Report Abuse

Basic Information

MemoryOS provides a memory operating system for personalized AI agents, designed to enable coherent, personalized and context-aware interactions by managing long-term, mid-term and short-term memory. The project implements a hierarchical storage architecture with four core modules: Storage, Updating, Retrieval and Generation, and it supplies both a Python package and a separate MCP server for integrating memory functionality into agent clients. It is intended for developers and researchers who want to add persistent dialogue history, user profiles and knowledge management to conversational agents and other LLM-based systems. The README documents project structure, installation options via PyPI and GitHub, Docker deployment, ChromaDB integration, supported embedding and LLM providers, configuration parameters such as similarity thresholds, and evaluation on the LoCoMo benchmark demonstrating substantial F1 and BLEU-1 improvements.

Links

Categorization

App Details

Features
The repository emphasizes SOTA memory management performance on long-term benchmarks with reported large gains on LoCoMo. It offers a plug-and-play architecture allowing interchangeable storage engines, update strategies and retrieval algorithms. MemoryOS includes a MemoryOS-MCP server that exposes core tools such as add_memory, retrieve_memory and get_user_profile for agent workflows. The system supports multiple LLM providers and embedding models, local model inference options, and vector database integration including Chromadb. Recent updates highlight Docker deployment, parallelization for faster performance, new configuration parameters like similarity_threshold, and support for specific embeddings such as BGE-M3 and Qwen3. The codebase organizes components into short_term, mid_term, long_term, retriever and updater modules with prompts and utilities for LLM interactions.
Use Cases
MemoryOS helps developers and teams add persistent, multi-scale memory to AI agents to improve contextual continuity and personalization in conversations. It provides ready-made APIs and an MCP server that let applications save conversational turns, retrieve relevant historical context, and build user profiles derived from dialogue. The modular design makes it straightforward to swap storage backends, tune retrieval and update strategies, and connect different LLMs or local inference engines. Deployment options include a PyPI package, example scripts, Docker images and a Chromadb integration for vector storage. The project also supplies evaluation and reproduction scripts used in the paper, configuration examples and tests, which assist researchers in benchmarking and adapting the memory stack to their agents.

Please fill the required fields*