A scanned PDF is a stack of photographs. OCR (optical character recognition) reads those photographs and writes an invisible text layer over them — the page looks identical, but you can now search, select and copy.
OCR is exactly the kind of document most people hesitate to upload: contracts, medical records, IDs. Kameleo runs OCR with WebAssembly on your own machine. It's slower than a data-center GPU — that's the honest cost of privacy — but nothing ever leaves your device.
Scan at 300 DPI, keep pages straight, and prefer black-on-white. Skewed, low-resolution or handwritten pages will produce errors — OCR reads print, not handwriting. After OCR, spot-check by searching for a few words you can see on the page.
Once the text layer exists, the rest of the toolkit lights up: extract tables to CSV, chat with the document, summarize it, or compress the scan without losing the new searchability (Balanced level only — Strong re-renders pages and drops the text layer).