PageIndex

Why we included this project

PageIndex takes a different route through RAG than most projects you will run into. Rather than embedding chunks and pushing them into a vector store, it builds a hierarchical in-context tree index of a document and lets the LLM reason over that structure to find the relevant passages, so you skip the usual chunking and similarity-search tuning. It is a self-hosted Python package that turns a PDF or markdown file into this index, with optional LLM-generated node summaries, and it also offers a hosted chat platform plus MCP and API access for heavier jobs like OCR on complex PDFs. Runnable agentic RAG examples and a file-system layer for searching across millions of documents mean it scales from a single report to a whole corpus. If vector retrieval keeps missing the right context in dense financial or technical documents, this is worth a close look.