How This Agentic Memory Research Unifies Long Term and Short Term Memory for LLM Agents

Researchers from Alibaba Group and Wuhan University have developed Agentic Memory (AgeMem), a unified framework enabling large language models (LLMs) to autonomously manage both long-term and short-term memory within a single policy. Unlike traditional systems that treat these memory types separatelyrelying on external stores, heuristics, or external controllersAgeMem integrates memory management directly into the model's action space, allowing the agent to decide when to store, retrieve, summarize, or forget information dynamically during text generation. This innovation addresses key limitations of existing LLM architectures, which often treat long-term and short-term memory