Standard RAG is a fixed pipeline: retrieve evidence once, pass it to a model, generate an answer. That is enough for many factual questions. It breaks down when the question requires planning, multiple searches, tool choice, or verification before answering.

In more complex scenarios, the system may need to search several sources, refine a query, inspect intermediate results, call a database, compare evidence, or decide that retrieval is unnecessary. This is where agentic RAG is useful.

Agentic RAG puts an LLM-controlled decision loop around retrieval. The system can plan, call tools, evaluate intermediate results, retry with a different strategy, and stop when it has enough evidence. The retrieval path becomes conditional rather than fixed.

This chapter covers router RAG, multi-step retrieval, adaptive retrieval, tool design, and when the extra cost and latency are justified.