OneSentence

A free, open-source macOS menu bar utility for local speech-to-text and text-to-speech. Dictation and playback run on your Mac. No account is required.

Install from GitHub Releases View source User guide

Requires macOS 13.3 or later on Apple Silicon. First-run model download needs about 1 GB of disk space. Whisper is optional and downloads only when you tap Download on a model.

License

What	License
Application source	MIT
Whisper weights	MIT
NVIDIA Parakeet weights	CC BY 4.0
Supertonic weights	OpenRAIL-M (use restrictions apply)
Material Design product sounds	CC BY 4.0

See License, LICENSE, NOTICE, and CREDITS.md. A notarized DMG is a mixed-license bundle, not MIT-only software.

Install

Download the latest notarized build from GitHub Releases.
Open the DMG and drag OneSentence to Applications.
On first launch, grant Microphone and Accessibility permission, then let the app download models.

The app checks for updates via Sparkle at https://onesentence.app/appcast.xml.

What it does

Listen (⌘5): record, transcribe on-device with Parakeet (default) or Whisper, copy text to the clipboard, and paste into the active field.
Read Selection (⌘6): speak selected English text with Supertonic voices.
Snippets (⌘⇧Space): fuzzy-search plain-text files from a folder you choose.
Optional local history: search past transcriptions stored on your Mac.

Details: User guide.

Privacy

Speech recognition and synthesis run on your hardware. The app does not send your voice, transcripts, or clipboard content to us. Network use is for model download (Hugging Face) and Sparkle updates. There is no in-app analytics SDK. Full policy: Privacy.

Build from source

Requirements: macOS 13.3+, Xcode 16+, Mise.

git clone https://github.com/snowy-0wl/onesentence-app.git
cd onesentence-app
mise run setup
mise run tg          # generate Xcode workspace (Tuist)
mise run build       # Debug build
open OneSentence.xcworkspace

Tests: ./scripts/dev/test.sh --tier 1. See CONTRIBUTING.md and AGENTS.md.

Models

Runtime ML weights are hosted on Hugging Face: snowy-0wl/onesentence-models. The app downloads them on first use. The dataset includes a NOTICE with third-party weight license summaries.

For local development:

git lfs install
huggingface-cli download snowy-0wl/onesentence-models --local-dir deps/assets/onesentence-models

This is a personal project

I built this for myself and published it in case it is useful. I am not looking for pull requests, feature requests, or a support queue. Forks are welcome. See CONTRIBUTING.md.

Questions

Is the whole product MIT?

No. Application source is MIT. Whisper, Parakeet, and Supertonic weights, plus Material sounds, keep upstream licenses. Do not treat the DMG or the Hugging Face dataset as MIT-only.

Does it work offline?

Yes, after models are downloaded. Dictation and playback do not require a network connection.

Where do I get help?

See the user guide and Support. This is a personal project; there is no support SLA.

OneSentence is a software utility, not a medical device. Local models can make mistakes.