GH Repository · MemTensor
MemOS
Self-evolving memory OS for LLM & AI Agents: ultra-persistent memory, hybrid-retrieval, and cross-task skill reuse, with 35.24% token savings and DeepSeek Harness support.
- stars
- 11,509
- 30-day movement
- +7224/day
- Related entries
- 60
- Connections
- 4
python/uvagent-apppython/poetryTypeScriptdsh-plugindeepseek-harnessmcptoken-savingsself-evolvingpythonllmclaudeagentmemory-managementhermesragskillsaiopenclawmemoryagentic-aimakedockerlong-term-memory
MemOS is a memory operating system for LLMs and AI agents, described as providing ultra-persistent memory, hybrid retrieval, and cross-task skill reuse. The repository ships agent configuration files (AGENTS.md, CLAUDE.md, .claude directory) and offers a DeepSeek Harness plugin, with TypeScript source managed via Python tooling (poetry, uv, Make, Docker).
You want a self-evolving memory layer so agents retain context across sessions and tasks while the author reports 35.24% token savings.
Use it to
- Add persistent long-term memory to LLM applications
- Reuse skills across agent tasks
- Reduce token spend via memory-based retrieval
- Integrate with DeepSeek Harness via the dsh plugin
- Replace or complement RAG pipelines with hybrid retrieval
For Developers building stateful agents and LLM apps
- Role
- agent-app
- Language
- TypeScript
- Licence
- Apache-2.0
- Forks
- 1,044
- Open issues
- 41
- Last push
- 2026-09-17
- Latest release
- v0.1.12 · 2025-07-08
topicsllm-memoryagentsmemory-managementraglong-term-memoryagentic-ai