How bai-lab/memoryos works
Project type and purpose
• MemoryOS is a full-stack, modular “memory operating system” for AI agents.
• It consists of:
• A core Python library (memoryos-pypi) implementing hierarchical memory management (short-, mid-, long-term).
• A microservice API server (memoryos-mcp) exposing endpoints (add_memory, retrieve_memory, get_user_profile).
• An optional vector-DB–backed variant (memoryos-chromadb).
• A web-based demo/playground (memoryos-playground).
• Evaluation scripts and a Docker deployment.
File structure
Top-level:
• Dockerfile, docs/, README.md, Paper-MemoryOS.pdf
• eval/ – benchmarking and reproduction scripts
• memoryos-chromadb/ – vector-DB storage provider implementation
• memoryos-mcp/ – MCP server, config.json, mcp.json, server_new.py, test scripts
• memoryos-playground/ – Flask (or similar) demo (“memdemo”) with templates, static data, app.py
• memoryos-pypi/ – core library modules: long_term.py, mid_term.py, short_term.py, retriever.py, updater.py, prompts.py, utils.py, memoryos.py
README for additional insights
• Confirms four core modules: Storage, Updating, Retrieval, Generation
• Describes plug-and-play support for multiple LLMs and embedding providers
• Details multiple “modes” (PyPI library, MCP server, Chromadb storage, Docker, Playground)