MatrixHub is an open-source, private AI model registry. It acts as a drop-in Hugging Face replacement optimized for vLLM and SGLang at scale.

• Inference Speed: "Pull-once, serve-all" caching enables 10Gbps+ distribution to 100+ GPU nodes.
• Air-Gap Ready: Seamless model "ferrying" into isolated networks via native HF_ENDPOINT support.
• Private AI model registry: Unified weight management with Tag locking and CI/CD for dev-to-prod consistency.
• Geo-Sync: Automated, resumable replication across global data centers for low-latency access.

• 🚀 Distribution: Transparent HF Proxy, on-demand caching, and direct-to-GPU NetLoader.
• 🛡️ Governance: Multi-tenant RBAC with SSO, audit logging, and malware scanning.
• 🌍 Infrastructure: S3/NFS support, Kubernetes-native (Helm), and P2P-ready scaling.

Early stage. Core features are under active development.