Definition
Storage that lets an AI agent remember facts, plans, and context across many separate conversations and tasks over time, instead of starting fresh each session. Persistent memory is what makes long-running agents useful, but it is also a durable target: poison the memory once and every future decision built on it is corrupted.
Why it matters
Organisations deploying long-lived agents are effectively granting them an institutional memory that will be attacked; protecting and auditing that memory is a growing governance obligation.