AI agents can now keep their identity, even when their memory fails

AI agents usually lose their sense of self when they run out of memory or forget old conversations. This new architecture lets them keep a core identity even when their memory fails.

Current AI agents often act like different entities from one conversation to the next, making them unreliable for long-term relationships or complex projects. This paper proposes a way for them to maintain a consistent "self" over time. It means AI could handle ongoing tasks, like personal assistants or project managers, without constantly needing to be reminded of past interactions or their own purpose.