MemOS: A Memory-Centric Operating System for Evolving and Adaptive Large Language Models

Researchers from MemTensor, Shanghai Jiao Tong University, Renmin University of China, and China Telecom have introduced MemOS, a memory-centric operating system designed to enhance the memory capabilities of large language models (LLMs). Unlike traditional LLMs that rely on fixed weights and transient context, MemOS employs MemCube, a unified memory abstraction that manages parametric, activation, and plaintext memory, enabling structured, traceable, and persistent memory handling across tasks and platforms. This innovation addresses critical limitations in current LLMs, such as forgetting past interactions and poor adaptability, by making memory a first-class