About

Why it exists

Text that passes through AI tools and publishing pipelines picks up characters nobody typed. They are invisible, they survive copy and paste, and they cause a long tail of small problems: broken exact-match search, noisy diffs, failed validation, mysterious syntax errors. There were plenty of tools promising to make AI text “undetectable” and very few that just told you which codepoints were actually in your document. This is the second kind.

How it works

The application is a Next.js App Router project in TypeScript. Text is processed as a codepoint stream against explicit tables of invisible and homoglyph characters. Files are classified by magic bytes rather than extension, then routed to a format-specific handler that rewrites the container without its metadata segments. Everything runs in memory for the duration of the request.

• Text: invisible Unicode scrub, space homoglyph normalization, optional confusable-Latin mapping.
• Images: PNG and JPEG C2PA, EXIF and XMP segment removal.
• Documents: SVG, PDF, DOCX, ODT, HTML and Markdown provenance stripping.

What it will not claim

It will not tell you that your text will pass an AI detector. Detectors score writing style; this tool does not change writing style. It will not claim to strip a statistical watermark, because nothing that inspects characters can. And when it cannot parse something in a file, it says so in the report rather than reporting success.

Privacy

Nothing you submit is stored. There is no database, no object storage and no content logging. See the privacy page for specifics.

Roadmap

• Batch and ZIP handling for whole directories.
• Optional deeper PDF and EXIF handling for stubborn provenance.
• A documented public API for scripted use.