VectifyAI/PageIndex

What it solves

Vector-based RAG often struggles with long, complex professional documents because it relies on semantic similarity, which does not always equal relevance. PageIndex addresses this by replacing similarity search with a reasoning-based approach that mirrors how humans read, ensuring retrieval is context-aware and traceable.

How it works

Instead of using a vector database and chunking, PageIndex uses a hierarchical tree index. The process involves two main steps:

1. Index: A tree-structure index is generated for each document (extracted from layout and refined by an LLM).
2. Retrieve: An LLM agentically searches through this tree, reasoning its way to the relevant information rather than searching for similar text fragments.

Who it’s for

It is designed for users dealing with long, complex professional documents, such as financial reports, legal documents, regulatory filings, technical manuals, medical literature, and academic textbooks.

Highlights

• Vectorless Architecture: Eliminates the need for vector databases and text chunking.
• Reasoning-based Retrieval: Uses LLM reasoning over a tree index to find relevant information.
• Traceable Results: Provides explicit references, making retrieval explainable compared to "vibe retrieval."
• Flexible Deployment: Offers both a local mode for text-based PDFs and a Cloud version that supports OCR, image understanding, and corpus-level indexing via the PageIndex File System.