MindMemOS
A portable, self-evolving memory layer for AI agents that stores user and task context and reuses it across agents, automatically distilling recurring experience into reusable skills.
MindMemOS persists user profiles, preferences, project facts, tool experience and skill candidates as reusable memory so that OpenClaw, Hermes, Claude Code, OpenHands and other agents can share or transfer the same long-term memory instead of each starting from nothing. It aims to improve memory quality over time through schema learning, offline consolidation of memories, and corrections drawn from ongoing interactions, feeding those improvements back into how it adds and searches memories.
Experience captured as memories can be distilled into candidate skills, and the results of running a skill, including its failures, flow back into the memory system so skills keep evolving. Plugins connect MindMemOS to specific agents and workflows, retrieving relevant memories before an interaction and writing the conversation back afterward; an OpenClaw plugin and a DeepSeek Harness plugin are currently available, with more planned.
It offers an official cloud service and local self-hosting, reachable through an HTTP API, a Python SDK or CLI, or an agent plugin, all speaking the same protocol so any combination of access method and deployment mode works together. It is written in Python.