Talk to your Mac.
It actually listens.

Hold a key. Speak. Release. Your words appear at the cursor — like magic, except it's just really good math running on your GPU.

100% local. 100% free. 100% open source. Your voice never leaves your machine. Pinky promise.

Download for macOS View source

What it does

The good stuff.

Everything you need. Nothing you don't.

What happens on your Mac, stays on your Mac

Zero network calls. Zero telemetry. Zero analytics. Your audio never touches a server. We don't even have a server.

Blink and you'll miss it

Metal GPU on Apple Silicon. Sub-second transcription. VAD trims the silence so Whisper only crunches the good bits.

Types anywhere you can type

Terminal, VS Code, Xcode, Slack, Chrome, that one app only you use. If it has a cursor, we can type there.

Speaks fluent developer

500+ built-in terms. Say "kubernetes" and get "Kubernetes", not "Cooper Netties". Add your own names and jargon too.

Cleans up after you

Auto-capitalizes. Fixes 100+ term casings. Adds punctuation. Turns "three hundred" into 300. All in under 5ms.

Obsessively tweakable

Pick your hotkey. Pick your model. Pick your mic. Toggle every correction on or off. It's your app.

Friendly to your wrists

Push-to-talk for quick zaps, or toggle mode (press once to start, once to stop) for long dictations. Carpal-tunnel and RSI friendly — configurable hold prevents accidental activation.

Free, MIT, no asterisks

Open source. Audit it, fork it, ship it. No paywall, no upgrade prompts, no "free tier". Just an app.

Safety

Don't trust us. Verify it yourself.

Every privacy claim on this page is something you can check in under a minute.

What we don't do

✗ Send your audio anywhere
✗ Send your transcriptions anywhere
✗ Track, analyze, or log anything
✗ Phone home, ever
✗ Read your clipboard
✗ Load web fonts or external scripts
✗ Need an account
✗ Need an internet connection

Every network call in the entire app, in one grep:

$ grep -rE "URLSession|http://|https://" --include="*.swift" WhisperDictation/

ModelManager.swift: "https://huggingface.co/..."
ModelManager.swift: URLSession.shared.download(from: model.url)

# That's it. All of them. Only fires when you click Download Model.

Read the full privacy breakdown →

Alternatives

How we stack up.

Spoiler: really well for something that costs nothing.

WhisperDictation

Open source

✓ 100% local & private
✓ 500+ dev terms built in
✓ Custom names & vocabulary
✓ Works offline, always
✓ MIT licensed

Willow / WisprFlow

Commercial

$8-15/mo

✗ Cloud-dependent
~ Some dev vocabulary
~ Varies by plan
✗ Needs internet
✗ Closed source

Apple Dictation

Built in

✗ Sends to Apple servers
✗ Poor with tech terms
✗ No customization
~ Partial offline
✗ Closed source

Models

Pick your speed.

Quantized models. Smaller, faster, nearly identical accuracy.

⚡

Base Q5

57 MB~0.2sGood accuracy

🎯

Small Q5

181 MB~0.4sBetter accuracy

Recommended

🧠

Medium Q5

515 MB~1.0sBest accuracy

Speeds measured on Apple Silicon with Metal GPU. Download models in the app — Settings > Model.

Get started

90 seconds. Tops.

We timed it. Okay, maybe two minutes if you read slowly.

1
Download WhisperDictation.dmg from Releases
2
Open the DMG, drag to Applications
3
If macOS blocks it: System Settings > Privacy & Security > Open Anyway, or run xattr -cr /Applications/WhisperDictation.app
4
Grant Microphone + Accessibility permissions
5
Settings > Model > download Small Q5 (181 MB)

Or build from source

# Clone with whisper.cpp submodule
git clone --recurse-submodules \
  https://github.com/sam-pop/WhisperDictation.git
cd WhisperDictation

# Build everything and launch
make whisper && make model && make run

Talk to your Mac.It actually listens.