Mouthy

Talk, and it types. Free, open-source dictation that runs entirely on your computer.
No account, no subscription, no cloud.

Download · mouthy.dev

Press a shortcut, speak, press it again. Your words appear wherever your cursor is.

Get started

Mac (macOS 26 or later, Apple silicon or Intel)

1. Download the Mac ZIP from Releases, open it and drag Mouthy into Applications.
2. Open Mouthy. The welcome screens ask for the microphone and for Accessibility, which lets Mouthy type into other apps.
3. Double-tap the right ⌘, talk, and double-tap it again. ⌃⌥Space and Hold Fn are in Settings → Shortcut. Press Return while you talk to send what you've said so far; Mouthy keeps listening (it asks for Input Monitoring the first time).

Updates arrive on their own: Mouthy checks once a day, downloads a new version in the background and installs it when you quit.

Windows (64-bit)

1. Download the Windows ZIP from Releases, unzip it, right-click install-windows.ps1 and choose Run with PowerShell.
2. Press Ctrl+Alt+Space, talk, and press it again.

Mouthy isn't code-signed on Windows yet. If Windows says it protected your PC, choose More info → Run anyway.

Linux (x86-64, Wayland or X11)

1. Download the Linux .tar.gz from Releases, unpack it and run ./install.sh.
2. Press Ctrl+Alt+Space, talk, and press it again. Wayland doesn't let apps register shortcuts: install wl-clipboard and wtype, and bind a key in your compositor to mouthy --toggle.

To hear about new Windows and Linux versions, choose Watch → Custom → Releases at the top of this page.

Every engine runs on your own computer. A model downloads once, when you choose it, and then works offline. See Speech engines.

If something isn't right

• Nothing types (Mac). Turn Mouthy on in System Settings → Privacy & Security → Accessibility.
• A word keeps coming out wrong. On the Mac, add it in Vocabulary. Anywhere, another speech engine in Settings may suit your voice better.
• Anything else. Open an issue. A short description and your platform are enough.

Highlights

• Only your voice (Mac). Train your voice once by reading three short sentences, even with a TV on, and Mouthy turns down a TV, music and other people while you dictate. Mouthy keeps a voiceprint, never the recording.
• Types anywhere. Text goes into whatever app has focus, with spacing and capitals fitted to what is already there (Mac and Windows).
• Enter sends, Escape cancels. Press Enter while talking to send what you have said so far and keep going.
• Speech only types. Nothing you say can send, click or run anything in other apps.
• Spoken punctuation and fixes. "comma", "new paragraph", and "scratch that" to drop the last sentence.
• Modes. Different styles per app or website, for example plain text for chat and code formatting in your editor.
• Meetings. Records your microphone and the call's audio, then writes a transcript.
• For AI agents. Claude Code and other agents can ask you a question and hear your spoken answer.
• Notch hub (Mac). Hover the notch while you talk to watch your words appear. Timers, notes, music and your calendar live there too.

Full list: Features.

Privacy

• Dictation audio stays in memory and is never saved.
• No telemetry and no account. Mouthy goes online only to download the speech models you picked (plus a small voice-activity model on the Mac) and, on the Mac, to fetch lyrics, artwork and weather for the notch hub and to check once a day for a new version (Settings → Privacy → Check for updates automatically). Local Only Mode turns all of it off.
• History is off by default.

Details: Privacy and security.

Documentation

• Features: everything Mouthy does, per platform
• Speech engines: which engine to pick, languages and download sizes
• Privacy and security
• AI agents: connect Claude Code and other MCP clients
• Command line
• Building and testing

Contributing

Bug reports and pull requests are welcome. Read CONTRIBUTING.md first. Report security problems privately as described in SECURITY.md.

License

Copyright © 2026 LayerMaker LLC. Mouthy is free software under the GNU General Public License v3.0.

Speech models and third-party libraries keep their own licenses: Mac, Windows and Linux.