Memory Matters More: Event-Centric Memory as a Logic Map for Agent Searching and Reasoning
Yuyang Hu, Jiongnan Liu, Jiejun Tan, Yutao Zhu, Zhicheng Dou
2026-01-12
Summary
This paper introduces a new way for AI agents, powered by large language models, to remember and use past experiences to make better decisions over longer periods of time.
What's the problem?
Current AI memory systems often just store experiences as a big, disorganized collection, making it hard to find relevant information when needed. Even systems that try to organize memories struggle to understand the *relationships* between different experiences, and still rely on simply finding things that seem similar, which isn't very effective for complex reasoning or understanding how events connect over time.
What's the solution?
The researchers developed a system called CompassMem that organizes memories like a map of events. It breaks down experiences into individual events and then links those events together based on how they logically relate to each other. This creates a 'logic map' that allows the AI to navigate its memories in a structured way, focusing on events that are relevant to its current goals instead of just relying on surface-level similarities.
Why it matters?
This is important because it helps AI agents become more capable of handling complex tasks that require remembering and reasoning about many past experiences. By improving how AI remembers and connects events, CompassMem allows for more intelligent behavior and better performance in tasks like answering questions about stories or navigating complex environments.
Abstract
Large language models (LLMs) are increasingly deployed as intelligent agents that reason, plan, and interact with their environments. To effectively scale to long-horizon scenarios, a key capability for such agents is a memory mechanism that can retain, organize, and retrieve past experiences to support downstream decision-making. However, most existing approaches organize and store memories in a flat manner and rely on simple similarity-based retrieval techniques. Even when structured memory is introduced, existing methods often struggle to explicitly capture the logical relationships among experiences or memory units. Moreover, memory access is largely detached from the constructed structure and still depends on shallow semantic retrieval, preventing agents from reasoning logically over long-horizon dependencies. In this work, we propose CompassMem, an event-centric memory framework inspired by Event Segmentation Theory. CompassMem organizes memory as an Event Graph by incrementally segmenting experiences into events and linking them through explicit logical relations. This graph serves as a logic map, enabling agents to perform structured and goal-directed navigation over memory beyond superficial retrieval, progressively gathering valuable memories to support long-horizon reasoning. Experiments on LoCoMo and NarrativeQA demonstrate that CompassMem consistently improves both retrieval and reasoning performance across multiple backbone models.