MemOS

Why we included this project

LLM agents that lose track of earlier turns or prior tasks are a common frustration, and MemOS addresses that with a memory layer that stores what an agent remembers as an editable, inspectable graph rather than an opaque embedding index. That design makes it easier to see why an agent answers the way it does and to correct outdated or wrong memories, and the same system handles text, images, tool traces, and user personas together. It fits long-running assistants, support bots that recall past tickets, and multi-agent setups that need shared context. You can use it as a hosted API or self-host it, with local plugins for OpenClaw and Hermes so you don't have to rebuild your stack, though the token-savings and benchmark numbers are worth testing on your own workload before you commit.