firecrawl/pdf-inspector

pdf-inspector is a Rust library built by Firecrawl that looks at a PDF file and figures out what kind of PDF it is before deciding what to do with it. It can tell whether a document is text based, scanned as an image, a mix of both, or image based, and it does this classification in about ten to fifty milliseconds by sampling the file's content rather than reading the whole thing. The point of doing this first is that a large share of PDFs, the README estimates around 54 percent, already contain real selectable text and do not need slow, expensive OCR processing at all, so those can be handled locally in under 200 milliseconds instead. For PDFs that are text based, the library extracts the text along with its position on the page, font information, and reading order, including handling multi column layouts the way a newspaper is laid out, and then converts all of that into clean Markdown. This includes recognizing headings by font size, bullet and numbered lists, code blocks, tables built either from the PDF's own drawing instructions or inferred from how text lines up, bold and italic formatting, and links. If it detects a broken or unusual font encoding it cannot reliably decode, it flags this so a calling application knows to fall back to OCR instead of producing garbled text. The library is written in pure Rust with no machine learning models and only a single external dependency for PDF parsing, and it ships bindings for Python, Node.js, and the browser through WebAssembly, plus a set of command line tools for converting PDFs to Markdown or running detection only. The README includes a benchmark against several other PDF parsing tools on a shared 200 document test set, where pdf-inspector reports the best overall score, the best reading order accuracy, the best table extraction accuracy, and the fastest total processing time among the local, non machine learning engines compared. It is intended as a strong default for reports, research papers, financial documents, invoices, and legal PDFs that need clean structured Markdown without the latency or infrastructure cost of running OCR. The project is released under the MIT license.