MouthKeys/Straight voice to text for macOS
GitHub ↗ Download

Straight voice to text for macOS.

Hold a hotkey, speak, and the text lands where you were typing. On your Mac, free and open source, built to land every word.

v0.1.0 · macOS 15 or later · signed and notarized · GPL-3.0

The MouthKeys overlay mid-dictation: the live transcript above a voice trace, a 13-second timer, 43 words and 203 words per minute in the foot row, and the active mic.

01 / Listen

You talk to your computer all day now. Prompts for agents, replies, notes.

Dictation should be the fastest way in, but it fails quietly. The text lands in the wrong window. A terminal swallows it. The clipboard you were holding is gone. You find out when you look up.

The problem is not hearing you. The problem is delivery.

MouthKeys runs the speech model on your Mac and puts the words where you were typing when you stopped. When it sees a paste fail, it tells you and puts the words on your clipboard.

02 / What you get

01

Paste you can trust

When a paste fails in a way the app can detect, a card shows and the text goes on your clipboard. Your own clipboard comes back afterwards.

02

A faster stop

In a benchmark with a 13,600-entry history, stop-path work outside the model fell from 145 ms to 4 ms. A stalled model never loses the recording.

03

Built for terminals

Ghostty and c11 always get Reliable Paste. With Spoken Send on, say "send it" and Return follows, in c11 only in the pane you stopped in.

04

Your voice, live

A trace of your voice, a timer, a word count and words per minute while you speak.

05

History

Recent dictations, each one click from copy or reprocess.

06

Local engines

Parakeet, Nemotron, Cohere Transcribe, Whisper or Apple Speech, on your Mac.

03 / Privacy

TelemetryNone

Analytics are hard-wired off and the keys are blank.

AccountNone

Nothing to sign up for.

Your voiceLocal

Audio and text stay on your Mac unless you choose otherwise.

UpdaterOff

The app never phones home to update itself.

The app goes online only by your choice: to download the speech model you pick, for optional AI cleanup through a provider you set up (unsupported), and for Apple ASR Legacy, which lets macOS choose where speech is recognized.

04 / Built on FluidVoice

MouthKeys is a fork of FluidVoice by altic-dev, who did the hard part: the speech pipeline, the model integrations, hotkeys and typing into other apps.

We simplified it down to one job, removed all telemetry and the updater, and tuned it for speed on Apple Silicon Macs. MouthKeys is not affiliated with or endorsed by altic-dev.

Fig. 1 · The grin · a mouth full of keysHover to talk