VectifyAI/PageIndex

PageIndex takes the vector database out of RAG entirely. Instead of chunking documents, embedding them, and hoping cosine similarity finds the right passage, it builds a structured index that a reasoning model can navigate directly. The bet is that LLMs are now good enough at logical lookup that vector search is the bottleneck, not the solution. That is a genuinely different architectural claim, not a config tweak. If it holds up, it cuts infrastructure cost and retrieval latency at the same time since there is no embedding pipeline, no index rebuild on updates, just document structure and a reasoning step. Early days and the repo is fresh, so caveat emptor on production hardening. But for anyone currently fighting chunking heuristics and low-recall retrievals on structured docs like contracts, manuals, or product specs, this deserves a Saturday morning. -> Best for: AI engineer building document-heavy RAG pipelines who suspects vector search is the wrong abstraction