Extract tables from any PDF — clean JSON or markdown

How it works

Your PDF goes through a three-step pipeline: a fast parser reads digital PDFs directly; pages it can't read confidently (like scans) are handed to a vision model; and every table then gets cleaned up — headers detected, merged cells filled in, tables that continue across pages joined back together — with an honest confidence score attached, so you know how much to trust each result.

We never store your documents

Your PDF is processed in memory and discarded the moment your tables are returned. No copies on disk, nothing in logs, no accounts, no tracking of document contents. That's not a settings toggle — the processing pipeline is built so there is nowhere for your document to be kept.

Use this in your pipeline

Everything this page does is one POST /v1/extract call — same engine, same output, plus CSV, page selection, and higher limits. Get the API →