$ utter --listen
Say it out loud.
Utter does the typing.
Hold a key, speak naturally, and polished text appears at your cursor — in any Mac app. WhisperKit, Qwen3-ASR and MLX run locally, so your voice stays on your Mac. Remote providers are opt-in.
Free & open source · macOS 26+ · Apple Silicon · MIT
um so basically can you send the report to sarah by friday and um tell her i'll be on the flight to tokyo
Send the report to Sarah by Friday, and let her know I'll be on my flight to Tokyo.
$ utter --demo
See it in action.
Three short films: the feature tour, a day at the office, and local dictation. The interface is an animated reconstruction of the real app; sample texts are illustrative.
Feature tour
Smart format, voice command, translation dictation, industry lexicon, writing styles, and the local or remote models you can choose.
A day at the office
Replying to chat, meeting notes, an English email, a voice-drafted reply, and a weekly report on a train with no signal.
Local voice input
Hold fn, speak, and cleaned-up text lands at the cursor — recognition and cleanup both run on this Mac.
$ utter pipeline --verbose
From voice to final text in one key release.
- 01
Hold fn
fnA global hotkey starts capture without leaving the current app.
- 02
Speak naturally
··|·|··Talk like you would to a person. Audio is captured on your Mac.
- 03
Local ASR
</>WhisperKit, Apple Speech, or local Qwen3-ASR transcribes on Apple Silicon.
- 04
LLM cleanup
≡The selected model removes filler, fixes grammar, and formats for context.
- 05
Insert anywhere
↵The final text is pasted at the cursor, then your clipboard is restored.
so basically we need to ship the update this week and test it before friday
We need to ship the update this week and test it before Friday.
$ utter features --list
The features are the pipeline.
- ⌘
Global hotkey
Press and hold fn (or your key of choice) in any text field — Mail, Slack, Xcode, Notes. Hold, speak, release.
- ✎
Verbatim or smart format
Get the transcript exactly as spoken, or let the LLM polish it into finished prose. Your call, every time.
- ⇄
One-key translation
A second hotkey turns what you say into a translation in your target language — no app switching, no pasting.
- ◇
Voice command mode
Reply, summarize, proofread, format a table, or list action items by voice — with what's on screen as context.
- ⌗
Screen context OCR
Reads nearby on-screen text to better understand names, project terms, and homophones.
- ▤
Memory & personal dictionary
Correction learning, an industry lexicon, and input history keep names and phrases exactly the way you write them.
- ⚙
Smart model recommendations
Checks your Mac's chip, memory, GPU, and disk space, then highlights which models run well — and warns about those that won't.
- ∿
Streaming recognition (beta)
Partial transcripts appear while you speak — near-instant feedback on supported engines.
- ◎
Bilingual UI
The interface switches between English and 简体中文, and recognition covers many languages.
- ⇥
Inserts anywhere
Text is pasted at your cursor with a simulated ⌘V, and your previous clipboard is restored afterwards. Accessibility access lets Utter read the selection and nearby text for context.
$ utter hardware --check
Pick the right model for your Mac.
A curated catalog of Qwen and Gemma models in MLX quantized variants. Utter pre-checks your chip, RAM, GPU, and disk space, flags models that won't fit, and you can also point it at your own local MLX model folders.
- Qwen3.5 0.8B
- 2 GB RAM · 0.7 GB disk · any Apple Silicon
- Qwen3 1.7B
- 2 GB RAM · 1.0 GB disk · any Apple Silicon
- Qwen3.5 2B
- 4 GB RAM · 1.8 GB disk · M1 or later
- Gemma 4 E2B
- 6 GB RAM · 3.6 GB disk · 8 GB Mac or above
- Gemma 4 E4B
- 8 GB RAM · 5.3 GB disk · 10 GB+ Mac recommended
- Qwen3.5 9B
- 10 GB RAM · 7.1 GB disk · 16 GB Mac recommended
- Gemma 3 12B
- 12 GB RAM · 8.1 GB disk · 16 GB Mac recommended
- Qwen3 30B-A3B
- 12 GB RAM · 17.2 GB disk · 18 GB+ Mac recommended
- Qwen3.5 35B-A3B
- 16 GB RAM · 20.5 GB disk · 24 GB+ Mac recommended
Custom MLX model folders are welcome, and remote LLM providers in OpenAI / Anthropic format are there when you want them.
$ utter stack
Native, local, and built for Apple Silicon.
$ utter privacy --local
Private by default. Local when you want it.
WhisperKit transcribes on-device. MLX refines text locally. Remote providers exist for when you want them — and are off unless you turn them on. Nothing is uploaded unless you choose it.
- Audio upload
- 0 required
- Cloud calls
- 0 required
- Remote providers
- optional
$ utter faq
Questions, answered.
What is Utter?
A native macOS menu bar app that turns your voice into polished text in any app. Hold a hotkey, speak, and the result is typed directly at your cursor.
How is Utter different from Apple's built-in dictation?
Utter adds LLM cleanup, screen context via OCR, one-key translation, voice command mode, and global hotkey insertion into any app — all with local models by default.
Does Utter work offline?
Yes. With WhisperKit, Apple Speech, or local Qwen3-ASR plus a local MLX model, Utter transcribes and refines text entirely on-device. Remote providers are optional.
Which macOS version and hardware does Utter require?
macOS 26 or later on Apple Silicon. Intel Macs are not supported. Utter checks your chip, RAM, GPU, and disk space and shows which models are compatible.
Which apps does Utter work in?
Any macOS app that accepts pasted text. Utter pastes at the cursor with a simulated ⌘V and then restores your clipboard. Password fields and apps with Secure Keyboard Entry turned on may block it.
Which languages does Utter support?
Speech recognition supports many languages via WhisperKit, Apple Speech, and Qwen3-ASR. The app UI is bilingual in English and Simplified Chinese.
Is Utter free and open source?
Yes. Utter is MIT-licensed and the full source is on GitHub.
$ utter install
Get Utter from GitHub Releases.
- 01
Open the DMG
Grab the latest Utter-<version>.dmg from GitHub Releases and double-click it.
- 02
Drag to Applications
Drag Utter.app into the Applications folder shown in the DMG window.
- 03
Launch from the menu bar
Open Utter once from Applications. It lives in your menu bar from then on.
Only needed if macOS blocks the first launch: self-signed releases are not Apple-notarized. Developer ID releases open normally.