Most PDF tools were built for a world of desktop installs and cloud storage you did not ask for. PDFwarp is a browser UI. The rewrite happens on our servers.
You pick a tool in the browser and drop a file. That upload goes to our API on Railway. The worker is a Docker image with Ghostscript, LibreOffice, Poppler, Noto and Liberation fonts, and the Python libraries named on each tool page. The result is streamed back as a download. Temp files for that request are deleted; we do not keep a user-facing document locker.
Thumbnails, drag-to-reorder, and the PDF Editor canvas run in the browser so you can see what you are about to send. The merge, the compress, the Office convert, and the AI call do not. Saying otherwise would be false.
Free PDF structure tools (merge, split, reorder, rotate, protect, unlock) copy or encrypt objects with pypdf, except unlock which prefers PyMuPDF so the unlocked file still opens in Ghostscript and LibreOffice afterwards. Compress is three Ghostscript presets plus an exact preview that runs those same commands. Office → PDF is headless LibreOffice. Images use Pillow (and pillow-heif for iPhone photos); PDF → Images uses Poppler.
HTML → PDF and Markdown → PDF are ReportLab typesetters, not Chrome. The PDF Editor is Pro and not AI: PyMuPDF extracts boxes, and saving rasterizes only the pages you edited at 200 DPI. AI tools extract text (or page images for Scan to Text) and send that content to xAI. Form Fill returns a field map, not a filled PDF. Originality does not search a paper database. Scan to Text writes a new typescript — it does not hide a text layer under your scan.
We will not invent in-browser processing for the rewrite itself, fake testimonials, or a tool count we do not ship. The pages share a layout so the product is learnable; the Under the hood section on each tool is where the engines, caps, and failure modes differ.
Unlimited conversions, 500 MB files, the PDF Editor, and 13 AI tools with a fair-use cap of 30 AI calls per day. Cancel any time.