MatrixHub

MatrixHub is an open-source, self-hosted model registry designed for large-scale enterprise AI inference. It functions as a private replacement for Hugging Face, optimized specifically for vLLM and SGLang workloads. The project describes itself as a drop-in Hugging Face alternative that lets organizations control their AI assets without relying on public internet infrastructure.

Key features include transparent Hugging Face proxy compatibility, on-demand caching that localizes models across clusters, role-based access control with audit logging, and storage flexibility across local filesystems, NFS, and S3-compatible backends. It supports air-gapped deployment, multi-region synchronization, and integrates with Kubernetes via Docker Compose or Helm. The material indicates the project was built using Claude-assisted development.