Blog/Technical notes

A PDF engine that never takes the file

Server-side PDF shops scale with RAM and disk. A 50 MB scan can need a few hundred megabytes to flatten. Doing that per concurrent user is why “free” tools watermark, throttle, or quietly keep copies. ProMaxPDF moves the bill to the user’s CPU instead — which is what students and shops actually want when the file is a CV or a contract.

Storage

Active bytes stay in RAM as ArrayBuffers. IndexedDB can hold a buffer if a tool needs to resume. localStorage only stores names, sizes, and timestamps — never file content.

Compression

Light and Medium try to recompress embedded images and strip metadata while keeping text as vectors. Heavy rasterizes if the file is still large. That is an honest trade: smallest files may lose selectable text.

OCR

Rebuilding a page from OCR boxes wrecks handwriting and tables. The searchable-scan path keeps the pixels and hides the text layer underneath, which is what Ctrl+F actually needs.

Office conversion

Without LibreOffice on a server, Word/PPT round-trips will not match Microsoft’s renderer. We still ship them because they are useful; we do not pretend they are lossless.