MemOS
An open-source AI agent memory system with on-device plugins for OpenClaw, Hermes Agent and DeepSeek Harness, plus a self-hosted Docker service.
MemOS gives AI agents long-term memory across conversations and tasks, with a choice of on-device storage, a self-hosted service or hosted cloud memory. It's for developers building assistants that remember user preferences, support history or lessons from earlier work. The project is open source under Apache 2.0.
The local plugin works with OpenClaw, Hermes Agent and DeepSeek Harness on macOS, Linux and Windows. It stores memory in SQLite on your machine without a cloud dependency for memory storage. It combines full-text and vector search to recall relevant context, removes duplicate memories and turns task experience into reusable skills. A local Memory Viewer lets you inspect what the agent has retained.
For applications and teams, the self-hosted service runs on your own machine or servers, including through Docker, with Neo4j and Qdrant. Its API handles adding, searching, editing and deleting memories. Graph-based memory remains inspectable and editable, and natural-language feedback can correct or replace stored information. It supports text and images alongside tool traces and personas. Separate knowledge bases let users, projects and agents keep memories isolated or share them selectively.
The hosted API stores data in MemOS Cloud and requires an account and API key. Cloud plugins for OpenClaw and DeepSeek Harness retrieve context before agent work and save new messages afterward. DeepSeek Harness can continue its current task if the cloud memory service is temporarily unavailable.
