The field of multi-agent systems is shifting towards the integration of knowledge graphs to enhance the reliability and coherence of Large Language Model (LLM) agent reasoning. Recent developments have focused on improving the retrieval and selection of agents and tools, as well as the integration of external knowledge to resolve entity ambiguities and improve named entity recognition. Notable advancements include the use of graph-based retrieval methods, structural graph constraints, and multi-agent frameworks that leverage knowledge retrieval, disambiguation, and reflective analysis. Noteworthy papers include:

• Agent-as-a-Graph, which introduces a knowledge graph retrieval approach that achieves significant improvements in recall and nDCG.
• Chatty-KG, a modular multi-agent system for conversational question answering over knowledge graphs that outperforms state-of-the-art baselines in both single-turn and multi-turn settings.
• Path-Constrained Retrieval, a retrieval method that combines structural graph constraints with semantic search to ensure retrieved information maintains logical relationships within a knowledge graph, achieving full structural consistency and strong relevance scores.