Agentic RAG

Agentic RAG lets model-driven decisions change whether, where, or how often a system retrieves evidence.

Tiny comparison

fixed RAG:     question → known retriever → answer
agentic RAG:   question → choose search → inspect evidence
                        → refine or search elsewhere → answer

Several sources do not automatically make a system agentic. Code can search known sources in parallel. The agentic part begins when model judgment changes the path from observations.

FDE note

Allowlist sources, preserve evidence provenance, cap rounds and cost, define an evidence-completeness rule, and test source selection separately from answer quality. Prefer a fixed workflow when the path is already known.

Continue with Agentic RAG earns its loop.