About and method
What RinsePDF is
RinsePDF is a free web tool that makes a clean copy of a document before you give it to an AI assistant. It exists because a PDF can contain text and active content that a person never sees and a language model reads in full. Instructions hidden this way are called indirect prompt injection.
The site is operated by RinsePDF. You can reach us at hello@rinsepdf.com.
The method, step by step
- Open. Your browser parses the PDF with Mozilla's pdf.js, with script execution and XFA forms switched off. Images and screenshots are decoded by the browser itself.
- Draw. Each page is painted onto a canvas at 110, 150 or 220 dots per inch, depending on the quality you choose. Only what a PDF reader would visibly paint ends up in the pixels.
- Discard. The original file's text layer, annotations, form fields, embedded files, scripts, actions and metadata are not copied anywhere. Photos are re-encoded, which drops their EXIF data such as GPS position.
- Rebuild. For a clean PDF, each canvas is saved as a JPEG and placed on a new page with pdf-lib. For text, Tesseract reads the pixels with optical character recognition (OCR) on your device. A searchable PDF combines both: pictures, plus an invisible layer containing only the words OCR read from those pictures.
- Compare (optional). When OCR runs on a PDF, the words OCR found are compared with the words in the original text layer. Runs of words that exist in the file but were not seen on the page are shown to you as possible hidden text.
What it does not do
- It does not remove instructions that are visibly printed on the page, including very small or low-contrast text that is technically visible.
- It is not antivirus software. It does not detect malware, and it does not change or repair your original file.
- It cannot guarantee how any AI system will behave. Treat it as one layer of protection, and review AI output about documents you did not write.
- OCR is imperfect. Check names, numbers and dates against the page when they matter.
- The hidden-text check can miss things and can raise false alarms when OCR misreads a passage.
Privacy by construction
All processing happens in your browser's memory. Files are never uploaded, and the site has no server-side code that could receive them. The libraries are served from rinsepdf.com itself, and after your first visit the tool works with your network connection switched off. See the privacy policy for details about hosting logs and advertising.
Open-source components
- pdf.js by Mozilla, Apache License 2.0
- pdf-lib, MIT License
- Tesseract.js and the Tesseract OCR engine, Apache License 2.0
- Bricolage Grotesque and Instrument Sans typefaces, SIL Open Font License 1.1
Corrections
If you find a file that RinsePDF handles incorrectly, or a claim on this site that is inaccurate, write to hello@rinsepdf.com. Please do not send confidential documents.