A language model does not remember anything between calls on its own. Whatever it should know has to be placed in the prompt each time. Agent memory is the set of techniques for deciding what to save and what to bring back.
Two kinds are common. Short-term memory is the running state of the current task: the conversation so far, tool results and notes the agent has made, all of which must fit in the context window. Long-term memory is stored outside the model, in a database or a vector database, and retrieved when it becomes relevant, usually with semantic search.
Memory needs care. Stale or wrong entries get repeated as if they were true, and saving too much crowds out what matters. Decide what is worth keeping and when it expires.
Related: AI Agent, AI Agent Context, Embeddings.