AtomMem : Learnable Dynamic Agentic Memory with Atomic Memory Operation

2026-01-13

AI Summary

Existing agent memory mechanisms predominantly rely on static, handcrafted workflows, which struggle to meet the dynamic demands of long-horizon tasks. This work reframes memory management as a learnable dynamic decision-making problem and introduces, for the first time, an approach that decomposes memory operations into atomic CRUD (Create, Read, Update, Delete) actions. By integrating supervised fine-tuning with reinforcement learning, the agent learns to autonomously develop task-adaptive memory scheduling strategies. The resulting AtomMem-8B model significantly outperforms current static memory methods across three long-context benchmarks and demonstrates structured, task-aligned memory management capabilities.

Technology Category

• Multiagent Systems: Mechanism Design
• Cognitive Modeling & Cognitive Systems: Agent Architectures
• Planning, Routing, and Scheduling: Planning with Language Models

Application Category

• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions
• Search and Retrieval-Augmented AI: Agentic search
• Economics, Online Markets and Human Computation: Cost models of using LLMs in production systems

Abstract

Equipping agents with memory is essential for solving real-world long-horizon problems. However, most existing agent memory mechanisms rely on static and hand-crafted workflows. This limits the performance and generalization ability of these memory designs, which highlights the need for a more flexible, learning-based memory framework. In this paper, we propose AtomMem, which reframes memory management as a dynamic decision-making problem. We deconstruct high-level memory processes into fundamental atomic CRUD (Create, Read, Update, Delete) operations, transforming the memory workflow into a learnable decision process. By combining supervised fine-tuning with reinforcement learning, AtomMem learns an autonomous, task-aligned policy to orchestrate memory behaviors tailored to specific task demands. Experimental results across 3 long-context benchmarks demonstrate that the trained AtomMem-8B consistently outperforms prior static-workflow memory methods. Further analysis of training dynamics shows that our learning-based formulation enables the agent to discover structured, task-aligned memory management strategies, highlighting a key advantage over predefined routines.

Problem

• agent memory
• long-horizon tasks
• static memory workflow
• learnable memory
• memory generalization

Innovation

• learnable memory
• atomic memory operations
• dynamic decision-making
• CRUD-based memory
• reinforcement learning for memory