LocalVoiceAI is a free, on-device push-to-talk voice transcription tool for macOS. Start the service once, then switch to any app — Claude, Cursor, VS Code, your browser, anything — hold Fn+F10, speak, and release. Your words appear as text at the cursor, instantly, powered entirely by a local Whisper model running on your Mac’s Apple Metal GPU. No cloud connection, no API key, no subscription required.Documentation Index
Fetch the complete documentation index at: https://mintlify.com/npateriya/LocalVoiceAI/llms.txt
Use this file to discover all available pages before exploring further.
Why LocalVoiceAI?
Most voice-to-text solutions either require a paid cloud subscription or send your audio to a remote server for processing. LocalVoiceAI solves both problems: it runs the Whisper speech recognition model locally on your Apple Silicon GPU, keeping your voice data completely private while delivering transcription in roughly one to two seconds — faster than most network round-trips.| LocalVoiceAI | Cloud STT (Whisper API, etc.) | |
|---|---|---|
| Cost | Free | Pay per minute |
| Privacy | Fully on-device | Audio sent to cloud |
| Latency | ~1–2s (Metal GPU) | Network dependent |
| Works offline | Yes | No |
| Setup | One-time | API key required |
System Requirements
Before installing, confirm your Mac meets these requirements:- Apple Silicon Mac — M1 or later (arm64 architecture required)
- macOS 14 Sonoma or later
- Homebrew — used to install
whisper-cppandportaudio
Key Features
On-Device Privacy
Every audio byte is processed locally. Nothing is ever transmitted to a remote server, making LocalVoiceAI safe for sensitive work, confidential conversations, and air-gapped environments.
Metal GPU Speed
Transcription is accelerated by Apple Metal, running the Whisper model on your GPU. A typical short phrase transcribes in approximately one to two seconds.
System-Wide Push-to-Talk
A C-level
CGEventTap monitors keypresses globally, so LocalVoiceAI responds to Fn+F10 no matter which app has focus — terminals, browsers, editors, chat apps, or anything else.LaunchAgent Service
LocalVoiceAI runs as a macOS LaunchAgent, independent of any terminal window or IDE. It starts automatically at login and continues working whether you launched your apps from Terminal, iTerm2, or VS Code.
Auto Model Download
On first run, the Whisper
ggml-small model (~244 MB) is downloaded automatically to ~/.cache/localvoice/. No manual model management needed to get started.Configurable Hotkey
The push-to-talk key defaults to F10 (keycode 109) but can be overridden with the
WHISPER_KEYCODE environment variable. Common alternatives include F8 (100), F9 (101), F11 (103), and F12 (111).