RAG Without Vectors: How PageIndex Retrieves by Reasoning

PageIndex is introduced as a vectorless RAG approach that uses a hierarchical, table-of-contents-style tree instead of embedding similarity.

It combines document structure with LLM-driven reasoning to retrieve relevant sections and answer queries more transparently.

The article shows PageIndex indexing and querying the Transformer paper, “Attention Is All You Need,” without vectors or embeddings.

The approach is positioned as a promising alternative for long, complex documents where accuracy, interpretability, and traceability matter, including research, legal, and financial use cases.