Dictation
Hold, talk, release. Text lands at the cursor. Tap instead to latch it on.
mac + windows / open source / runs offline
Hold a key. Say the thing. Send it without editing.
Transcribed on your own machine. Cleaned up before it lands. No account, no upload, no subscription.
- Transcription
- On device
- Audio uploaded
- None
- Price
- Free, MIT
- Requires
- macOS 14+ or Win 10+
01 / the gap
You already know
talking is faster.
3× faster. Always has been.
So why do you still type? Because raw speech-to-text hands you something you would never send to another person.
What you get
um so i was thinking uh maybe we could push the launch to next week
What you want
I was thinking we could push the launch to next week.
That gap is the entire product.
02 / the part nobody shows
Watch it clean up
after you.
Keep scrolling
Are we still shipping Thursday?
Message
✕
✓
Raw transcript
What FreeFlow inserts
03 / everything in the box
Nine things, one app.
Commands
Select text, say “cut this in half”. The rewrite replaces it.
Meetings
Records both sides off your Mac's audio. No bot joins the call.
Style
Slack and Pages shouldn't sound the same. Per-app rules.
Vocabulary
Your name, your colleagues, the words it keeps mangling.
Snippets
Say two words, insert a paragraph. Expanded before cleanup runs.
Scratchpad
Always open. For the thought with nowhere to live yet.
Insights
How much you said, how fast, and where it went.
Your key, your call
fn by default. Chords are ignored, so ⌘C still copies.
fnright ⌥right ⌃right ⇧
04 / it never leaves
Your voice does not
go anywhere.
The model runs on your machine. There is no account, because there is nothing to log in to.
- Mic16 kHz mono
- Whisperon device
- Your wordsat the cursor
edge of your Mac
Turn cleanup off and it is fully offline
One switch and it makes zero network requests. Pull the wifi. It still works.
You can read every line of it
All of it on GitHub under MIT. No credentials in the tree.
Cleanup uses a key you own
Bring your own Groq key. Free tier, no card. Or skip it entirely.
05 / get it running
Download it on Windows.
Build it on a Mac.
Windows lets you run software it hasn't been paid to trust. macOS doesn't, and a warning-free Mac app needs a paid Apple account this project doesn't have.
First build / unsigned / 0.1.0
FreeFlow for Windows
Windows 10 or 11, 64-bit. Hold Right Alt, talk, let go. Dictation and cleanup; the meeting recorder is Mac-only for now.
Download the installerNothing is code signed, so Windows will say "Windows protected your PC" the first time. Click More info, then Run anyway. A certificate costs a few hundred dollars a year, and this is free.
- Run the installer. FreeFlow lands in your system tray. No account, nothing to sign up for.
- Let it fetch the model. About 500MB of speech model on first launch, once. The panel shows progress.
- Allow the microphone when Windows asks. If you miss it: Settings, Privacy, Microphone.
- Hold Right Alt and talk. Let go and the text lands where your cursor is. Double-tap to lock it on.
Didn't end with the app open?
If the last step opened a folder, or Terminal said no such file or directory: ./install.sh, you're in the wrong folder or on an older copy of the code. Install Xcode from the App Store and open it once, then paste this single line into Terminal. It fetches a fresh copy, builds it and opens the app.
Full detail, including how to stop macOS forgetting your permissions on every rebuild, lives in the README.
Rather build it yourself? The source, the licences and the rough edges are in the Windows README.
06 / straight answers
Things you
were going
to ask.
Is it actually free, or free for now?
Free. MIT, no paid tier planned. Cleanup runs on Groq's free tier, which doesn't ask for a card.
Does my voice get uploaded?
No. Whisper runs locally. With cleanup on, only the text goes to Groq. Turn it off and nothing leaves.
Why is there a download for Windows but not for Mac?
Gatekeeper-free distribution costs $99/yr. Windows just warns you once and lets you through, so there's an installer. Building the Mac app takes ten minutes.
Is the Windows version the same app?
Same trigger, same local transcription, same cleanup. Dictation only so far: the meeting recorder, commands and snippets are Mac. It's also new, so expect rougher edges.
How is this different from the dictation already in macOS?
Apple's transcribes. It doesn't punctuate, cut fillers, learn names, or change tone per app. That pass is the point.
Will it fight with my existing shortcuts?
No. The event tap observes without consuming, and a trigger only counts when held alone.
Does it work in Electron apps?
Yes. Accessibility write first (clipboard untouched), falling back to ⌘V. Slack, Discord, VS Code.