PageIndex is an open-source reasoning-based RAG framework that retrieves information from long documents without relying on a vector database or conventional fixed-size chunking. It builds a hierarchical tree representation similar to an LLM-oriented table of contents and lets models reason over that structure to locate relevant sections.

The project is based on the idea that semantic similarity is not the same as relevance, especially in professional documents where questions may require contextual understanding and multi-step reasoning. Retrieval is performed through tree search, producing explicit section and page references that make results easier to trace and explain. A file-system layer extends the approach from individual documents to large collections.

PageIndex is aimed at document-analysis tasks involving financial reports, legal documents, regulatory filings, technical manuals, medical literature, academic material, and other long structured texts. It can be self-hosted and also has hosted, API, MCP, and enterprise deployment options, making it useful for teams exploring alternatives to embedding-heavy vector RAG.