Vox 🎙️

Free, on-device voice dictation. Hold a key, talk, release — your words appear wherever the cursor is (terminal, editor, browser, any text field).

Everything runs locally. No audio ever leaves your machine, no API keys, no subscription, no cloud.

• macOS — native menu-bar app using Apple's on-device SpeechAnalyzer (macOS 26).
• Windows — Python port using local Whisper (faster-whisper). See windows/.

A free, private alternative to cloud dictation tools. Tiny and focused.

The sections below cover the macOS app. For Windows, see windows/README.md.

Features

• Push-to-talk — hold a key, speak, release. That's the whole interaction.
• On-device — uses Apple's SpeechAnalyzer (macOS 26). Private and offline.
• Pastes anywhere — works in terminals too (great for CLIs like Claude Code).
• Menu-bar only — no Dock icon, no window. A mic glyph that turns red while listening.
• Multilingual — switch language from the menu (ships with English + Portuguese).
• Free & MIT-licensed.

Requirements

• macOS 26 (Tahoe) or later — required for the on-device SpeechAnalyzer API.
• Xcode 26 / Swift 6 toolchain to build from source.

Install

Build from source

git clone https://github.com/tdzapps-dev/vox.git
cd vox
./scripts/make-app.sh          # builds build/Vox.app (release)
cp -R build/Vox.app /Applications/
open /Applications/Vox.app

make-app.sh signs with your Apple Development identity if you have one (so the macOS permission grants persist across rebuilds), and falls back to ad-hoc signing otherwise.

If macOS blocks an ad-hoc build on first launch: right-click the app → Open, or run xattr -dr com.apple.quarantine /Applications/Vox.app.

Permissions (first launch)

macOS will ask for three — all required:

• Microphone — to capture speech.
• Speech Recognition — to transcribe on-device.
• Accessibility — to listen for the global hotkey and paste the result. (System Settings → Privacy & Security → Accessibility → enable Vox.)

Vox auto-arms the hotkey the moment Accessibility is granted — no restart. The speech model for each language downloads once on first use.

Usage

1. Click into any text field.
2. Hold ⌃ Control, speak, then release.
3. The transcribed text is pasted into the focused app.

Menu-bar icon states: mic idle · red mic.fill listening · waveform transcribing.

Changing the hotkey

The trigger is a single constant in Sources/Vox/HotKeyMonitor.swift:

private let triggerKeyCode: UInt16 = 59          // 59 = Control
private let triggerFlag: NSEvent.ModifierFlags = .control

Set it to another modifier (e.g. 61 + .option for right Option) and rebuild.

How it works

• Hotkey — a global NSEvent monitor for flagsChanged (Accessibility only, no Input Monitoring needed).
• Capture — AVAudioEngine taps the mic and converts buffers to the analyzer's format.
• Transcription — SpeechAnalyzer + SpeechTranscriber on-device, streaming.
• Insertion — copies to the pasteboard and synthesizes ⌘V, then restores the previous clipboard.

Source layout:

File	Role
main.swift	Entry point (menu-bar-only agent)
MenuBarController.swift	Status item, menu, dictation flow
HotKeyMonitor.swift	Push-to-talk via global key monitor
DictationEngine.swift	Audio capture + on-device SpeechAnalyzer
TextInjector.swift	Pasteboard + synthetic ⌘V
Permissions.swift	Mic / speech / accessibility

Windows

A Python port lives in windows/ — same idea (hold a key, speak, release, paste anywhere), running Whisper locally via faster-whisper. It runs from the system tray and is installed with a couple of .bat scripts. Full instructions in windows/README.md.

Roadmap

• Configurable hotkey from the UI (currently a one-line code change).
• Double-tap-to-lock for hands-free long-form dictation.
• Better punctuation (commas in particular) and an optional cleanup pass.

PRs welcome.

License

MIT — see LICENSE.