The Open Source
Hub for AI Models
MatrixHub is an open-source, self-hosted AI model registry engineered for large-scale enterprise inference. It serves as a drop-in private replacement for Hugging Face, purpose-built to accelerate vLLM and SGLang workloads.
MatrixHub is to Hugging Face what Harbor is to Docker Hub.
Stop relying on public internet for mission-critical AI. Control your assets, accelerate your pipelines.
Core Features
Infrastructure designed for Scale
Built for the specific needs of SREs and Algorithm Engineers managing massive model weights.
Transparent HF Proxy
Drop-in replacement for Hugging Face. Point your HF_ENDPOINT to MatrixHub and keep all training/inference code unchanged.
On-Demand Caching
Pull once, cache forever. Automatically localizes public models to slash redundant traffic and accelerate cluster-wide distribution.
RBAC & Audit Logs
Fine-grained permissions, project-based isolation, and comprehensive audit trails for every upload and download.
Storage Agnostic
Compatible with local filesystems, NFS, and S3-compatible backends (MinIO, AWS). Scale to unlimited model capacity.
Key Use Cases
How organizations use MatrixHub in production.
Zero-Wait Distribution
Eliminate bandwidth bottlenecks with a 'Pull-once, serve-all' cache. Achieve 10Gbps+ speeds across 100+ GPU nodes simultaneously.
Air-Gapped Delivery
Securely ferry models into isolated networks with integrity protection, malware scanning, and comprehensive audit trails.
Private Registry
Centralize fine-tuned weights with tag locking and CI/CD integration. Guarantee consistency from development to production.
Global Multi-Region Sync
Automate asynchronous, resumable replication between data centers for high availability and low-latency local access.
SEAMLESSLY INTEGRATED WITH
vLLM
SGLang
KubernetesReady to take control of your AI Models?
Deploy MatrixHub in minutes using Docker Compose or Helm. Open source and free for the community.