MemTensor/MemOS

An AI memory operating system for LLM and agent systems, enabling persistent skill memory for cross-task reuse and evolution.

Core Features

Unified, graph-structured Memory API for store/retrieve/manage

Native Multi-Modal Memory support (text, images, tools, personas)

Multi-Cube Knowledge Base Management for isolation and composability

Persistent Skill Memory with auto task summarization and self-upgrading skills

Flexible deployment: Cloud-hosted service or 100% On-Device Local memory

GitHub Repo Full Preview Documentation

Quick Start

pip install MemoryOS

Detailed Introduction

MemOS is an advanced Memory Operating System designed for Large Language Models (LLMs) and AI agent systems. It provides a unified platform for storing, retrieving, and managing long-term memory, enabling context-aware and personalized interactions. By supporting multi-modal data, graph-structured memory, and composable knowledge bases, MemOS facilitates persistent skill memory, cross-task reuse, and continuous skill evolution, significantly enhancing agent performance and reducing token usage.