Is my text uploaded to a server?

No. Translation runs in your browser. The server only delivers the page and model files.

Why can first use take a while?

The browser needs to download and cache the on-device model. Later visits reuse the browser cache.

Does it work in China?

Yes. Model files are fetched from zhanghe.dev through Cloudflare/R2, not directly from Hugging Face.

After the model is downloaded and cached, translation can continue during poor connectivity. Clearing site data requires a new download.

product / on-device translation

Translate

A browser-first AI translator. Model files are served from zhanghe.dev on Cloudflare R2, and translation text stays in your browser.

Hy-MT1.5 1.8B / Q2_K on-device modelText translation v1Local history stays in your browser

architecture

Privacy And Performance By Default

The page is server-rendered first, then the model runtime loads only after you press Prepare or Translate. Weak networks and data saver mode still avoid accidental model downloads.

worker

Browser inference

Translation runs inside a Web Worker. The main thread stays responsive, and your text is not sent to a third-party translation API.

China-friendly delivery

Model, WASM, and tokenizer assets are served from same-origin Cloudflare/R2 paths instead of Hugging Face CDN.

runtime

Runtime compatibility

The runtime prefers self-hosted Transformers.js ONNX files and falls back to a browser-side GGUF/WASM adapter when needed.

related workflows

Native TranslateOffline Chrome translation extension using Chrome 138+ built-in Translator and Language Detector APIs MangoFlowAI chat sidebar for smarter web browsing and translation workflows

official links

Official links

Use these canonical links for citations, installation, source review, privacy checks, and support.

Website Support

Model Delivery

R2 object access is restricted to the models/translate/ prefix. Responses support Range, ETag, and long immutable caching for large GGUF artifacts.

/models/translate/*

Henry Zhang