MemAgent improves task accuracy by managing diverse memory sources
MemAgent: Learning to Manage Heterogeneous Memory Providers for LLM Agents
Artificial Intelligence
Summary
Many AI agents struggle to remember past tasks in ways that help with new problems. The authors found that using a single type of memory doesn't work well across different tasks. Instead, they created MemAgent, which decides which kind of memory to use for each situation and when to save new experiences. This approach improved performance by 10% and made agents work faster on average across several benchmarks.
What this means in practice
- •For ai system developers: Build LLM agents that better handle multiple memory types for improved long-term performance.
- •For conversational ai engineers: Enhance dialogue systems with adaptive memory routing to recall and use relevant past interactions effectively.
Authors
Yongxian Wei, Yilin Zhao, Runxi Cheng, Xinrui Chen, Chun Yuan, Yaoru Wang, Jiahong Yan, Dian Li
Abstract
Current agents remain largely stateless across tasks, limiting their ability to continually improve from prior interactions and making memory essential for long-horizon agentic behavior. Existing memory methods seek to reuse past experience, but most rely on a single memory representation (e.g., trajectories, reflections, skills, structured knowledge) whose effectiveness varies across task distributions. Rethinking this design space, we evaluate 13 memory methods and find that no single method generalizes across benchmarks, revealing the potential of managing heterogeneous memory providers. We formulate agent memory as a routing problem in which a memory agent decides which memory provider to retrieve from, whether to inject short-term memory, and which providers should store the resulting experience. Based on this perspective, we propose MemAgent, featuring a content-aware routing architecture and a training-data synthesis pipeline. The routing architecture combines content-aware probing before retrieval, short-term memory gating during execution, and selective multi-provider storage, while the training pipeline synthesizes phase-specific supervision for routing decisions. Across GAIA, WebWalkerQA, and xBench-DS, MemAgent improves average accuracy by 10.0% and outperforms every individual memory method across all three benchmarks. These gains come with less than 0.3% routing overhead and a 12% reduction in average task steps.