Anyone who works with documents knows how frustrating it is to convert DOCX, PDF and other formats into clean Markdown. Misaligned tables, missing headings, scattered footnotes—you end up re-adjusting everything every time you switch formats.

AnyDoc takes a straightforward approach: **500 DOCX files processed in 1.7 seconds, less than 5 milliseconds per file.**

In the video above, a scanned copy of an English non-disclosure agreement is quickly recognized and converted into well-structured Markdown, with headings, clauses, and signatory information all clearly presented.

How does it pull this off?

There are two core principles:

**First: Unified document model.** All supported formats—14 in total, including Word, PPT, Excel, PDF, RTF, EPUB, CSV—are first parsed into a single intermediate model, then output via a unified Markdown serializer. That means table escaping, heading anchors, footnotes and all other formatting behave exactly the same no matter what input format you use. You don’t need to write separate processing logic for different formats.

**Second: Content-based format recognition, no dependency on file extensions.** It will convert correctly even if the file extension is wrong. What does that mean for you? All those messily-named files with wrong extensions you get from clients? Just drop them in, and it will process them just fine.

Some people have asked if it can handle complex tables properly. The project is open-source on GitHub at [github.com/firecrawl/anydoc](https://github.com/firecrawl/anydoc), so if you’re interested you can pull it down and test it yourself.

The value of this tool isn’t just that it *can convert*—there are plenty of tools that can do that. Its value lies in **speed** and **consistency**. When you need to process hundreds of documents in bulk, saving tens of seconds per file adds up to hours of total time saved. And having a consistent unified output eliminates the time you’d otherwise spend patching formatting over and over again during downstream processing.

For a utility project, this hits the mark perfectly.