Product · OCR
Read text from scans or image-only PDFs.
A PDF may have no text layer. Webfold then falls back to optical character recognition. It returns the text as Markdown. It is the same call that reads normal PDFs.
See it work
Try it from wherever you work.
1webfold pdf scan.pdf
2webfold pdf scan.pdf --chunk 4000
What you get
What OCR does.
Automatic fallback
A PDF page with no text goes through OCR, and so does a scan whose only text is a stamp. The command line, the MCP tool and the library decide page by page. Pass --no-ocr to the command line to turn it off.
Same output as everything else
Markdown, chunks and the document layout graph, so a scanned contract and a born-digital one flow into the same pipeline.
Runs where your files are, or hosted
Run OCR with Poppler and Tesseract on your own machine and nothing leaves it, or call the hosted endpoint for up to 25 MiB per file.
A layout graph in development
A document graph with regions and provenance exists, and an OCR and graph-fidelity test harness is in place.
How it works
Three steps.
- 1
Check for text
Read the page's own text layer first.
- 2
Rasterise and recognise
If there is none, render the page and run Tesseract.
- 3
Return Markdown
Text per page, ready to chunk and index.
Limits up front.
Know the limits
- Locally, OCR needs
pdftoppm(Poppler) andtesseractinstalled; without them, image-only pages return nothing. The hosted endpoint limits pages (200 at most) and runs a few jobs at a time, and answers 503 when busy. - OCR was scored on eight small fixtures and on the scanned sets of two public benchmarks. It matches Tesseract run alone, so it cannot beat it. Treat the output as a first pass.
- Mixed born-digital and scanned pages, and unusual PDF filters, need more test coverage.
Keep reading.
FAQ
Questions.
Does it OCR every PDF?
No. A page with readable text is read from its text layer. OCR fills in the pages that have none, and the scans whose only text is a stamp, in the library, the command line and the MCP tool.
Which OCR engine does it use?
Tesseract, driven through Poppler's page renderer. The test harness can score other engines, but they are not wired into the product.
Use OCR with Webfold.
Run the command-line tool today. Join the waitlist for the hosted API.