PageIndex: A New RAG Approach That Bypasses Traditional Vector Databases

Key Info

Researchers developed PageIndex, a new RAG approach that eliminates vector DBs, data embeddings, chunking, and similarity search, using a tree index for LLM reasoning. It scores 98.7% on FinanceBench and outperforms all vector RAGs on the leaderboard, and is free and open source.

Highlights

• PageIndex uses a tree index instead of vector databases, letting the LLM reason through documents like a human reading a book.
• It bypasses traditional RAG dependencies: no vector DB, no embeddings, no chunking, no similarity search.
• Achieves 98.7% on FinanceBench and outperforms every vector RAG on the leaderboard.
• Offered as 100% free and open source software.