Typelite

OraType is a voice keyboard for macOS. Hold or tap a shortcut and speak. OraType turns your speech into text, removes fillers, applies your self-corrections, adds punctuation and pastes the result where you are typing. It can also translate what you say, or answer a question about the text you have highlighted.

Quick start

Download

OraType 1.0 runs on Macs with Apple Silicon (M1 or later).

1. Download OraType_X.Y.Z_aarch64.dmg from the latest release. To check it, download SHA256SUMS.txt too and run shasum -a 256 -c SHA256SUMS.txt --ignore-missing.
2. Open the DMG and drag OraType to Applications.
3. Start it once as described in First launch.

Build from source

You need Rust (rustup), Node.js 20 or later, the Xcode Command Line Tools and CMake.

git clone https://github.com/sennett-lau/oratype.git
cd oratype
npm ci
npm run build:app      # builds llama-server, then the app
open src-tauri/target/release/bundle/macos/OraType.app

More detail is in the Contributing guide.

First launch

1. Downloaded builds are not signed by Apple yet, so macOS blocks the first launch. Right-click OraType.app and choose Open (on macOS 15 and later: System Settings → Privacy & Security → Open Anyway), or run xattr -dr com.apple.quarantine /Applications/OraType.app once.
2. Allow Microphone and Accessibility when macOS asks. Accessibility lets OraType paste.
3. Follow the guided setup: microphone, speech, AI, then a short hands-on tutorial.

OraType is macOS first and runs best on Apple Silicon.

What makes OraType different

Language presets and a router

Each of your languages can carry its own writing instructions. Pick a preset (for example Cantonese (Hong Kong), English or Mandarin (Taiwan)) from the shared preset catalogue. Your edits are kept, and updates are offered, never forced.

When you dictate, the router works out which language you spoke. It uses the language that speech recognition detected, plus hint words a preset carries, such as Cantonese particles. Only that language's instructions are added, so English dictation never gets Cantonese rules.

Presets are small Markdown files. If you speak a language well, you can improve them: see Contributing a language preset.

Translate with quick language switching

Choose up to three target languages. While you speak, press the Switch key (⇧ Shift by default) or click the language name on the pill to change the target. The recording keeps going. Highlight text and press Translate to replace it with its translation.

Built-in setup in one click

On Apple Silicon, both halves can run on your computer with nothing else to install. whisper.cpp handles speech recognition inside the app, and llama.cpp's server handles AI polish. Press Set up and OraType picks models that suit your hardware, downloads them and checks each download against a SHA-256.

Bring your own server or key

Prefer a GPU computer on your network or a cloud key? OraType works with any OpenAI-compatible speech or chat server (whisper.cpp server, Speaches, Ollama, LM Studio, OpenAI, Groq, OpenRouter and more) and with the Qwen Cloud and ElevenLabs speech APIs. Cloud services are opt-in and use your own key. See the Guides.

How it compares with Typeless

OraType is inspired by Typeless, a polished commercial product.

	Typeless	OraType
Source	Closed	Open source (MIT)
Price	Free tier and paid plans	Free
Account	Sign-in required	None
Where speech and AI run	The Typeless cloud	Your computer, your own server, or a cloud service you pick
Languages	Built in	Presets from a public library, plus a router
Translate targets	Up to 3, switch from the pill	Up to 3, switch with a key or the pill while recording

Features

Feature	What it does	Default keys
Dictate	Speak, and clean text is pasted where you are typing.	Fn
Translate	Speak, and it is written in another language.	Fn + ⇧
Ask anything	Ask a question, or highlight text and say "make this shorter". Answers show in a panel above the pill.	Fn + Space
Switch language	Change the translation target while you speak.	⇧
Cancel	Stop the current recording or processing. Nothing is pasted.	esc
The pill	Shows recording, transcribing and polishing on the screen you are working on. It never takes focus.
Copy pill	When there is nowhere to paste, the result stays in the pill with a Copy button.
Presets	Save several speech and AI setups, switch between them, and export or import them.
Insights	Your speaking speed against your typing speed, time from speech to text, and how your presets compare.
Typing nudge	After a minute of steady typing, a small hint suggests saying it instead. At most once a day.

Keys are shown as macOS menus show them (⇧ is Shift). Every shortcut can be changed, and hold-to-talk is available too.

Screenshots

The images are rendered from the app's own components and the website's demos; see docs/media.

Documentation

Document	Covers
Guides	Choose a setup: built-in, your own server, or a cloud key
Speech recognition	Connections and supported services
AI polish	Connections, services and turning off thinking
Choosing models	Speech and polish models by hardware
Languages	Guides per language, translation languages, presets and recognition
Sharing presets	Export and import your speech and AI presets
Benchmarks	Polish and speech speed on real setups, and how to measure yours
Code performance	Application baseline, runtime evaluations, recorded decisions and performance PR workflow
Preset catalogue	Every language preset in the library
Plans	How features are designed before they are built

Privacy

• No accounts, no analytics, no telemetry.
• No history of what you say. Insights keep only timings and a keystroke count (never which keys). You can turn typing measurement off, and Settings → System → Clear insights data deletes them.
• Logs (~/Library/Logs/OraType/oratype.log) hold timings and errors, never your text.
• Audio goes only where you point it: the built-in models, your own server, or a service you chose.

Contributing

Code, services, language presets and docs are all welcome. Start with the Contributing guide.

Licence

MIT. Third-party components and models are listed in the Third-party notices.