PageIndex is an open-source retrieval engine that answers questions about long documents by having a language model reason through a tree of their sections, with no vector database.
In plain words
Ordinary RAG fetches passages by how similar they are to the question, which can miss text that matters but looks different. PageIndex builds a tree index of each document and lets the model search it, much as an expert turns to the right part of a long report, so answers trace back to clear references. Its Flash engine builds the tree from page layout, so a model only has to write the summaries.
A quick example
Imagine a 300-page company report. Instead of cutting it into chunks and matching them by similarity, a tree-based system follows chapters and sections down to the one about debt to answer a question on borrowing.
Where you'll meet it
- Question answering over long professional documents.
- A retrieval tool added to AI agents built with agent SDKs.
Related
Sources
- DocsPageIndex README, VectifyAI on GitHub · read 12 Oct 2026
- DocsPageIndex v0.4.0 release notes, VectifyAI on GitHub · read 12 Oct 2026
Checked by WhatIsAI against the sources above · 12 Oct 2026. A full five-minute explainer with a diagram may follow. Spotted an error? Report it.