The capacity of a system to retain and reuse information within or across interactions.
Memory in AI can refer to several different ideas, but in prompt engineering it usually means how much information the model can retain within the active context or how an application stores information for later reuse.
Short-term memory is often constrained by the context window, while longer-term memory may come from application logic, retrieved documents, or saved state outside the model itself.
Understanding memory matters because users often expect continuity. If the system cannot retain the right information, prompts may need to restate context or rely on retrieval workflows.
Memory strategies can include summarization, retrieval, state management, and selective reinsertion of important information into the prompt context.