What it does is break a PDF down into Markdown while trying to preserve the information and structure inside it. The impressive part is that it is not limited to reading lines of text. It is designed to handle tables, diagrams, flows, graphs, and other visual elements that ordinary OCR can easily flatten or misinterpret.

That distinction matters. A PDF is often not just a sequence of sentences. It can be a report with a table, a diagram explaining a process, a chart carrying the actual argument, or a page whose meaning depends on how all of those pieces fit together. Unlimited OCR is interesting because it appears to understand that document-level context rather than simply extracting isolated characters.

It runs locally, like a local model on your laptop. The download is not tiny, and I had roughly 6 GB in mind while setting it up, but I like the idea of being able to process documents locally instead of sending everything to a remote service.

The first setup was a little difficult for me. At one point I thought it had stopped working, when it was probably still doing something in the background. Once it was running, though, the results were impressive.

It also has a practical advantage for working with AI assistants. If you convert a complex PDF into clean, structured Markdown first, you can give the model a much more useful representation of the source. That can mean fewer repeated image uploads, less wasted context, and fewer tokens spent trying to reconstruct the document’s layout from scratch.

I am still playing with it, so I am not going to pretend it is perfect. But it is very good at the part that matters: understanding the document well enough to turn it into usable Markdown without losing the relationships between the text, tables, diagrams, and other elements.

Worth checking out:

• Unlimited OCR on GitHub