Good tools should be clear about how they work.
This release provides local English editing rules. Generative AI rewriting is currently unavailable.
Local editing, with clear limits
The working engine shortens supported wordy phrases, fixes a focused set of common mechanics, and makes limited style changes. It is deterministic editing, not generative AI. If no supported rewrite applies, the tool says so. Quotes, URLs, numbers, code, and existing Markdown structure are protected.
Formal and professional modes expand supported contractions. Casual and friendly modes adjust supported expressions. Confident mode makes supported proposals more direct without removing uncertainty. Concise and simplifier modes reduce tested redundancies. Clear & Structured retains existing layout and simplifies wording; it does not invent headings or reorganize arguments. Academic rewriting is unavailable.
Why generative AI is unavailable
We ran real WebAssembly inference in Chromium using ONNX Community Qwen2.5 0.5B and Xenova FLAN-T5 base, both derived from Apache-2.0 models. Loading and inference worked, but the tested rewrites did not meet our quality requirements.
On synthetic examples, Qwen returned unrelated plans, a request for more details, or a refusal. FLAN-T5 omitted a sentence or repeated the input unchanged. We withheld those drafts and removed the model engines from the public application. This release downloads no model files and sends no text for remote inference.
This evaluation does not establish that all browser models are unsuitable. It explains why these candidates are not offered here. General sentence restructuring and broader paraphrasing remain gaps. No detector-evasion or human-authorship claims are made.
Patterns are not proof
The experimental pattern analyzer measures sentence length, repeated first words and selected formulaic phrases across the full input. It is not a validated AI classifier and never gives an authorship probability.
You have the final say
Local rules have limited coverage and cannot establish semantic equivalence or comprehensive grammar correctness. Review facts, specialized terms, tone and intent. There is no account, paid AI API, database, subscription or remote draft history. See our privacy policy and terms.