It reasons, not just retrieves.
Agentic RAG plans a complex question into up to four sub-queries, runs them, and synthesises one grounded answer. Hybrid retrieval blends pgvector similarity with BM25 keyword ranking, so nothing relevant slips through.