transcription → @-file tagging → it types the command into the active terminal where Claude Code runs. An animated overlay shows the live state — blue listening, red recording, green sent.

Setup

  1. Install this extension.
  2. Install the backend (one time) — click "Install backend" when prompted, or run:
    brew install whisperkit-cli pipx python@3.12
    pipx install claude-purr --python $(brew --prefix python@3.12)/libexec/bin/python
    
    (Pins Python 3.12 — the on-device deps have no Python 3.14 wheels yet.)
  3. Reload the window. Grant Microphone permission when macOS asks. The first run downloads the WhisperKit speech model (one time).
  4. Open a terminal, start claude, then say "Hey Claude …".

Controls

Click claude_purr in the status bar for the menu: Pause (microphone off) · Resume · Restart · Stop · Show log. The status icon shows the state: $(mic) active · $(debug-pause) paused · off.

Settings

  • Language — transcription language for your dictation (default German; the UI stays English).
  • Wake sensitivity — how easily "Hey Claude" triggers (lower = more sensitive).
  • Silence threshold, Overlay on/off, and an optional Backend path (advanced).

Privacy

All speech recognition runs locally on-device. No audio or text is sent to any server.


Made by DKS-Analytics GmbH — www.dks-analytics.de

  • Contact us
  • Jobs
  • Privacy
  • Manage cookies
  • Terms of use
  • Trademarks
© 2026 Microsoft