Getting started
- Open Settings and choose a provider: Built-in Local, Ollama, LM Studio, an OpenAI-compatible endpoint, or Chrome Translator for the browser extension.
- Pick a target language, then type, paste or share the text you want translated.
- Use local history to revisit, copy or remove recent translations. Nothing is uploaded when you do.
Choose the languages you use
On Mac, open Settings → Languages. Your selection appears in the input window, menu bar popover and Glossary.
On iPhone and iPad, open Settings → Target Languages. The app, Share extension, Translation extension and Glossary use the same list.
If you have never chosen target languages, the menu starts with the 10 interface languages. Enable any of the other targets in Settings. Your default target language always stays in the menu.
You can also choose your languages during first-run setup. On iPhone and iPad, the new three-step setup covers model downloads, permissions and target languages; it is skipped if a translation model is already installed.
Set up voice input
Voice input is here on Mac, iPhone and iPad. Whisper recognizes your speech on your device, then PrivyTranslate translates the text with your chosen provider.
Open Settings → Voice, download a speech model, and allow microphone access when prompted. Speech models are a separate, one-time download from translation models.
- macOS 15+. Whisper Small (about 488 MB) is free. Whisper Large v3 Turbo (about 574 MB), the most accurate option, is included with Professional.
- iOS / iPadOS 18+. Whisper Base (about 148 MB) is free. Whisper Small (about 488 MB), for higher accuracy, is included with Professional.
- Tap the microphone. On Mac, use the input window or menu bar popover. On iPhone and iPad, open the Translate tab.
- Speak, then tap again. Tap again to translate. Recordings stop automatically after 60 seconds. If you speak the target language, the target switches automatically.
- Audio stays with you. Recognition runs entirely on your device. Audio is never uploaded, and the recording is deleted immediately after recognition.
Press Esc on Mac or tap ✕ on iPhone and iPad to cancel. On iOS, moving the app to the background cancels recording; voice input is available in the app only, not in Share or Translation extensions. Recognition of Burmese, Khmer and Lao is experimental.
Voice recognition has its own language coverage: Burmese, Khmer and Lao are experimental. Whisper does not recognize Zulu; selecting it for voice input falls back to automatic language detection.
Provider connection issues
Work down this list in order — the first step rules out the app itself, so everything after it is a network or configuration question.
- Built-in Local. Try this first. It needs no network and no configuration, so if it works the translation interface is fine and the problem is in the provider you had selected.
- Ollama or LM Studio. Confirm the server is running, that the address you configured is reachable, and that the model you selected has actually been pulled.
- Mobile devices. The phone and the machine running the model server must be able to reach each other on the same network. A guest or isolated Wi-Fi network will block this.
- Chrome extension. Overlay, Replace and Side-by-Side each translate the page differently — try another mode if one behaves oddly on a particular site. Chrome Translator additionally requires a Chrome build where the Translator API is available.
- OpenAI-compatible endpoints. Verify the endpoint URL, the credentials and the exact model name with your provider. A wrong model name usually surfaces as an immediate error rather than a timeout.
Platform availability
Privacy
PrivyTranslate collects nothing: no translation text, no prompts, no identifiers, no usage analytics. Text either stays on your device in Built-in Local mode, or goes directly to the provider you configured — never to us. History, glossary entries and API keys stay on the device.
Feedback and bug reports
Tell us what broke, what you expected, and which provider you had selected — that combination is usually enough to reproduce a problem. Feature requests are welcome through the same form.
Send feedback or report a problem
For anything you would rather not put in a form, email [email protected].