Agentic RAG describes retrieval-augmented generation systems that can make choices about their own retrieval process.
Short definition
Instead of running one search and passing the results to a model, an agentic RAG system can decompose a question, search multiple sources, inspect results, and decide whether it needs more evidence.
How it works
A planner proposes a retrieval path, tools access the relevant sources, and an evaluator checks whether the collected evidence supports the answer.
When to use it
Use it when questions span systems or require iterative investigation. For simple lookup, conventional RAG is often faster and easier to operate.
