Agent memory refers to the mechanisms by which intelligent agents store, retrieve, and utilize information during task execution, and is an important component in building long-term interactive capabilities. This concept has gradually emerged with the development of agent systems based on large language models.

Agent memory aims to provide agents with the ability to store and retrieve information across tasks and multiple rounds of interaction, enabling them to continuously utilize historical experience and maintain contextual consistency during execution.

This mechanism is typically built upon a large language model, dividing memory into short-term and long-term layers: short-term memory, relying on a context window, is used for immediate reasoning and decision-making in the current task; long-term memory is stored in an external system and dynamically retrieved and injected into the model when needed. Through this hierarchical memory structure, the agent can not only process the current input but also perform continuous reasoning by combining historical information, thereby improving task consistency, decision accuracy, and adaptability in complex scenarios.