Graph-based agentic RAG

This family combines knowledge graphs with agentic reasoning. The survey highlights Agent-G as a representative: it integrates structured graph knowledge bases with unstructured document retrieval, using modular retriever banks, a critic module, and feedback loops.

In Agent-G, graph knowledge bases supply relationships and hierarchies (for example, disease-to-symptom mappings in healthcare), unstructured documents add contextual detail, a critic module evaluates the relevance and quality of what was retrieved (flagging low-confidence results for re-retrieval), and feedback loops refine the process iteratively. The strength of this family is strong multi-hop reasoning over structured relationships combined with the flexibility of text retrieval. It suits domains where relationships are central, such as healthcare and legal analysis.