Free & open source — part of the Weedlabs suite

Type by voice.
Locally. Privately. Free.

TypeFlow turns your voice into text on macOS — no cloud, no account, no monthly fee. Press a key, speak, paste anywhere. Whisper runs entirely on your Mac.

Download for macOS Star on GitHub

macOS 13 or later · Apple Silicon recommended · ~500 MB

Ready
Press \ to record
What's inside

Everything you need. Nothing you don't.

Built on open-source Whisper. No subscription, no telemetry, no API key.

Press, speak, paste

A configurable global hotkey starts recording from any app. Release and your transcript is on the clipboard, ready for ⌘V.

100% on your machine

Whisper runs locally via faster-whisper. No audio leaves your Mac. No account. No telemetry. Works on a plane.

99 languages, one model

Whisper natively supports nearly every language. English & Bahasa Indonesia exposed in the UI — code-switching works.

Pick your accuracy/speed tier

From a 240 MB small model that turns words into text in under a second, up to large-v3-turbo for near-perfect transcripts.

Optional AI polishing

A tiny local LLM (Qwen 2.5) can rewrite your transcript in a more fluent, friendly, or professional tone — still on-device.

Open source, MIT licensed

Every line of code is public. Fork it, audit it, modify it, ship your own version. The source on GitHub is the product.

How it works

Three steps. Less than two seconds.

No setup. No prompts. No "Did you mean…"

Press your hotkey

Default is ⌘ \. Rebind to anything you want from inside the app — TypeFlow listens system-wide.

Speak naturally

The orb pulses with your voice. Mix English and Indonesian freely. Pause when you need to. Press the key again to stop.

Paste anywhere

Your transcript is already on the clipboard. ⌘ V in Slack, Notion, Gmail, your terminal — wherever you were just typing.

Pick your tier

Trade size for accuracy. Your call.

All models run locally on Apple Silicon. Downloaded once, then offline forever.

Model Disk Speed Best for
Small 240 MB ~0.5s Quick notes, English-heavy speech Default
Medium 1.5 GB ~1s Daily use, mixed-language
Large v3 Turbo 1.5 GB ~1.5s ChatGPT-grade transcription Recommended
Large v3 3 GB ~3s Long meetings, max accuracy
FAQ

Answers, before you ask.

Is it really free?
Yes. MIT-licensed open source. There's no paid tier, no usage limit, no "pro" version. Models download from Hugging Face on first use and stay on your machine forever.
Does my voice get sent to a server?
No. All transcription runs locally via faster-whisper. The only network activity is the one-time model download. Set HF_HUB_DISABLE_TELEMETRY=1 in your env if you want to suppress even that detail.
Why is the app unsigned?
Apple charges $99/year for a Developer ID, which isn't reasonable for a free hobby project. The first time you open TypeFlow, right-click → Open instead of double-clicking, then click Open in the warning dialog. After that, it opens normally.
Does it work on Intel Macs?
Probably, but it's only been built & tested on Apple Silicon. The Whisper models will run, just slower. The optional LLM polishing step requires mlx-lm which is Apple Silicon only.
What about Windows or Linux?
macOS-only for now. The UI is built on native AppKit (rumps + PyObjC). Cross-platform would mean rewriting the menu bar / global hotkey / overlay layers — meaningful work, not in scope yet.
How do I rebind the hotkey?
Open the TypeFlow window, click the hotkey button at the bottom, then press whichever key combination you'd like. Escape cancels. The new binding is saved to ~/.typeflow.json.
Can it translate?
Whisper transcribes in the source language. For "voice in English → text in Indonesian" (or vice versa), enable the optional ToneTranslate to… option, which runs a small local LLM after transcription.
I want a feature / found a bug.
Open an issue on the GitHub repo. PRs welcome — see README.md for the build instructions.

Stop typing. Start talking.

Download TypeFlow, grant a few permissions, and you'll be back to your terminal in 90 seconds.