Agentic RAG replaces the classic single-shot "query → embed → fetch → answer" pipeline with one in which an Agent decides at runtime which tools to call and when. Query rewriting, sub-question generation, multi-source retrieval across multiple Vector Databases, and self-critique with another retrieval pass are all decisions made on the fly. Llamaindex was among the first frameworks to name the pattern in 2024; it builds on Tool Use, the Reflection Pattern, and Self-Critique. It can clearly outperform vanilla RAG on complex multi-document questions, at the cost of higher latency and spend.
MEVZU N°124ISTANBULYEAR I — VOL. III
Glossary · Advanced · 2024
Agentic RAG
A RAG architecture in which an agent decides retrieval steps instead of a fixed pipeline.
- EN — English term
- Agentic RAG
- TR — Turkish term
- Ajan Davranışlı RAG