When I started building agentic workflows in n8n, I kept running into performance and token consumption problems processing formats like epub, pdf, and the ms office *.docx/pptx/xlsx. Stripping them down to markdown was the obvious fix since the content mattered more than the form.
I ended up with a node to process each format into markdown. It dramatically improved accuracy and let me run the better models without sacrificing my credit card (and marriage) over to the token gods, or to use local models without brown outs or melted cables at home.
Last night I ripped it all out and replaced it with anydoc (link in comments). Incredible improvement. It’s lightning fast, provides consistent markdown across every format (or at least the ones I care about), and it replaced early workflow cruft/techdebt with almost zero effort.
Awesome tool to drop into your loops.
</>$ pip install firecrawl-anydoc </>$ cargo add anydoc </>$ npm install @firecrawl/anydoc