MemoryOS
A memory operating system for personalized AI agents, using an operating-system-style hierarchical storage architecture to keep long-term context coherent across conversations.
MemoryOS provides a memory operating system for personalized AI agents, drawing on memory-management principles from operating systems to enable coherent, personalized, context-aware interaction. It organizes memory through four core modules, storage, updating, retrieval and generation, and on the LoCoMo long-term-memory benchmark it reports average improvements of 49.11 percent in F1 and 46.18 percent in BLEU-1 over a baseline.
Its memory modules, including storage engines, update strategies and retrieval algorithms, are designed to be plugged in and swapped independently, and an accompanying MCP server, MemoryOS-MCP, injects long-term memory capabilities into other AI applications by exposing modular tools an agent can call. It integrates with a range of LLMs, including OpenAI, DeepSeek and Qwen models, and the work behind it was presented as an oral paper at EMNLP 2025. It is written in Python and licensed under Apache 2.0.