Indexes long documents into hierarchical page structures and uses LLM reasoning for vectorless retrieval, offering developers an alternative to embedding-based RAG.