VectifyAI/PageIndex

Project overview

PageIndex is a vectorless, reasoning-based RAG system that uses a hierarchical tree index and LLM reasoning to provide traceable and context-aware retrieval over long documents. It is designed for developers who need to avoid chunking and vector databases when processing simple PDFs and Markdown files.

Project type

AI Agent · RAG · Document Processing

Use cases

Knowledge Q&A

Deployment

Refer to project documentation

License

MIT

Best for

• Developers and AI engineers who need to build RAG applications over simple PDFs and Markdown files.

Key capabilities

• Transforms lengthy PDF or Markdown documents into a semantic tree structure similar to a table of contents, optimized for use with LLMs and AI agents.
• Performs reasoning-based, context-aware retrieval over long documents using LLMs without requiring a vector database or chunking.
• A simple, end-to-end agentic vectorless RAG example using self-hosted PageIndex with the OpenAI Agents SDK.

Limitations and risks

• The open-source repository is designed for generating the tree structure for simple PDFs using standard PDF parsing.
• Complex PDFs are hard to parse by classic Python tools and may lose document context and hierarchy.

Getting started

• Install dependencies using pip3 install --upgrade -r requirements.txt. Create a .env file with your LLM API key, then run python3 run_pageindex.py --pdf_path /path/to/your/document.pdf. Requires an LLM, with the default being gpt-4o-2024-11-20.

Evidence and sources

• GitHub project description: 📑 PageIndex: Document Index for Vectorless, Reasoning-based RAG
• Release: v0.3.0.dev3
• README: Are you frustrated with vector database retrieval accuracy for long professional documents? Traditional vector-based RAG relies on semantic *similarity* rather than true *relevanc…
• README: > PageIndex is a vectorless, reasoning-based RAG engine that mirrors how humans read, delivering traceable, explainable, and context-aware retrieval, without vector databases or c…
• README: uses LLMs to **reason** *over that index* for **agentic, context-aware retrieval**. The retrieval is *traceable* and *explainable*, with no vector DBs or chunking.

Reviewed at: 2026-07-17T10:18:17Z