🇬🇧 English | 🇰🇷 한국어 | 🇨🇳 中文 | 🇯🇵 日本語

OST — On-Screen Translator

OST is a menu bar app for macOS 26+ that transcribes system audio and shows on-device translation subtitles in a floating overlay. It is designed for videos and meetings, where disappearing, blinking, or duplicated text is a serious usability failure.

Screenshots

Combined overlay

General settings

Highlights

• Apple Speech and Apple Translation are the default on-device providers.
• Apple Speech supports English, Chinese (Simplified and Traditional), Japanese, and Korean when the corresponding system locale is available.
• Optional Qwen3 ASR and translation models run locally with MLX.
• Confirmed transcript and translation history stay visible while two separate lines are reserved for the current preview.
• Each transcript and translation area supports 2–10 confirmed lines; the window resizes when this value changes.
• Left alignment is the default, with center and right alignment available.
• Transcript, confirmed translation, and current translation preview have separate font/color controls and live examples.
• EPD is used as a soft translation boundary and does not force a new visible line for every pause.
• The app UI supports English, Chinese, Japanese, and Korean. English is the default.
• Optional per-session transcript and translation files are off by default and are saved only to a folder selected by the user.

Privacy

Audio, transcripts, and translations are processed on the Mac. OST has no feature that uploads or sends audio, transcripts, translations, settings, or saved session files outside the computer.

Internet access is used only when the user chooses to download an MLX model. The sandboxed downloader receives model files; user content is never passed to it.

Requirements

• macOS 26.0 or later
• Apple Silicon

Install 0.3.1

Download OST-0.3.1-macos-arm64.zip from the v0.3.1 release, extract it, and move OST.app to Applications.

The current binary is ad-hoc signed because a Developer ID certificate was not available in the release environment. If macOS quarantine reports that the app cannot be verified or is damaged, remove the quarantine attribute after confirming that you downloaded it from this repository:

xattr -dr com.apple.quarantine /Applications/OST.app

The release ZIP SHA-256 is published by the release workflow and can be compared with:

shasum -a 256 OST-0.3.1-macos-arm64.zip

Start transcribing

1. Launch OST and allow System Audio Capture when macOS asks.
2. Select the OST captions icon in the menu bar.
3. Choose the input and target languages, then choose Start.
4. Use Settings > Overlay to unlock, move, resize, or change the number of confirmed lines.

Microphone input

In Settings > General > Microphone, select Off, Transcribe only, or Transcribe and translate. Restart capture after a change.

System audio stays enabled. Both inputs use the selected input language and providers. macOS requests microphone permission only when you enable microphone input.

In Settings > Appearance > Microphone subtitles, choose a separate text color. Microphone transcripts, translations, and previews use this color and a microphone symbol. Each enabled input has its own two-line preview. Saved microphone text uses the [Microphone] prefix.

Capturing both inputs uses more memory and processing time. Use headphones to avoid transcribing speaker audio twice.

Keyword boosting

Keyword boosting gives Qwen3 ASR preferred spellings for spoken terms. In Settings > Models, select Qwen3 ASR as the transcription provider. Under Keyword boosting, enter one short term per line. Restart capture after changes. Clear the list to disable the hints.

OST stores up to 4,000 characters locally and uses the first 100 unique terms. Qwen3 ASR uses these hints for confirmed transcripts. Live previews and Apple Speech do not use them. Hints do not guarantee recognition or force text replacements.

Optional session files

Open Settings > Privacy, enable Save each session to text files, and choose a folder. One capture start-to-stop interval is one session. OST creates separate timestamped transcript and translation files. This option is off by default and never saves audio.

Build and test

The project uses SwiftPM plus an Xcode app/XPC project.

swift package --disable-sandbox resolve
swift test --disable-sandbox --disable-automatic-resolution --jobs 4
script/test.sh
script/build_and_run.sh run

See docs/manual-qa.md for the release checklist.

License

MIT