MemOS
A memory operating system for LLMs and AI agents that unifies storing, retrieving and managing long-term memory as an inspectable graph, with multimodal support and plugins for several agent runtimes.
MemOS is a memory operating system for large language models and AI agents, unifying storing, retrieving and managing long-term memory to enable context-aware, personalized interactions across knowledge bases, multimodal content and tool-use history. A single API adds, retrieves, edits and deletes memory that is structured as an inspectable, editable graph rather than an opaque embedding store, and covers text, images, tool traces and personas together in one memory system.
Multiple knowledge bases can be managed as composable 'memory cubes' with isolation and controlled sharing across users, projects and agents, memory operations run asynchronously with millisecond-level latency under high concurrency, and existing memories can be corrected, supplemented or replaced through natural-language feedback rather than only added to.
It connects to DeepSeek Harness, Hermes Agent and OpenClaw through local or cloud plugins that add automatic recall and background capture without changing the host agent's own core, and the project reports raising one agent's average task-completion rate from about 37 percent to about 51 percent across five tasks using its memory.