100% Local & Private

AI Document Cleaner

Strip hidden characters, normalise punctuation and whitespace, and get a transparent AI-writing pattern report — without changing a single word. Runs entirely in your browser.

Words UntouchedTypography only
Fully ReversibleDiff, undo, restore
Honest ReportPattern score, not proof
Never UploadedNo servers, no APIs

Drag & drop a document

TXT, Markdown, HTML, DOCX, ODT, RTF or PDF. Your file is read in this tab and never leaves your device.

Clean the file, not the writing

Every change is technical or typographic, shown in a diff, and individually reversible.

  • Strict Text Preservation

    No rewriting, paraphrasing or spelling fixes. Words, numbers, citations and structure stay exactly as they are.

  • Hidden Character Removal

    Finds zero-width spaces, stray byte order marks, control and tag characters, while keeping joiners, bidi marks and non-breaking spaces that real text needs.

  • Configurable Em Dashes

    Replace em dashes with commas, en dashes or hyphens, with spacing handled for you. Number ranges and dialogue dashes are treated sensibly.

  • Transparent Pattern Report

    Sixteen documented indicators, each with its points, measurement and flagged passages. It's a pattern score, not a verdict.

  • Diff, Undo & Restore

    Compare original and cleaned side by side, reject any single change, undo any step, or restore the original instantly.

  • Nothing Leaves Your Device

    Parsing, cleaning, scoring and export all run in JavaScript in this tab. No uploads, no cloud AI and no detection APIs.

Frequently Asked Questions

Will this make my document pass AI detectors?
No tool can promise that, and this one doesn't try. It only removes technical artefacts and changes punctuation you choose. Detectors mostly judge word choice and how predictable the sentences are, and the cleaner never touches either. Removing invisible characters or em dashes does not show that a human wrote the text.
Does the cleaner change any of my words?
No. It never rewrites, paraphrases, corrects or shortens text. It only changes em dashes (to your chosen alternative), invisible and control characters, unusual or repeated whitespace and line endings. Encoding artefacts and formatting debris change only if you switch those options on. Every change appears in the diff and can be rejected.
What is the AI-pattern score?
It's the sum of 16 documented heuristics, such as formulaic transitions, assistant phrases, uniform sentence lengths and em dash frequency, each with a fixed maximum. It is not a probability of AI authorship. All of these patterns also occur in human writing, so treat each finding as something to review, not as evidence.
Why didn't my score change after cleaning?
The original and the cleaned text are scored with exactly the same method. Hidden characters and spacing are ignored during scoring, so removing them has no effect. Only visible patterns count, so the score moves only when a change affects one (for example, replacing em dashes).
Are AI phrases like “I hope this helps” removed?
No. Assistant phrases, preambles, placeholders and similar text are flagged in the report with the passage, the reason and a confidence rating, but they are never removed or changed automatically. Edit them yourself if you want to.
Is my document uploaded anywhere?
No. The file is read with your browser's File API, and DOCX/ODT archives, RTF, HTML and PDF are all parsed in this tab. The PDF engine's worker is served by this site, not a third party. There is no backend processing, no cloud AI, no detection API and no document telemetry.
Which formats keep their formatting?
DOCX, ODT, HTML, RTF, Markdown and plain text are edited in place, so styles, tables, links, images and metadata are preserved. PDF text can't be edited in place: the text is extracted and exported as a .txt file, and the layout is lost. The tool warns you before you export.
Which invisible characters are kept on purpose?
Non-breaking and narrow no-break spaces, ideographic spaces, zero-width joiners and non-joiners (needed for emoji and many scripts), directional marks for right-to-left text, invisible math operators, variation selectors and flag-emoji tag sequences. These are reported but never removed automatically.