PageIndex
ActiveDescription
Open-source vectorless, reasoning-based RAG that builds a hierarchical tree index from long documents and uses LLMs to reason over it, achieving 98.7% on FinanceBench without vector DBs or chunking.
Key Features
- Vectorless retrieval — no vector DB or chunking needed; uses document structure and LLM reasoning
- Hierarchical tree index — automatically generates TOC-style tree structure from long documents
- Reasoning-driven — every retrieval result is traceable to explicit page and section references
- Context-aware — retrieval depends on full conversation context and incorporates new information dynamically
- Tree-search algorithm — LLM performs multi-step reasoning over the index tree to find relevant sections
- High accuracy — achieves 98.7% on FinanceBench, vastly outperforming vector-based RAG
Use Cases
💡 High-accuracy Q&A on long financial reports, regulatory filings, and academic textbooks
💡 Building explainable RAG systems where every retrieval result is traceable to the source page
💡 Replacing traditional vector retrieval to avoid similar-but-irrelevant false positives
💡 Powering enterprise knowledge bases that require multi-step reasoning over professional documents
💡 Providing Agentic tool-calling for LLMs to complete document analysis tasks
Strengths & Limitations
✅ Strengths
- • Actively maintained, recent updates
- • High community interest (35.2k stars)
- • Permissive open-source license (MIT)
- • Responsive to issues, low backlog
Categories
Quick Start
git clone https://github.com/VectifyAI/PageIndex.git
cd PageIndex
pip install -r requirements.txt
export OPENAI_API_KEY=sk-...
python examples/agentic_vectorless_rag_demo.py