VectifyAI has released PageIndex, an open-source system designed to enable AI models to work with long documents like PDFs and legal texts more like human experts rather than traditional search engines. Unlike typical RAG systems that rely on chunking and vector search, PageIndex builds a hierarchical semantic tree of the document structure, allowing AI to perform logical reasoning and tree-based search.

Key benefits include better handling of complex and large documents by preserving context and document structure, eliminating the need for vector databases, and improving search accuracy through step-by-step structural analysis.

The framework includes scripts and Jupyter notebooks for generating the tree from PDFs or Markdown files and supports direct reasoning-RAG without external vector databases.

For more details, visit the GitHub repository and the official blog post.