MatrixHub is an open-source, self-hosted, Hugging Face-compatible AI model registry and distribution layer for enterprise inference. It accelerates vLLM and SGLang on Kubernetes with pull-once-serve-all caching, air-gapped delivery, RBAC and multi-tenancy, and cross-region replication.
MatrixHub was accepted to CNCF on September 28, 2026 at the
Sandbox maturity level.
