Section Anything
Section headings and the text under them, for PDFs with a text layer. Built for RAG
An open-source Python library that finds the real section headings in a PDF and their levels, so your chunks follow the document's structure. Pick a setting from free to model-assisted: . Works with LightRAG.
How it works
One library, several settings. Each one finds headings and their levels; they differ in what helps the rules decide.
Benchmarks (READoc)
Where others win
We list these on purpose. Pick the tool that fits your documents.
RAG pilot
Examples
Real documents where the heading tree goes wrong, next to READoc's answer key.