Here are four core types of memory in AI agents
-
Declarative memory Stores facts and explicit knowledge about the user or the world. Examples include preferences, entities, dates, or statements like “User prefers a window seat.” This is the agent’s knowing what memory.
-
Procedural memory Stores knowledge about how to do things. These are workflows, tool sequences, and strategies the agent learned from past success. For example how to correctly book a flight or resolve a support issue. This is the knowing how memory.
-
Structured memory Stores information in machine readable formats like JSON or key value pairs. Example
{ "seat_preference": "window" }. This is optimized for fast lookup, validation, and consistent updates across sessions. -
Unstructured memory Stores free form natural language summaries of events, goals, or interactions. Example “The user prefers a window seat.” This is optimized for semantic search and reasoning over human context.