Source: Singh et al. 2025: Agentic Retrieval-Augmented Generation, a Survey
Survey that classifies agentic RAG architectures (router, multi-agent, hierarchical, corrective, adaptive, graph-based, document workflows).
Contents
- What: survey paper, arXiv:2501.09136
- Author: Aditi Singh, Abul Ehtesham, Saket Kumar, Tala Talaei Khoei, Athanasios V. Vasilakos
- Date: 2025-01-15 (v4 2026-04-01)
- Original: arXiv:2501.09136
Summary
Surveys systems that embed agents in the RAG pipeline, using reflection, planning, tool use and multi-agent collaboration to manage retrieval dynamically.
Key claims
- Single-agent router: one agent routes each query to SQL, vector search, web search or another source.
- Multi-agent and hierarchical: specialized retrieval agents per source, coordinated by a master or by tiers.
- Corrective: agents grade retrieved documents, refine queries and fall back to external search.
- Adaptive: a classifier estimates query complexity and chooses no retrieval, single-step or multi-step retrieval.
- Graph-based and agentic document workflows extend the pattern to knowledge graphs and end-to-end document processing.
- Open challenges: evaluation, coordination, memory, computational overhead and latency, governance.