Your words, any language, zero cloud. Instantly.
A privacy-first, AI-powered desktop translator that lives in your menu bar.
Press a hotkey → type → get your translation → send it. All offline. All local. All yours.
You're chatting with a colleague in Tokyo. Mid-conversation, you need to reply in Japanese — but your Japanese is… creative at best. You could open a browser, navigate to some translation site, copy-paste, wait for the response, copy-paste again…
Or you could just press Ctrl+Alt+T.
LiveLingo pops up as a sleek floating overlay — right where you're working. Type your message, get an instant AI translation, and hit Send. The translated text lands directly into the app you were using. No tab-switching. No copy-paste gymnastics. No data leaving your machine.
💡 Think of it as having a personal interpreter sitting inside your keyboard.
Your text — and your voice — never leaves your computer. LiveLingo runs local AI models for both translation and speech recognition: Gemma 4 26B-A4B MoE (Mixture-of-Experts, ~26 B-class quality at ~4 B-class latency) for translation, and sherpa-onnx (Cohere Transcribe 14-Lang or SenseVoice CJK-tuned) for speech-to-text. No API keys, no cloud services, no "we may use your data to improve our products" surprises.
Translation in ~300ms. The overlay appears instantly with a global hotkey. Type, translate, send — all in one fluid motion.
Chinese, English, Japanese, Korean, French, German, Spanish, Russian, Arabic, and Portuguese. Mix and match source and target languages freely.
Two modes to fit your workflow:
- Paste Only — drops the translation into your clipboard
- Paste & Send — pastes AND hits Enter for you (perfect for chat apps)
Talk to LiveLingo instead of typing. Powered by sherpa-onnx with two interchangeable engines:
- Accuracy First — Cohere Transcribe (14 languages, top of the Open ASR Leaderboard, ~1.6 GB)
- Multilingual First — SenseVoice Small (中 / 粤 / 英 / 日 / 韩, on-model language detection, ~228 MB)
Switch engines anytime from Settings → Speech — the active model card shows exactly what will run, with a one-click jump to the Models tab when a download is missing.
A frosted-glass overlay that floats above everything. Drag it, resize it, adjust its transparency from 10% to fully opaque. The text automatically adapts contrast based on your opacity setting — readable on any desktop background.
Native experience on both Windows and macOS. Platform-specific hotkey hooks, text injection, and UI polish on each.
Built-in update system via Velopack keeps you on the latest version without lifting a finger.
|
Translation Overlay 
|
Long Text Translation 
|
|
Settings — General 
|
Settings — AI Models 
|
Download the latest release for your platform:
| Platform | Format | Download |
|---|---|---|
| Windows | .exe (auto-update) |
Releases |
| Windows | .msi (traditional) |
Releases |
| macOS | .pkg |
Releases |
- Download the AI models — LiveLingo will prompt you on first run. Translation needs Gemma 4 26B-A4B MoE (~17 GB; falls back to Gemma 4 E4B ~5.4 GB on low-RAM devices). Voice input is optional — pull either Cohere Transcribe (~1.6 GB) or SenseVoice Small (~228 MB) from Settings → Models when you're ready.
- Set your hotkey — Default is
Ctrl+Alt+T(Windows) orOption+Cmd+T(macOS). Customize it anytime. - Start translating — Press the hotkey anywhere, type or speak your text, and go.
macOS users: Grant Accessibility and Input Monitoring permissions when prompted. LiveLingo needs these to register global hotkeys and inject text.
Contributions are welcome! Whether it's a bug fix, new language support, or a feature idea — check the Contributing Guide to get started.
Stop copy-pasting between translator tabs.
Start LiveLingo-ing. 🚀