Say it aloud. It types.
Aloud is a macOS menu bar dictation app. Hold a key, speak, let go — your words appear in whatever app you're typing in. That's the whole product.
- Private by design. Speech is transcribed entirely on your Mac. Audio never leaves your machine — no cloud, no account, no subscription, no telemetry.
- Works everywhere. Terminal, browser, notes, chat — anywhere a cursor blinks.
- Fast. Powered by the Apple Neural Engine; a sentence transcribes in a blink.
- Invisible until you need it. A quiet menu bar icon, a small indicator while you speak, nothing else.
- Download
Aloud.dmgfrom the latest release. - Drag Aloud to Applications and open it.
- Follow the short setup: allow the microphone, allow accessibility, and let the one-time voice model download (~500 MB). After that, Aloud works fully offline.
Requires macOS 14 (Sonoma) or later on Apple Silicon.
Hold right ⌥, speak, release. Text appears where your cursor is. Press Esc while holding to cancel. Double-press the key to keep listening hands-free — edit, click around, keep talking — then press Esc (or double-press again) to finish.
Aloud tidies as it types — removes "um"s, honors "scratch that", applies your personal word fixes — at a clean-up level you choose in Settings (or turn it off for word-for-word). The exact words you said are always kept in History, so nothing is ever silently rewritten. Settings also has the hotkey, microphone, launch-at-login, and a Vocabulary list for names and terms it should always get right. Everything stays on this Mac.
Aloud updates itself: when a new release is out, an "Update available" item appears in the menu — one click, it swaps itself and relaunches. No background nagging.
Speech recognition runs on-device with NVIDIA Parakeet TDT 0.6B v3 (CC-BY-4.0), converted to CoreML and driven by the FluidAudio SDK (Apache-2.0) on the Apple Neural Engine. The engine is an implementation detail behind a small interface and may change as better on-device models appear.
The only network calls Aloud ever makes: the one-time model download (Hugging Face) and a daily GitHub check for new releases. Nothing is sent, no identifiers, nothing else. Grep the source — it's small.
git clone https://github.com/theabecaster/aloud && cd aloud
swift build && swift test
.build/debug/Aloud --selftest # headless subsystem checks
bash scripts/make-app.sh # stage dist/Aloud.appBranches: work lands on dev; merges to main cut a signed, notarized release automatically (version derived from Conventional Commits).
Copyright © 2026 Abraham Gonzalez.
The app source — including all previously published releases — is offered under the GNU AGPL-3.0: you're free to use, study, modify, and redistribute it, but any derivative — including one offered as a network service — must be released under the AGPL with its full source available.
The speech model weights are licensed separately by their authors (currently CC-BY-4.0 — attribution above).