MEM-KB
A self-hosted knowledge workspace for teams that organizes documents, notes, and a knowledge graph for retrieval-augmented question answering, with built-in RAG evaluation and nightly self-consolidation.
MEM-KB is an open-source knowledge workspace that organizes documents, images, notes, past conversations, and a knowledge graph inside per-team or per-person workspaces, answering questions through hybrid retrieval, reranking, and traceable citations back to source documents. It runs scheduled RAG evaluations against a knowledge base that report recall, accuracy, citation correctness, and hit or miss documents with failure reasons, and a nightly consolidation job that fills in entity relationships, fixes broken citations, and reorganizes knowledge structure.
It is built for self-hosting, with support for multiple tenants, role-based access control, PostgreSQL row-level security, local or S3-compatible object storage, and an OpenAI-compatible model gateway so any compatible provider can supply the underlying language model. The stack is a React web frontend, a Fastify API, PostgreSQL with pgvector, Redis with BullMQ for background jobs, and an optional long-term knowledge engine component; without that component configured, the core knowledge base and question-answering interface still run, with the long-term engine features degraded. It requires Node.js, npm, Bun, and Docker, and ships scripts for infrastructure setup, database migration and seeding, type-checking and testing, end-to-end tests, and dependency auditing.