We introduce a sleep phase in continual learning, enabling large language models to continuously integrate new facts and enhance reasoning while mitigating catastrophic forgetting or static knowledge cutoffs.