Context
LLM-generated text comes with a quiet trail. Zero-width spaces. Curly quotes where straight ones belong. Em-dashes that pattern-match as a “tell.” Sometimes tracking parameters from a clipboard manager. None of it shows up visually — but it shows up to anything that looks closely.
CopyCleanse is a small, focused tool for stripping those fingerprints out of pasted text — fast, in the browser, with nothing leaving the machine.
How it works
Paste text. Click clean. Copy the result. That’s it.
The whole interface — paste, clean, copy. Diff hidden by default; on demand if you want to see what changed.
Everything happens in the browser. No signup, no upload, no server-side storage of any kind. The entire app is shipped as static assets and runs in the user’s tab.
The cleaning rules cover:
- Zero-width and other invisible Unicode characters.
- Smart-quote and dash normalization (smart → straight, when that’s what you want).
- Common tracking parameters in pasted URLs.
- A growing rule list driven by what’s actually showing up in real-world LLM output.
More coming
The full case study with the rule architecture and how the tool keeps up with new fingerprints is being written. Stack and scope are accurate. For specifics, book a call.