arXiv:2511.13593v2 Announce Type: replace 
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.

يتناول المقال O-Mem، وهو إطار ذاكرة جديد مصمم لتحسين أداء الوكلاء المدعومين من LLM في الحفاظ على التفاعلات طويلة الأمد. تواجه أنظمة الذاكرة التقليدية تحديات في الاتساق السياقي والتخصيص، مما يؤدي إلى ضوضاء في الاسترجاع. يتناول O-Mem هذه المشكلات من خلال إنشاء ملفات تعريف نشطة للمستخدمين وتحديث خصائصهم وسجلات الأحداث ديناميكيًا بناءً على التفاعلات. يسمح ذلك بالاسترجاع الهرمي لسمات الشخصية والسياق، مما يؤدي إلى استجابات أكثر اتساقًا وتكيفًا. حقق O-Mem درجة 51.67% في معيار LoCoMo، مما يدل على تحسين كبير.

El artículo presenta O-Mem, un nuevo marco de memoria diseñado para mejorar el rendimiento de los agentes impulsados por LLM en el mantenimiento de interacciones a largo plazo. Los sistemas de memoria tradicionales a menudo enfrentan desafíos en la consistencia contextual y la personalización, lo que conduce a ruido en la recuperación. O-Mem aborda estos problemas mediante el perfilado activo de usuarios y la actualización dinámica de sus características y registros de eventos basados en interacciones. Esto permite la recuperación jerárquica de atributos de persona y contexto, resultando en re…

Cet article présente O-Mem, un nouveau cadre de mémoire conçu pour améliorer les performances des agents alimentés par des LLM dans le maintien d'interactions à long terme. Les systèmes de mémoire traditionnels rencontrent souvent des difficultés avec la cohérence contextuelle et la personnalisation, entraînant du bruit dans la récupération. O-Mem aborde ces problèmes en profilant activement les utilisateurs et en mettant à jour dynamiquement leurs caractéristiques et leurs enregistrements d'événements en fonction des interactions. Cela permet une récupération hiérarchique des attributs de per…

The article discusses O-Mem, a new memory framework designed to enhance the performance of LLM-powered agents in maintaining long-term interactions. Traditional memory systems often struggle with contextual consistency and personalization, leading to retrieval noise. O-Mem addresses these issues by actively profiling users and dynamically updating their characteristics and event records based on interactions. This allows for hierarchical retrieval of persona attributes and context, resulting in more coherent and adaptive responses. O-Mem has achieved a score of 51.67% on the LoCoMo benchmark, …

O-Mem: Omni Memory System for Personalized, Long Horizon, Self-Evolving Agents

arXiv:2511.17208v1 Announce Type: new 
Abstract: LLM-based conversational agents still struggle to maintain coherent, personalized interaction over many sessions: fixed context windows limit how much history can be kept in view, and most external memory approaches trade off between coarse retrieval over large chunks and fine-grained but fragmented views of the dialogue. Motivated by neo-Davidsonian event semantics, we propose an event-centric alternative that represents conversational history as short, event-like propositions which bundle together participants, temporal cues, and minimal local context, rather than as independent relation triples or opaque summaries. In contrast to work that aggressively compresses or forgets past content, our design aims to preserve information in a non-compressive form and make it more accessible, rather than more lossy. Concretely, we instruct an LLM to decompose each session into enriched elementary discourse units (EDUs) -- self-contained statements with normalized entities and source turn attributions -- and organize sessions, EDUs, and their arguments in a heterogeneous graph that supports associative recall. On top of this representation we build two simple retrieval-based variants that use dense similarity search and LLM filtering, with an optional graph-based propagation step to connect and aggregate evidence across related EDUs. Experiments on the LoCoMo and LongMemEval$_S$ benchmarks show that these event-centric memories match or surpass strong baselines, while operating with much shorter QA contexts. Our results suggest that structurally simple, event-level memory provides a principled and practical foundation for long-horizon conversational agents. Our code and data will be released at https://github.com/KevinSRR/EMem.

تم اقتراح نهج جديد للذاكرة الحوارية طويلة الأمد في وكلاء نماذج اللغة الكبيرة (LLM) يركز على تمثيلات قائمة على الأحداث تجمع بين المشاركين، والإشارات الزمنية، وسياق الحد الأدنى. تهدف هذه الطريقة إلى تحسين التماسك والتخصيص في التفاعلات على مدى عدة جلسات، مع معالجة قيود نوافذ السياق الثابتة وأنظمة الذاكرة التقليدية.

Se ha propuesto un nuevo enfoque para la memoria conversacional a largo plazo en agentes de modelos de lenguaje de gran tamaño (LLM), centrado en representaciones basadas en eventos que agrupan participantes, señales temporales y un contexto mínimo. Este método busca mejorar la coherencia y la personalización en las interacciones a lo largo de múltiples sesiones, abordando las limitaciones de las ventanas de contexto fijas y los sistemas de memoria tradicionales.

Une nouvelle approche de la mémoire conversationnelle à long terme dans les agents de modèles de langage de grande taille (LLM) a été proposée, se concentrant sur des représentations centrées sur les événements qui regroupent les participants, les indices temporels et un contexte minimal. Cette méthode vise à améliorer la cohérence et la personnalisation des interactions sur plusieurs sessions, en s'attaquant aux limitations des fenêtres de contexte fixes et des systèmes de mémoire traditionnels.

A new approach to long-term conversational memory in large language model (LLM) agents has been proposed, focusing on event-centric representations that bundle participants, temporal cues, and minimal context. This method aims to enhance coherence and personalization in interactions over multiple sessions, addressing limitations of fixed context windows and traditional memory systems.

O-Mem: Omni Memory System for Personalized, Long Horizon, Self-Evolving Agents

Was this article worth reading? Share it

Teilord

MindPrism AI

Ai doll