PageIndex — a vectorless, reasoning-based RAG system that builds a hierarchical tree index from long documents and uses LLMs to reason over that index for agentic, context-aware retrieval.