Think out loud.
Larynx writes it down.

Free dictation for the Mac that hands back what you meant.

Larynx is a Mac app. Open this page on your Mac to install it.

curl -fsSL https://raw.githubusercontent.com/StopBanning/larynx/main/scripts/install.sh | zsh

Paste it into Terminal on an Apple Silicon Mac with macOS 15 or later. Nothing you say leaves your Mac.

Can you move the standup to Friday?

Said with an um, an uh, a repeat and a change of mind. Written in 0.15 s.

It shows its work

Every number here comes from a test you can run yourself.

Speed

From letting go of the key to the text, the full pipeline.

The speed test

0.15 s

Words wrong

On 57 real dictations picked at random.

The benchmark

0.87%

Sent anywhere

Speech to text runs with the network blocked. Nothing you say leaves your Mac.

The privacy test

0 bytes

What it needs

Apple Silicon
M1 or newer. The speech model runs on the Neural Engine.
macOS 15
Or later.
About 600 MB
For the speech model, downloaded once.
No account
No subscription, no word limits, no ads.

Questions

Is it really free?

Yes. MIT licence, no account, no paid tier, no word limits and no ads. The code is on GitHub. The default speech model, Parakeet Ultra, is CC BY 4.0; the credits list every model and its licence.

Why does it need Accessibility?

To put your words where your cursor is: Larynx pastes into the app you’re using and checks the text arrived. Input Monitoring lets it see the talk key, and the microphone hears you. To spell names right it also reads the text near your cursor when you press the key; that stays in memory for that one dictation, is never kept, and can be turned off. Details in PRIVACY.md.

Does it work offline?

Yes. After the speech model downloads once, everything runs on your Mac’s Neural Engine. The privacy test turns the network off and transcribes anyway.

Intel Macs?

No. Larynx needs an Apple Silicon Mac (M1 or newer) with macOS 15 or later, because the speech model runs on the Neural Engine.

Windows?

Not yet. The cleanup engine already passes its tests on Windows and the speech model runs there; the app is planned.

Why is there no download button?

A signed download needs a paid Apple developer account. Until then the one line above builds the same app from source on your Mac, in a few minutes the first time. Run it again to update.

Try it on your Mac.

curl -fsSL https://raw.githubusercontent.com/StopBanning/larynx/main/scripts/install.sh | zsh