Agent memory

Agent memory is the state an AI agent keeps across steps or sessions, such as scratchpad notes, retrieved facts, user preferences, or task history.

Agent memory can be short-term state inside a single run, long-term memory stored in a database, or retrieved context from vector search. Useful memory improves continuity and task performance, but it must be scoped, refreshed, evaluated, and protected so stale, private, or incorrect information does not leak into future actions.