Agentic RAG

What Is Agentic RAG?

Classic RAG fires a vector search on every user query. Agentic RAG lets an LLM agent reason first: route trivial questions to direct generation, run multi-step retrieval for complex ones, and re-query when the first chunks are insufficient. Tools used include LangGraph state machines, function-calling, and self-reflection prompts. Cuts cost on cheap queries and improves accuracy on hard ones.