A Rust tool now decides if your PDF needs OCR

Firecrawl's pdf-inspector picked up 589 stars in one day, pushing it to 18,260 total and second place on GitHub's Rust trending board. The tool doesn't extract text from PDFs. It classifies them first, sorting scanned pages from text-based ones so a system knows which documents actually need OCR.

That sorting step is where document processing burns money. Running OCR on every PDF regardless of whether it's scanned is slow and expensive. Skipping OCR on scanned pages produces garbage text. Getting the classification right before extraction starts sets the cost of everything downstream.

The project's recent releases point at hardening rather than new features. August 17's v1.15.0 added selective OCR for scanned and mixed PDFs. Four days earlier, v1.14.2 shipped hardened parsing for what the changelog calls pathological PDFs, the malformed or adversarial files that break naive parsers.

That sequence, mixed-format handling first, then defensive parsing against bad input, reads like a team that hit real-world PDFs that didn't fit the clean cases and had to build for them. Anyone processing documents that OCRs by default is paying for compute on files that don't need it. A classification pass ahead of extraction is a cheap fix for an expensive habit, and pdf-inspector is one bet on what that pass should look like. Whether it holds up against PDFs deliberately built to break a classifier, rather than just confuse one, isn't something the release notes address yet.