Open source · Free · Local-first dictation for Mac

Speak. It writes.

Scribe puts Whisper inside your Mac. No account, no subscription. Use your voice to write emails, notes, code, and messages — anywhere there is a cursor.

macOS 14+ MIT License No cloud transcription

Hold Fn and speak. Release, and the text lands at your cursor.

Works inside the apps you already use

Notes Mail Messages Slack Telegram Gmail Arc VS Code Cursor GitHub Raycast ChatGPT Claude Notion Obsidian Linear

Everywhere

It is not another input method — it is a voice layer you can use anytime.

Scribe does not ask you to switch into a special editor. It does one thing: it pastes what you just said, cleanly, at the current cursor.

Push to talk

Hold Fn anywhere

No window switching. No recording panel. Wherever the cursor is, that is where your words appear.

Local first

Whisper runs on-device

After the model is downloaded, inference happens offline. Audio never leaves the machine — write emails, code comments, or private notes safely.

Multilingual

Mixed languages, no problem

Built for bilingual workflows. Handles English jargon, product names, and on-the-fly language switches without losing context.

Fast 140 MB
Balanced 210 MB · current
High Quality 600 MB

Designed for flow

Less interface, more momentum.

A small menu-bar presence, models you download on demand, a global hotkey. You stop noticing it — until you realize you no longer want to type long sentences.

Privacy by default

Your voice never leaves this Mac.

Audio is used locally, just for the current transcription. No accounts. No telemetry. Nothing sent to a third-party service to come back as text.

01 Audio stays local

Recording is used only for this transcription.

02 No account

No user IDs, no subscriptions, no cloud workspace.

03 Offline after setup

Once the model is downloaded, it runs offline.

Tiny surface

Just one small capsule, to show it is listening.

Open source

You can audit how it works — and fork your own version.

Scribe is MIT licensed. For a tool that touches your voice input, "you can read the code" matters more than "trust us".

Open the GitHub repo

FAQ

A few things you might want to know first.

Does it need an internet connection?

Only to download the Whisper model the first time. After that, recognition and pasting all happen on your Mac.

Why the Fn key?

Fn sits in a comfortable spot on a Mac keyboard, rarely conflicts with other shortcuts, and feels right as a hold-to-talk trigger.

Does it work on Intel Macs?

The project targets macOS 14+. Apple Silicon is the better experience, especially for on-device model inference speed.

How do I uninstall?

Delete /Applications/Scribe.app. To clear the models too, remove ~/Library/Application Support/Scribe.

Give your keyboard a break.

Download, grant mic and accessibility access, then hold Fn and start speaking.