Repolode
Submit
VectifyAI

VectifyAI/PageIndex

A vectorless RAG engine that builds a hierarchical tree index of long documents and lets an LLM reason through it to retrieve answers.

View on GitHubWebsite
Stars
40,134
Forks
3,454
Open issues
151
Last commit
21 hours ago

What is it?

PageIndex is a retrieval framework for long, complex documents such as financial reports, legal filings, technical manuals and textbooks. Instead of splitting text into chunks and storing embeddings in a vector database, it generates a hierarchical tree index for each document and has an LLM search that tree to find the relevant sections, much as a human reader would navigate a table of contents.

The project offers a Python SDK (pip install pageindex) with a local mode that builds and stores indexes on your machine using your own LLM API key, and a cloud mode where indexing, OCR, image understanding and storage are handled by PageIndex Cloud. Local mode targets text-based PDFs, while cloud mode also handles scanned and image-rich documents. Answers include references back to the source, at page level locally and block level in the cloud.

The README reports benchmark results on FinanceBench and a cost comparison against passing whole PDFs to a model, but these are the authors' own measurements.

Key features

  • 01Builds a tree-structure index per document without a vector database or chunking.
  • 02Retrieves by LLM reasoning over the tree rather than semantic similarity.
  • 03Local mode runs on your machine with your own LLM API key.
  • 04Cloud mode adds OCR, image understanding and managed storage.
  • 05Answers include traceable references to source pages or blocks.

Use cases

  • Question answering over long financial reports or regulatory filings.
  • Querying legal contracts or technical manuals where exact section references matter.
  • Building RAG over textbooks or academic papers without managing a vector store.

Quick start

  1. 1

    Install the PageIndex package

    pip install -U pageindex
  2. 2

    Set your OpenAI API key and build a tree index from a PDF, then ask a question

Things to know

  • Needs an LLM API key (such as OpenAI) or a Claude subscription for indexing and chat.
  • Local mode is aimed at text-based PDFs; scanned or image-rich documents need the paid Cloud service.
  • Query accuracy and cost depend on the chat model you choose.

In collections

Open-source AI agent frameworks & apps

Topics

#rag#llm#retrieval#document-qa#information-retrieval#python#reasoning#agentic-ai#agents#ai

FAQ

What is PageIndex?
A vectorless RAG engine that builds a hierarchical tree index of long documents and lets an LLM reason through it to retrieve answers. PageIndex is a retrieval framework for long, complex documents such as financial reports, legal filings, technical manuals and textbooks. Instead of splitting text into chunks and storing embeddings in a vector database, it generates a hierarchical tree index for each document and has an LLM search that tree to find the relevant sections, much as a human reader would navigate a table of contents.
Is PageIndex free to use? What license does it have?
PageIndex is open source under the MIT license. Check the license terms before using it in commercial products.
Is PageIndex still maintained?
Actively maintained. The last commit to PageIndex was on Nov 6, 2026.
What can I use PageIndex for?
Question answering over long financial reports or regulatory filings. Querying legal contracts or technical manuals where exact section references matter. Building RAG over textbooks or academic papers without managing a vector store.
How do I install PageIndex?
Install the PageIndex package: pip install -U pageindex Set your OpenAI API key and build a tree index from a PDF, then ask a question
Where can I get PageIndex?
The source code and installation instructions are on GitHub: https://github.com/VectifyAI/PageIndex

Summary generated by AI from the project's README. GitHub stats refresh daily.

Related projects