Translate between languages without sending text anywhere.

No. Everything runs in your browser. The translation models download once from Hugging Face and cache locally; after that, translation is 100% on-device.

Meta AI's M2M-100 model (MIT-licensed) by default — a multilingual many-to-many model that translates directly between 100 languages. Pro users can switch to High Quality, which loads Google's MADLAD-400 (3B, Apache-2.0) for higher accuracy on a WebGPU-capable GPU. Pro users can also enable Max Quality, which loads Tencent's Hy-MT2 1.8B for the highest accuracy (requires WebGPU and a Pro plan).

English, Chinese, Japanese, Korean, Spanish, French, German, Italian, Portuguese, Russian, Dutch, and many more — 30+ in total. A single multilingual model handles every pair directly, including non-English ones like Chinese → Japanese.

One multilingual model (~430 MB, quantized) downloads on first use and caches afterward. Later runs are fast and fully offline.

Chrome 113+, Edge 113+, and other Chromium browsers with WebGPU give the fastest results. Safari and Firefox fall back to WebAssembly — slower, but fully functional.

Up to 5,000 characters per run for free and 50,000 with Pro — enough for long articles. Longer passages are split internally and translated piece by piece.

When the source is set to Auto, the tool detects the language on your device — fully offline, using a built-in heuristic — and then translates directly to the target. No text leaves your device.

Yes. The free tier covers everyday translation with a per-run character limit. Pro raises that limit and keeps a longer local history. Both tiers are private and on-device, and commercial use is included.