Press F5. Speak.
Ship.

Hands-free coding for macOS. Hold F5, dictate to your AI in natural speech, get on-device transcription in under two seconds. Your AI speaks back. You never leave your editor.

Get started View source
Recording
Transcribed · whisper.cpp · 1.2s

Split the JWT validation into its own middleware function. Add refresh token rotation and keep the cookie fallback for the admin dashboard.

11 MCP tools
546 tests passing
<1.5s local STT
Capabilities

Built for devs who think faster than they type

On-device STT

whisper.cpp large-v3-turbo runs entirely on your Mac. Sub-1.5s transcription, Hebrew + English in the same sentence. No cloud, no API key, no latency.

VoiceBar

Native SwiftUI menu bar app. F5 push-to-talk via hidutil, live waveform pill, real-time teleprompter. Collapses to a dot when idle. Draggable anywhere.

MCP tools

voice_speak and voice_ask cover the core speak-and-listen loop, while aliases like qa_voice_announce, qa_voice_brief, qa_voice_consult, and qa_voice_ask keep backward-compatible QA shortcuts across all 11 MCP tools.

How it works

F5 → text at your cursor in <2 seconds

F5
Hold F5
0ms
hidutil relay triggers VoiceBar recording
Speak
~3s
Silero VAD detects speech & silence
Transcribe
~1.1s
whisper.cpp large-v3-turbo, on-device
Paste
instant
Auto-pastes into your active field
Live demo

A voice conversation with your AI

voicelayer — session
stt: idle tts: idle lang: —
Idle
VoiceBar connected | mode: standby | buffer: 20 slots | vad: silero
Works with

Any MCP client

Claude Code
Claude Code
Cursor
Cursor
Zed
Zed
VS Code
VS Code
Codex
Codex
Kiro
Kiro
Gemini CLI
Gemini CLI
Getting started

Three steps

01

Install from npm

02

Launch VoiceBar

03

Press F5 and start talking

Stop typing. Start talking.

One install. On-device STT. Works with any MCP client.