Skip to content
Local-first dictation for macOS

You shout. We write.

Just say what you mean. Shout turns your speech into clean, ready-to-send text in any Mac app — spelling, punctuation, and polish handled.

Download for macOS View on GitHub

Apple Silicon · macOS 26+ · Free & open source (GPL-3.0)

100% on-device by default German & English Free & open source
How it works

Speak. It types. That's it.

Three stages, all on your Mac — from key-press to clean text in whatever app you're using.

Step 1

Hold fn & speak

Press and hold the fn key and talk naturally in German or English. Double-tap to go hands-free; Esc cancels anytime.

Step 2

Transcribed on-device

Your words become text instantly, right on your Mac — audio never touches the network or the disk. Under the hood: whisper.cpp (large-v3-turbo), Metal-accelerated on Apple Silicon.

Step 3

Polished & inserted

Apple Intelligence cleans it up on-device — removing "ähm"s, false starts, and fixing grammar — then Shout types the clean text wherever your cursor is.

The pill

One glanceable status pill.

A floating capsule, or docked to the Dynamic Island — with an optional live transcript while you speak. It also takes voice commands, so you can steer Shout without touching a key.

Switch profiles by voice.

Profiles change what Shout writes — clean it up, make it professional, summarize it, translate it to English. Pick one without touching a key:

  1. Begin with "use profile". The pill lists your profiles, numbered.

  2. Say a name or a number. It's forgiving — "professional" is enough, and "cancel" backs out.

  3. Keep talking. The command never lands in your text — the rest comes out in that style.

One breath works too: "use profile summarize, quick recap of today's standup…"

Say "use profile"
Shout showing a spoken profile list: Clean-up, Professional writing, Prompt engineer, Summarize, Translate to English
"Summarize" — switched, still listening
Shout's classic capsule switching to the Summarize profile by voice while recording
Features

Small menu-bar app. Serious dictation.

Everything you need to talk instead of type — and nothing you don't.

Push-to-talk & hands-free

Hold fn, or double-tap to lock hands-free recording. Esc cancels while recording, transcribing, or polishing.

German & English

Auto-detected as you speak, and fully fluent in both. Force a language in Settings if you prefer.

On-device speech-to-text

whisper.cpp large-v3-turbo, Metal-accelerated on Apple Silicon. Works fully offline once the model is downloaded.

AI cleanup

Apple Intelligence fixes fillers, false starts, and grammar — right on your Mac.

Local history

Every dictation (raw + cleaned) is kept locally and searchable. Optional, and never uploaded — turn it off or clear it anytime.

Glossary

Teach Shout the exact spelling of names, products, and jargon so they come out right every time.

Inserts anywhere

Types into any app via a concealed paste, then restores whatever was on your clipboard.

Pauses your music

Pauses Spotify or Apple Music the moment you start dictating, and picks it back up when you're done.

Bring your own model

Power users can route cleanup through any OpenAI-compatible endpoint — and give each profile its own model and prompt.

Privacy

Private by default.

Most dictation tools stream your voice to the cloud. Shout doesn't. Everything happens on your Mac unless you deliberately decide otherwise.

  • Transcription & cleanup are on-device. whisper.cpp and Apple Intelligence both run on your Mac.
  • Works fully offline. After the one-time speech-model download, dictation needs no internet connection at all.
  • History never leaves. It's stored locally and is never uploaded.

The one exception: if you connect a remote cleanup endpoint, that dictation's transcript is sent there to be polished. Shout marks it in the model list, under the picker, and on the pill — and your audio still never leaves the device.

Your Mac
Your voice
Transcribe + polish, locally
Clean text at your cursor
Get Shout

Download it, or build it yourself.

Requirements

  • macOS 26 or later, on Apple Silicon
  • ~2 GB free disk for the on-device speech model
  • Apple Intelligence enabled — optional, for the cleanup pass
  • Building from source? Xcode 26+

Download for macOS

Prebuilt app · Apple Silicon

Download
First launch
Shout is open source and self-signed — not notarized through a paid Apple account — so macOS asks you to confirm the first launch. Open System Settings → Privacy & Security and click Open Anyway, or clear the quarantine flag from a terminal: 
xattr -dr com.apple.quarantine /Applications/Shout.app

Build from source

Clone the repo and let the Makefile do the rest.

git clone https://github.com/Andreas-Menzel/shout-ai.git shout-ai
cd shout-ai
make setup   # whisper framework + speech model (~1.6 GB)
make cert    # optional: stable signing so permissions survive rebuilds
make run     # build build/Shout.app and launch it
Repository & README Contributing
FAQ

Questions, answered.

Is it really private?
Yes. Your audio is captured in memory only and never written to disk. Transcription and cleanup run on your Mac. The only network request Shout makes on its own is the one-time model download — unless you deliberately connect a remote cleanup endpoint, which is always clearly flagged.
Do I need Apple Intelligence?
Apple Intelligence powers the on-device cleanup pass — removing fillers and false starts and fixing grammar. Transcription itself works without it, and if cleanup is off or unsure, Shout inserts the raw transcript so dictation never blocks. You can also point cleanup at a different model or a remote endpoint in Settings.
Does it work offline?
Once the speech model is downloaded, yes — transcription and on-device cleanup run without a network connection.
Which languages are supported?
German and English, auto-detected as you speak. You can also pin a specific language in Settings.
Is it free? What's the license?
Free and open source under the GNU General Public License v3.0-or-later. You may use, study, share, and modify it; derivative works stay under the GPL.
Can I use my own cleanup model?
Yes. Cleanup can be routed to any OpenAI-compatible endpoint instead of the on-device default. When a model would send text off your Mac, Shout marks it in the list, under the picker, and on the pill — it never happens silently. Your audio still never leaves the device. You can also create your own profiles, each with its own specific prompt, to tailor cleanup to different kinds of writing.
What are the system requirements?
macOS 26 or later on Apple Silicon, plus about 2 GB of free disk for the speech model. Building from source additionally needs Xcode 26+.
Why won't it open on first launch?
Shout is open source and self-signed — not notarized through a paid Apple account — so macOS asks you to confirm the first launch. Open System Settings → Privacy & Security and click Open Anyway, or clear the quarantine flag from a terminal: 
xattr -dr com.apple.quarantine /Applications/Shout.app

Talk. Let Shout do the typing.

Local-first dictation for macOS — private, fast, and free.

Download for macOS View on GitHub