Dictation for Apple Silicon Macs

Say it messy.
Send it clean.

Hold a key and talk the way you think. When you let go, Herga drops the ums, applies your “no, actually”s and puts clean, ready-to-send text into whatever app you’re in. None of it leaves your Mac.

Free and open source · Apple Silicon · No account

How it works

Hold. Speak. Let go.

  1. 1
    right ⌘ right ⌥

    Hold the chord

    From any app, anywhere. A small pill shows you’re being heard. Tap Space to keep going hands-free.

  2. 2

    Talk like you think

    Ramble, restart, change your mind. Herga transcribes while you’re still speaking, so there’s little to wait for.

  3. 3
    ✓ Pasted into Slack

    Let go, it’s there

    Clean text lands in the field you started in, even if you’ve clicked somewhere else since.

What you say → what you get

It knows what you meant.

Speech is faster than typing, but it comes out full of detours. Herga cleans them up without summarizing or putting words in your mouth.

Changes of mind

“send it Tuesday, no actually Wednesday”

Send it Wednesday.

Lists

“make a list of flights, hotel and the rental car”

- Flights - Hotel - The rental car

Spelling

“my username is m, r, g, n, capital H”

My username is mrgnH

Symbols

“wrap it in open bracket x close bracket”

Wrap it in [x]

Line breaks

“thanks for the help new paragraph Morgan”

Thanks for the help. Morgan

Your clipboard

“here is the link paste from clipboard”

Here is the link: https://…

Private by design

Your voice never leaves the room.

There is no Herga server to send your audio to. Speech recognition, cleanup and everything learned from your corrections run on your Mac’s own chip and stay in a folder you can open.

  • No account, sign-in or subscription
  • Works offline once your models are downloaded
  • No analytics, telemetry or crash reporting
  • Recordings and history stay on disk, and you choose how long to keep them
  • Open source, so you can check every line
NativeBuilt with Tauri and Rust, not Electron. The hotkey, microphone and paste run in native code.
StreamingRecognition and cleanup start while you speak, a sentence at a time.
On-deviceWhisper and Qwen3 run on the Apple Silicon GPU through MLX.

Open source

Built in the open.

Herga is free, MIT licensed and developed on GitHub. Star the repository to follow along, and tell us when something doesn’t work the way you expect. Every report makes dictation better.

Questions

Good to know.

More in the docs.

Does anything I say leave my Mac?

No. Recording, transcription, cleanup and learning all run locally. Your audio and text never leave your Mac. Herga only goes online to download its speech and language models from Hugging Face, and to check GitHub for new versions.

How is this different from macOS dictation?

Built-in dictation types what it hears. Herga cleans it up the way you’d have typed it: filler words and false starts go, spoken corrections are applied, and punctuation follows your habits. It also learns your vocabulary and the fixes you make.

Will it rewrite what I meant?

No. Cleanup keeps your words and your meaning. It never summarizes and never adds anything you didn’t say. If you want a real rewrite, that’s what Command Mode is for.

Which Macs does it run on?

Macs with Apple Silicon (M1 or later). Whisper and the cleanup model run on Apple’s MLX framework on the GPU, which is what keeps it fast without the cloud.

Does it work in every app?

It works in any app with a text field: chat, email, docs, browsers, editors and terminals. Herga writes into the field you were in when you started. When it has to paste, it puts your clipboard back afterward.

What languages does it understand?

English, Spanish, French, German, Japanese, Chinese and Hindi, or let it detect the language for you.

What does it cost?

Nothing. Herga is free and open source under the MIT license. There’s no account and no subscription.

Stop typing everything.

Free, with no account to make, and every word stays on your Mac.