MuopDB is an on-disk vector database designed for machine learning workloads. It supports hybrid search combining text search with stemming and filtered vector search. The system utilizes multiple index types including HNSW, IVF, SPANN, and Multi-user SPANN, all stored on disk. It features product quantization and supports different I/O models such as memory-mapped files and asynchronous I/O with optional io_uring support on Linux. A key feature is multi-user support within collections, allowing individual user indices while sharing a knowledge base, targeting LLM memory applications.