O-Mem: Omni Memory System for Personalized, Long Horizon, Self-Evolving Agents
Piaohong Wang, Motong Tian, Jiaxian Li, Yuan Liang, Yuqing Wang, Qianben Chen, Tiannan Wang, Zhicong Lu, Jiawei Ma, Yuchen Eleanor Jiang, Wangchunshu Zhou
2025-11-24
Summary
This paper introduces a new memory system called O-Mem for AI agents, aiming to make them better at remembering details about users over long conversations and adapting to their preferences.
What's the problem?
Current AI agents powered by large language models struggle to maintain consistent and personalized interactions over extended periods. They often rely on grouping information by topic, which can cause them to miss important, but seemingly unrelated, details about a user or introduce irrelevant information when trying to recall past interactions.
What's the solution?
The researchers developed O-Mem, which actively builds a profile of each user by tracking their actions and characteristics during conversations. This profile isn't just based on what topics are discussed, but also on specific details the user shares. O-Mem then uses this profile to quickly find relevant information when generating responses, leading to more personalized and coherent conversations. It's designed to efficiently retrieve both general user traits and specific details related to the current topic.
Why it matters?
This work is important because it significantly improves the performance of AI agents in remembering and utilizing user information, as demonstrated by better scores on standard tests. This advancement paves the way for creating more helpful, efficient, and human-like AI assistants that can truly understand and adapt to individual users over time.
Abstract
Recent advancements in LLM-powered agents have demonstrated significant potential in generating human-like responses; however, they continue to face challenges in maintaining long-term interactions within complex environments, primarily due to limitations in contextual consistency and dynamic personalization. Existing memory systems often depend on semantic grouping prior to retrieval, which can overlook semantically irrelevant yet critical user information and introduce retrieval noise. In this report, we propose the initial design of O-Mem, a novel memory framework based on active user profiling that dynamically extracts and updates user characteristics and event records from their proactive interactions with agents. O-Mem supports hierarchical retrieval of persona attributes and topic-related context, enabling more adaptive and coherent personalized responses. O-Mem achieves 51.67% on the public LoCoMo benchmark, a nearly 3% improvement upon LangMem,the previous state-of-the-art, and it achieves 62.99% on PERSONAMEM, a 3.5% improvement upon A-Mem,the previous state-of-the-art. O-Mem also boosts token and interaction response time efficiency compared to previous memory frameworks. Our work opens up promising directions for developing efficient and human-like personalized AI assistants in the future.