A unified benchmark for evaluating continual agent memory in LLM-based systems across 5 evaluation modes (Online, Offline, Replay, Transfer, Repair) and 6 interactive tasks, supporting both system and personal memory mechanisms.