Skip to content
Search

AI workspace · Runs on your device

AI Writing & Paraphrasing Studio

Paraphrase, fix grammar and change tone, with tracked changes. Runs on your device.

On-device AI

Not loaded

Rewriting runs in this browser tab. Your text is never sent to TodayAI or an AI service. The model files are downloaded once from Hugging Face and then cached on this device.

Choose a model
Mode

0 words · 0 / 12,000 characters · 0 min read

Load the on-device model above to rewrite. The rule-based checks below work without it.

Rule-based checks · no AI

Spelling slips, repeated words, missing apostrophes, spacing, capitals and very long sentences are flagged here as you type.

Export includes only accepted changes.

Small on-device models make mistakes and can miss errors. Every change is shown for you to accept or reject; warnings mark numbers, citations and terms that changed.

Find a real reference

Look up a DOI or a paper's title in Crossref, the free scholarly metadata registry. References are formatted from Crossref's record only; nothing is invented. Always check against the original publication.

How to use it

  1. Paste your text. Rule-based checks for common slips appear straight away, with no download.
  2. Load the on-device model (a one-time download of about 0.6 GB for Light or 1.4 GB for Quality), then pick a mode.
  3. Review each paragraph's highlighted changes, accept or reject them, then copy or export the result.

Good for

  • Fixing grammar in an email or cover letter
  • Paraphrasing a paragraph in your own words
  • Making a draft more formal or more academic
  • Cutting an abstract down to a word limit

Privacy and limits

Your text is processed in this browser tab and never sent to TodayAI or an AI provider. Reference lookup is off by default; when you switch it on, only your search words go to Crossref.

  • Up to 12,000 characters per run, processed paragraph by paragraph
  • Small on-device models are less accurate than large hosted ones and can miss errors; the Light model is basic
  • Best in English; other languages work less reliably
  • Needs a recent Chrome, Edge or Safari; without WebGPU the Light model runs slowly on the CPU

Engine: Qwen3.5 0.8B or 2B (Apache-2.0, ONNX q4/q4f16) via Transformers.js, on WebGPU or WebAssembly in a Web Worker; rule-based grammar checks; Crossref for optional reference lookup. Results can contain mistakes. Check anything important.