What is matrixhub?

This software provides a private, self-hosted model registry that replaces Hugging Face for enterprise AI teams. It accelerates inference engines like vLLM and SGLang by caching model weights on first pull and distributing them across GPU clusters at high throughput. The registry speaks the Hugging Face API, so existing workflows work unchanged when the endpoint is redirected. It adds project-level access control with audit trails, works with local disks, NFS, or S3-compatible object stores, and can replicate models across data centers. You can run it via Docker Compose on a single machine (SQLite by default) or on Kubernetes with Helm (MySQL). It also connects with llm-d, Dynamo, and ModelExpress for advanced model distribution and GPU-to-GPU weight sharing. The latest release is v0.2.0, still in pre-1.0 development.

Should you self-host matrixhub?

matrixhub scores 46/100 on selfhost.directory (how scoring works) and ranks #70 of 125 in AI. It has 362 GitHub stars.

It shipped 3 stable releases in the past year. The latest, v0.2.0, came out on 21 September 2026. New versions land roughly every 2 months.

Setup is moderate: expect some configuration beyond a single command. An official Docker image is available. It is Apache-2.0-licensed and written mostly in Go.

Its closest alternative on selfhost.directory is ComfyUI (93/100, 143 741 stars). It scores higher overall, so compare the two before you pick. See matrixhub vs ComfyUI side by side.