Your voice. Your machine. Your text.

Suniye is open-source dictation for macOS. Speech recognition runs locally — nothing leaves your machine. Hold one key, speak, release.

Install with Homebrew Recommended
brew install --cask kishanhitk/tap/suniye

or download the app directly

Suniye
A real, unedited take.

How it works

01

Hold your hotkey

Press and hold Fn/Globe — or any shortcut you choose. A small indicator pops up.

02

Speak naturally

Talk at your normal pace. Your voice is captured through your mic and stays on your Mac.

03

Text appears

Release the key. Your words turn into text on your Mac and appear at your cursor.

Built for real work

Runs on your Mac

Speech recognition happens entirely on your Mac. Choose from a built-in library of models — no internet needed to turn your voice into text. No server. No cloud.

Instant results

No trip to the cloud. Your voice becomes text in an instant — the model is always loaded and ready to go.

Works in any app

Your words appear right at your cursor — Mail, Slack, Notes, your browser, your editor. Everywhere you type.

AI cleanup, on your Mac

Optional cleanup tidies grammar and punctuation using Apple's built-in AI or a model on your Mac — no cloud needed. Or connect your own provider. Off by default.

One simple shortcut

Default is Fn/Globe. Change it to any key combo you like. Hold to talk, release to get your text. No menus, no buttons.

Open source, MIT licensed

Full source on GitHub. Read every line. No tracking, no surprises. Build it yourself if you want.

Choose your model

Ten speech models. Pick the trade-off you want — speed, accuracy, or languages.

Every model runs entirely on your Mac. Download more than one and switch instantly from the app — no restart, nothing sent to a server.

Parakeet TDT 0.6B v3 Recommended

25 European languages

Balanced Best ~680 MB
Parakeet TDT 0.6B v2

English

Balanced Better ~482 MB
SenseVoice Multilingual

Chinese · English · Japanese · Korean · Cantonese

Balanced Better ~240 MB
Moonshine Base

English

Fast Good ~285 MB
Whisper Large v3 Turbo

Multilingual

Fast Best ~564 MB
Whisper Distil Large v3

Multilingual

Balanced Best ~529 MB
Whisper Large v3

Multilingual

Slower Best ~1.7 GB
Whisper Small (English)

English

Balanced Better ~636 MB
Whisper Base (English)

English

Fast Good ~209 MB
Whisper Tiny (English)

English

Fast Basic ~118 MB

Magic Format

Cleanup that can stay entirely on your Mac.

Optionally tidy up punctuation, capitalization, and formatting after each dictation. It stays off until you turn it on.

01 On your Mac

Apple Intelligence

On your Mac · no setup

Uses Apple's built-in AI on supported Macs. Nothing leaves your machine, nothing to set up, no cost.

02 On your Mac

A model on your Mac

On your Mac · works offline

Runs a small AI model right on your Mac (about 3.4 GB). Download it once, then clean up text even with no internet.

03 Optional

Your own provider

Optional · uses the internet

Connect your own AI provider — like Gemini or GPT — with your own key. Only text is sent, and only if you pick this option.

Your audio is never sent anywhere. Magic Format only ever works on the text you dictated, and only the last option uses the internet.

Every word you dictate could be training data.

Unless it never leaves your machine.

The problem

Cloud dictation services process your voice on remote servers. Your private notes, medical records, business plans — all passing through infrastructure you don't control. That audio can be stored, analyzed, and used to train models without your explicit consent.

Suniye's approach

Speech models run directly on your Mac — pick one from the built-in library. Voice is processed in memory, transcribed locally, and never touches a network. The only network calls Suniye makes are to check for app updates and download models.

Even the optional cleanup can stay on your Mac, using Apple's built-in AI or a local model. Only if you pick the online option does your text (never your audio) leave the Mac — and it's off by default.

More in the box

Floating indicator

A draggable capsule shows live audio levels while you talk. Park it in any corner — or hide it when idle.

History & dashboard

Every dictation is saved on your Mac with word counts and timing. Copy or delete from a simple dashboard.

Any microphone

Built-in, USB, or Bluetooth — switch freely. Suniye tunes itself to each device automatically.

Automatic updates

New versions arrive on their own. Stay on stable releases, or opt in to early builds to try new features first.

Launch at login

Start Suniye quietly in the background so it's ready the moment you press your key.

Safe insertion

Your text lands at the cursor and your clipboard is kept intact. Optionally press Return to send automatically.

Learns from your edits

Correct a misheard name right after dictating and Suniye quietly adds it to your vocabulary — undoable, marked with a ✨, and easy to switch off.

How it compares

Local-first, by design.

Suniye Cloud dictation Built-in dictation
Audio stays on your Mac some
Choice of speech models
Works in many languages
AI cleanup that can run on your Mac cloud
Inserts into any app some
Open source
No tracking varies varies
Price Free Subscription Free

"Cloud dictation" and "built-in dictation" describe common categories — capabilities vary by product.

Questions

Is my dictation really private? +

Yes. Your voice is captured and turned into text entirely on your Mac — it never touches the internet. The only times Suniye goes online are to check for updates and to download a model the first time.

Does the optional AI cleanup see my audio? +

Never. Cleanup only works on the text, never the audio. By default it runs right on your Mac, using Apple's built-in AI or a local model. Only if you turn on the online option does your text get sent to the provider you choose — and it's off by default.

What languages can it understand? +

It depends on the model you pick. Some cover 25 European languages, others handle Chinese, Japanese, Korean, English, and Cantonese, and the largest understand a broad mix. Several English-only models are smaller and faster.

Does it work offline? +

Yes — once you've downloaded a model, dictation needs no internet at all. The only thing that ever goes online is the optional cleanup provider, if you choose to turn it on.

Apple Silicon or Intel? +

Both work for dictation. The on-Mac cleanup model needs an Apple Silicon Mac (M1 or later), and Apple's built-in AI needs a Mac that supports it. The online cleanup option works on any Mac.

How much space does it need? +

A speech model is anywhere from about 118 MB to 1.7 GB, so you can pick one that fits. The optional on-Mac cleanup model adds about 3.4 GB.

Can I install with Homebrew? +

Yes — it's the easiest way. Run brew install --cask kishanhitk/tap/suniye. Homebrew downloads the latest release, moves Suniye into your Applications folder, and clears the macOS quarantine for you, so there's nothing extra to do on first launch. Updates still arrive automatically inside the app.

Homebrew asks me to trust the tap — is that expected? +

Yes. Homebrew 6 asks you to approve any third-party tap before it runs, because a tap can run code on your Mac. Running the full command — brew install --cask kishanhitk/tap/suniye — trusts just this one cask. If you'd like to read it first, run brew cat kishanhitk/tap/suniye.

Why does macOS say the app is damaged? +

This only affects the direct .dmg download. Suniye isn't signed with a paid Apple developer certificate, so macOS quarantines it the first time. Run the command shown above to allow it — or install with Homebrew, which clears the quarantine for you. You can also build it yourself from the source on GitHub.

Is it free? +

Completely. Free and open-source under the MIT license. No subscription, no account, no tracking.

Try Suniye

Install with Homebrew, download the latest release, or build from source. Requires macOS 14 (Sonoma) or later. Pick a speech model from the built-in library on first launch.

Install with Homebrew Recommended
brew install --cask kishanhitk/tap/suniye

Alpha — expect rough edges