arXiv:2602.06025v3 Announce Type: replace-cross 
Abstract: Memory is increasingly central to Large Language Model (LLM) agents operating beyond a single context window, yet most existing systems rely on offline, query-agnostic memory construction that can be inefficient and may discard query-critical information. Although runtime memory utilization is a natural alternative, prior work often incurs substantial overhead and offers limited explicit control over the performance-cost trade-off. In this work, we present \textbf{BudgetMem}, a runtime agent memory framework for explicit, query-aware performance-cost control. BudgetMem structures memory processing as a set of memory modules, each offered in three budget tiers (i.e., \textsc{Low}/\textsc{Mid}/\textsc{High}). A lightweight router performs budget-tier routing across modules to balance task performance and memory construction cost, which is implemented as a compact neural policy trained with reinforcement learning. Using BudgetMem as a unified testbed, we study three complementary strategies for realizing budget tiers: implementation (method complexity), reasoning (inference behavior), and capacity (module model size). Across LoCoMo, LongMemEval, and HotpotQA, BudgetMem surpasses strong baselines when performance is prioritized (i.e., high-budget setting), and delivers better accuracy-cost frontiers under tighter budgets. Moreover, our analysis disentangles the strengths and weaknesses of different tiering strategies, clarifying when each axis delivers the most favorable trade-offs under varying budget regimes.

تم تقديم إطار جديد يسمى BudgetMem لتحسين إدارة الذاكرة في الوقت الفعلي لوكلاء نماذج اللغة الكبيرة (LLM)، مما يسمح بالتحكم الصريح والواعي في الأداء وتكلفة الذاكرة من خلال هيكل ذاكرة متعدد المستويات. يستخدم هذا النظام جهاز توجيه خفيف الوزن لتحسين معالجة الذاكرة عبر مستويات ميزانية مختلفة، مما يعالج عدم الكفاءة في أنظمة الذاكرة التقليدية.

Se ha presentado un nuevo marco llamado BudgetMem para mejorar la gestión de la memoria en tiempo de ejecución para agentes de Modelos de Lenguaje de Gran Tamaño (LLM), permitiendo un control explícito y consciente del rendimiento y costo de la memoria a través de una estructura de memoria por niveles. Este sistema utiliza un enrutador ligero para optimizar el procesamiento de la memoria en diferentes niveles de presupuesto, abordando ineficiencias en los sistemas de memoria tradicionales.

Un nouveau cadre nommé BudgetMem a été introduit pour améliorer la gestion de la mémoire en temps réel pour les agents de modèles de langage de grande taille (LLM), permettant un contrôle explicite des performances et des coûts liés à la mémoire grâce à une structure de mémoire par niveaux. Ce système utilise un routeur léger pour optimiser le traitement de la mémoire à travers différents niveaux de budget, répondant ainsi aux inefficacités des systèmes de mémoire traditionnels.

A new framework named BudgetMem has been introduced to enhance runtime memory management for Large Language Model (LLM) agents, allowing for explicit, query-aware performance-cost control through a tiered memory structure. This system utilizes a lightweight router to optimize memory processing across different budget tiers, addressing inefficiencies in traditional memory systems.