Justspeak.
Free, open-source voice-to-text for Mac.
No subscription. No cloud. No account. Ever.
macOS 14+ · Apple Silicon · MIT License
How it works
Three steps. Zero friction.
No account. No API key. Install, press your shortcut, go.
Press your shortcut
Tap to toggle, or hold for push-to-talk. Any shortcut you want.
Speak naturally
Talk at your normal pace. Voxtral runs on your Mac. Nothing leaves your machine.
Hey team, I just pushed the latest changes to the feature branch. Can someone review the PR when they get a chance?
Text appears instantly
Clean text drops into whatever app you're using. Emails, code, notes. No hands needed.
Two modes.
Your choice.
Exact transcription when you need every word. AI-polished output when you want clean prose. Switch anytime.
“Um, so, I've been thinking and, you know, we should probably refactor the entire auth module because it's, like, getting really messy. And, uh, the tests are basically broken at this point. So, yeah, we need to fix that too before the next release.”
Um, so, I've been thinking and, you know, we should probably refactor the entire auth module because it's, like, getting really messy. And, uh, The tests are basically broken at this point. So, yeah, We need to fix that too before the next release.
Features
Everything they charge for. Free.
Native Swift & MLX
Not a wrapper. Petal runs Voxtral natively through Apple's MLX framework, using the GPU and Neural Engine on your Mac.
Smart transcription
Verbatim captures every word. Smart mode strips the filler and tightens your phrasing. On-device, either way.
um, so basically I was thinking that we should probably like refactor the authentication module because it's getting kind of messy and hard to maintain
I think we should refactor the authentication module. It's getting messy and hard to maintain.
Two input modes
Hold to record, or tap once to toggle. Pick whichever fits your flow.
Push to talk
Hold to record
Toggle mode
Tap to start & stop
Auto-paste anywhere
Text lands directly in whatever app you're using. Speak, release, done.
100% offline
Your voice never leaves your Mac. No server, no account, no internet required. Ever.
Models
One model, three sizes
Voxtral running natively on Apple Silicon via MLX. Three sizes so you pick the trade-off that fits your machine.
Voxtral Mini 3B
Full precision. Maximum accuracy. Pick this if you have the RAM.
Voxtral Mini 3B
Half the size. Nearly identical quality. The right pick for most people.
Voxtral Mini 3B
Under 2 GB. Runs on any M-series Mac. Still handles everyday dictation with ease.
Agentic coding
Talk to your agents
Use Petal with Cursor, Claude Code, Amp, Codex, or any agentic coding tool. Speak your intent, skip the typing.
Works where you code
Drive an entire fleet of agents faster than you can type. Petal pastes directly into any terminal or editor.
More context, less effort
Talk through your change with full nuance. Let agents one-shot your next feature from a spoken prompt.
~ claude
Claude Code v5.2
Opus 4.6 · ~/Developer/project
> Refactor the auth module to use JWT tokens instead of session cookies. Update the middleware and add refresh token rotation.
transcribed by petal · 4.2s of audio
Works everywhere you type
Petal transcribes your speech and pastes it directly into the frontmost app. No copy-paste needed.
Free and open source
Other apps charge $8/month for voice-to-text. Petal is free. The code is public. The models are on Hugging Face. Inspect it, improve it, or ship your own version.
Support
Frequently asked
questions
Start typing less.
Download Petal, pick a model, start talking. Up and running in under two minutes. No account. Free forever.
macOS 14+ · Apple Silicon required · MIT License