MemOS

Active
GitHub TypeScript Apache-2.0

Description

MemOS is a Memory Operating System for LLMs and AI agents that unifies store, retrieve, and manage for long-term memory, with built-in KB, multi-modal, and tool memory support.

Key Features

  • Unified memory API for add, retrieve, edit, and delete — graph-structured, inspectable, not a black box
  • Multi-modal memory supporting text, images, tool traces, and personas in one system
  • Multi-Cube Knowledge Base management for isolated, shareable, composable memory across agents
  • Memory feedback and correction via natural-language for continuous refinement over time
  • Asynchronous ingestion via MemScheduler with millisecond-level latency at production scale
  • +43.70% accuracy vs OpenAI Memory on benchmarks, saving 35.24% memory tokens

Use Cases

💡 Adding persistent long-term memory to AI agents and chatbots
💡 Building personalized AI assistants that remember user preferences across sessions
💡 Managing multi-agent memory sharing with knowledge base isolation
💡 Storing and retrieving tool usage history for agent planning and reasoning

Tags

agent-memory memory-management long-term-memory rag llm knowledge-base

Categories

🧩 Agent Memory

Quick Start

Deploy via Docker or use the cloud service at memos-dashboard.openmem.net. Install the local plugin: `npm install @memtensor/memos-local-plugin`. For self-hosted: `docker compose up`. Use the unified API: `POST /memory/add`, `POST /memory/retrieve`, `POST /memory/edit`.
Visit GitHub Visit Website View Docs

Related Projects