What Is Agentic RAG?

Agentic RAG Explainer

Agentic retrieval-augmented generation, or agentic RAG, gives an AI system control over the retrieval process. Instead of always issuing one query, the system can decide whether retrieval is needed and choose a source or search method.

The agent may rewrite a query, split a question, inspect results, identify missing evidence, and search again. It can route factual lookup to a database, current events to web search, and internal procedures to an enterprise index.

This flexibility helps with questions that require several sources or changing search strategies. Evidence evaluation can also prevent weak passages from being treated as sufficient just because a retriever ranked them highly.

The additional loop increases latency, cost, and risk. An agent can search the wrong collection, accept low-quality evidence, or keep retrieving without progress. Systems need source permissions, budgets, citation tracking, stopping rules, and tests for answer support.

Agentic RAG lets an AI system decide when, where, and how to retrieve information and whether the evidence is sufficient before answering.

Key terms include RAG, agent, retrieval, query rewriting, routing, evidence, source, search loop and stopping rule.

The agent chooses whether to search, which source to use, and how to reformulate the request.

It can inspect evidence and repeat retrieval when important gaps remain.

More retrieval steps increase opportunities for bad routing, weak evidence, and wasted resources.

Budgets, permissions, provenance, and evidence checks should constrain the loop.

Explore more "Explainers"

Discover additional explainers across politics, science, business, technology, and other fields. Each explainer breaks down a complex idea into clear, everyday language—helping you better understand how major concepts, systems, and debates shape the world around us.