Agentic Memory: Learning Unified Long-Term and Short-Term Memory Management for Large Language Model Agents

TL;DR

This paper introduces Agentic Memory (AgeMem), a unified memory management framework for LLM agents that integrates long-term and short-term memory handling into the agent's policy, improving long-horizon reasoning.

Contribution

AgeMem is the first unified framework that enables autonomous, end-to-end memory management in LLM agents through tool-based actions and a specialized reinforcement learning strategy.

Findings

AgeMem outperforms existing memory-augmented baselines on five benchmarks.

It achieves better task performance and more efficient context usage.

AgeMem provides higher-quality long-term memory management.

Abstract

Large language model (LLM) agents face fundamental limitations in long-horizon reasoning due to finite context windows, making effective memory management critical. Existing methods typically handle long-term memory (LTM) and short-term memory (STM) as separate components, relying on heuristics or auxiliary controllers, which limits adaptability and end-to-end optimization. In this paper, we propose Agentic Memory (AgeMem), a unified framework that integrates LTM and STM management directly into the agent's policy. AgeMem exposes memory operations as tool-based actions, enabling the LLM agent to autonomously decide what and when to store, retrieve, update, summarize, or discard information. To train such unified behaviors, we propose a three-stage progressive reinforcement learning strategy and design a step-wise GRPO to address sparse and discontinuous rewards induced by memory operations. Experiments on five long-horizon benchmarks demonstrate that AgeMem consistently outperforms strong memory-augmented baselines across multiple LLM backbones, achieving improved task performance, higher-quality long-term memory, and more efficient context usage.