Chinese Researchers Launch MemOS, the First “Memory Operating System” for AI
A team from leading Chinese universities (Tsinghua, Shanghai Jiao Tong, Renmin) alongside MemTensor and China Telecom debuted MemOS, a pioneering system that treats memory as a core, manageable resource for AI.
Key features Of MemOS
Persistent memory: Designed so AI can recall past conversations, preferences, and updates across days or weeks—solving the “memory silo” problem.
Three-memory layers wrapped in “MemCubes”: Activations (short-term), Plaintext (editable, mid-term), Parameters (long-term, baked into model weights).
OS-like architecture: Interface for API calls, Operation with MemScheduler and governance modules, Infrastructure for memory storage and policy enforcement.
Why It Matters
MemOS shifts from short-lived context to dynamic, personal memory—essential for intelligent assistants in areas like healthcare, education, and enterprise collaboration.
With built-in policy, privacy, and governance layers, it paves the way for secure, accountable AI systems.
Availability and Pricing
GitHub release & license :MemOS 1.0 “Stellar” preview is available now under the Apache 2.0.
Platform support: Runs on Linux today, with upcoming support promised for Windows and macOS .
No fees: The software is free, with full access to code, APIs, and documentation.
While paid enterprise extensions or “memory modules” may arrive later, there’s currently no cost to deploy or use it.
News Gist
Chinese researchers have launched MemOS, the first open-source memory operating system for AI.
It enables long-term memory, boosts reasoning by 159%, and is free to use on Linux.
Windows/macOS support and future paid modules are coming soon.
