A Go backend for running your own private model hub

What it does

CSGHub Server is the Go backend behind the CSGHub platform. It exposes REST endpoints for managing users, organizations, models, and datasets, and it tracks activity like downloads and likes. The server also handles file delivery—including Git LFS objects—and can preview dataset files such as .parquet directly in the browser.

The interesting bit

Rather than reinvent version control, it plugs into existing Git servers like Gitea (with GitLab planned) and uses standard Git LFS for large model files, while letting you swap in any S3-compatible store for object storage. It also auto-tags models and datasets, and can hand off to an inference runner for one-click model deployment.

Key highlights

• Treats models and datasets as Git repositories with LFS support, so you can use normal Git workflows for multi-gigabyte files.
• Pluggable storage: local disk or any S3-compatible service like MinIO.
• Built-in moderation hooks for text and images, optional and swappable with third-party providers.
• Auto-extraction of metadata tags for models and datasets to keep the catalog searchable.
• Integrates with OpenCSG’s inference stack for one-click model serving.

Caveats

• Only Gitea is supported as a Git backend today; broader Git server support remains on the roadmap.
• Native model-format conversion is still an unchecked roadmap item.
• The README notes testing only on Ubuntu 22, so portability is unverified.

Verdict

Teams that want an on-premise catalog for LLM weights and datasets—and already run Gitea or S3-compatible storage—will find this a practical starting point. If you are looking for a fully managed SaaS experience or deep MLOps experiment tracking, this is not it.