Skip to main content

Overview

MatrixHub is an open-source, self-hosted, Hugging Face-compatible model hub for enterprise inference infrastructure.

Target Users​

  • 🏗️ Platform Engineers: Operate enterprise AI infrastructure.
  • 📦 ML Infrastructure Teams: Manage internal model distribution.
  • 🛡️ AI Operations Teams: Own release control, audit, and replication.
  • 🔬 Research Teams: Work in isolated or regulated environments.

Use Cases​

  • 🚀 Intranet Inference Acceleration: Cache models once and distribute them efficiently across internal GPU clusters.
  • 🔐 Air-Gapped Model Delivery: Transfer approved models into isolated or regulated environments.
  • 📦 Enterprise Model Governance: Control releases, audit activity, and manage model versions from development to production.
  • 🌍 Multi-Region Replication: Replicate models between data centers with resumable transfers.

Key Capabilities​

🚀 Model Distribution​

  • Hugging Face-Compatible Access: Point HF_ENDPOINT to MatrixHub to use hf download and hf upload, and connect inference engines such as vLLM, SGLang, and Dynamo.
  • On-Demand Proxy Cache: Cache public Hugging Face models on first request.
  • Repository Workflows: Create and manage model repositories.

🛡️ Access and Project Management​

  • Project Isolation: Organize repositories by project and permissions.
  • Authentication and Roles: Manage users, roles, tokens, and robot accounts.
  • Git Access: Access repositories through Git, Git LFS, and HTTP.

🌍 Deployment and Replication​

  • Replication Policies: Run manual or scheduled pull and push synchronization, with a foundation for resumable transfers.
  • Flexible Storage: Use local PVC or NFS storage.
  • Deployment Options: Deploy with Docker Compose or Kubernetes Helm charts.

For planned capabilities, see the MatrixHub roadmap.

Live Demo​

Try MatrixHub instantly at demo.matrixhub.ai—no setup required.

Sign in with the public demo credentials: username admin, password changeme. The demo is for evaluation only, resets every six hours, and may also be reset during maintenance.

Getting Started​

MatrixHub is easy to deploy using Docker Compose or Kubernetes. The entire infrastructure is open source and free for the community.

👉 Ready to get started? Follow the Getting Started guide to deploy MatrixHub and begin using it.