How to use
- Press Download model once (it is cached for next time).
- Paste text, pick a tone, length and number of versions, then press Rewrite; the text streams in and Stop keeps what has been written.
- Switch between versions, open Changes to see added and removed words, and copy the one you like.
Worked example
“hey can u send the report by fri” in Formal is rewritten as a polite request along the lines of “Could you please send the report by Friday?”; the exact wording differs between versions and runs, and some versions need editing.
Supported formats and limits
| Input | Text up to 4,000 characters |
|---|---|
| Output | Up to 3 rewritten versions with a word-level comparison |
| Limits | The first use downloads the SmolLM2-360M-Instruct model from Hugging Face (about 390 MB for the CPU build, about 276 MB with WebGPU), cached afterwards. Fast with WebGPU; without it, expect a few words per second. Your text never leaves this tab. |
| Engine | SmolLM2-360M-Instruct (Apache-2.0) running in your browser with transformers.js in a Web Worker (WebGPU when available, otherwise WebAssembly on the CPU); one worked example per tone, passages of about 120 words |
Limitations
- A small model: it can change the meaning, drop details or reply to the text instead of rewriting it (such versions are retried once and flagged).
- English works best; other languages are unreliable.
- Without WebGPU, rewriting runs on the CPU and can take a minute or more per paragraph.
Questions
Which model writes the versions?
SmolLM2-360M-Instruct (Apache-2.0), a small open language model, downloaded once from Hugging Face (about 390 MB, or 276 MB with WebGPU) and run in this tab. Your text is not uploaded.
Can I use the output as it is?
Read it first. The rewrites are machine-generated by a small model that can change the meaning, drop details or answer the text instead of rewriting it. Suspicious versions are retried once and flagged, but you should still check every one.
How fast is it?
Quick with WebGPU. Without it the model runs on the CPU and a paragraph can take a minute or more. Use English text; other languages are unreliable.
Privacy
On-device model. Processing runs in this browser. The open-source model files are downloaded once from the model host (Hugging Face) and cached; your content is not uploaded.
- Hugging Face: Only after you press Download model: requests for the model files (SmolLM2-360M-Instruct, about 390 MB, or 276 MB with WebGPU) from huggingface.co and its CDN, which see your IP address. Your text and files are never sent.
See the privacy policy for how toolsdocks handles data.