Traditional RAG systems follow a relatively simple pipeline: retrieve relevant documents and pass them to a language model to generate an answer. While effective, this approach assumes that a single retrieval step is enough to gather all the information needed to answer a query.

In more complex scenarios, the system may need to search multiple sources, refine the query, gather additional evidence, or perform multiple retrieval steps before producing a final response. This is where Agentic RAG comes in.

Agentic RAG introduces an LLM-driven agent that can plan and execute multiple actions such as retrieving documents, reformulating queries, calling external tools, or verifying intermediate results. Instead of a fixed pipeline, the system dynamically decides what to retrieve, when to retrieve it, and how to use the information.

In this chapter, we will explore how agent-based approaches can extend traditional RAG systems, enabling more flexible workflows and helping AI systems handle complex, multi-step information tasks.