Private Translator
Translate between 100+ languages with a real neural model that runs entirely on your device. Your text is never uploaded, and once it loads, it works with your internet off.
🔒 100% private — nothing leaves your deviceHow it works
When you translate for the first time, the page downloads a quantized copy of Meta’s NLLB-200 model, a 600-million-parameter neural network trained to translate between 200 languages. The model file is stored in your browser’s cache. From then on, every translation is computed locally: your sentence is tokenized, run through the encoder and decoder on your own CPU using WebAssembly, and turned back into text. There is no server in the loop, which is why nothing you type is ever sent anywhere.
Why translate on your own device?
Most translation sites send whatever you paste to a company’s servers. For a quick phrase that is fine, but for a contract clause, a medical note, a private message, or anything covered by a confidentiality agreement, it means handing your words to a third party. Running the model in your browser keeps the text on your machine. It also means no rate limits, no sign-up, no API key, and it keeps working on a plane or anywhere without a connection once the model has cached.
Good things to try
Paste a paragraph and watch it translate sentence by sentence. Use the swap button to flip the direction and translate a reply back. Switch the target to a language you are learning and check your own writing. Translate the same sentence into several languages in a row to compare. Because it is all local, you can do this as many times as you like, on text you would never paste into a website.
Is my text uploaded anywhere?
No. The model runs in your browser. Your text is processed on your own device and never sent to a server, which is why it is private and works offline.
How good are the translations?
NLLB-200 is a strong open research model from Meta covering 200 languages, including many that big translators handle poorly. The in-browser build is quantized to keep the download reasonable, so for very nuanced or literary text a large cloud model may still edge it out, but for everyday translation it is very capable.
Why does the first translation take a moment?
The first time, it downloads the model (about 320 MB) and warms it up. That happens once. Your browser caches it, so later visits start fast and run with no connection.
Which languages are supported?
The model supports 200 languages. This tool exposes 100+ of the most widely used, with correct scripts, and you can pick any pair in either direction.
Do I need a fast computer or a GPU?
No GPU is needed; it runs on your CPU through WebAssembly. A modern laptop or desktop handles it comfortably. Very long passages translate sentence by sentence so you see progress as it goes.
I build tools like this every day.
Senior full-stack engineer, available for senior or contract work, fully remote. See the rest of the lab or get in touch.