Search papers, labs, and topics across Lattice.
The paper introduces Agent-Radar, a training-free context management method for multi-agent LLM systems that addresses performance degradation caused by long conversation histories. Agent-Radar dynamically steers each agent's attention toward relevant context using a novel temporal and spatial decay mechanism to prioritize information. Experiments across five benchmarks show Agent-Radar outperforms state-of-the-art methods by up to 7.64 absolute points, maintaining robustness with increasing agents and interaction rounds.
LLM-powered agents can actually maintain focus in long conversations, thanks to a simple attention steering trick that boosts performance by 7.64 points.
LLM-based multi-agent systems have demonstrated remarkable performance on complex tasks through collaborative reasoning. However, these systems tend to rapidly accumulate extremely long conversation histories during interaction. As conversations lengthen, relevant information is increasingly diluted by irrelevant context, leading to degraded performance. In this work, we present Agent-Radar, a training-free context management method that dynamically steers each agent's attention toward relevant context with a novel temporal and spatial decay mechanism. Our experiments demonstrate that Agent-Radar outperforms state-of-the-art methods across five different benchmarks, yielding gains of up to 7.64 absolute points. Furthermore, our analysis shows that Agent-Radar remains effective and robust as the number of agents and interaction rounds increases. Finally, the ablation study shows that core components in Agent-Radar are crucial to performance and generalizable in different settings.
