GitHub - MemTensor/MemOS: Build memory-native AI agents with Memory OS — an open-source framework for long-term memory, retrieval, and adaptive learning in large language models. Agent Memory

MemOS is an open-source Memory Operating System for LLMs and AI agents that provides unified long-term memory management with store, retrieve, and manage capabilities. It features a unified memory API with graph-structured storage, multi-modal memory support (text, images, tool traces), multi-cube knowledge base management for isolation and sharing, asynchronous ingestion via MemScheduler for production stability, and memory feedback mechanisms for refinement. The framework claims 43.70% accuracy improvement over OpenAI Memory and 35.24% token savings. Available as both a cloud-hosted API and self-hosted deployment via Docker or CLI, with support for Neo4j, Qdrant, and Redis Streams.