CSGHub Server is the backend of CSGHub, OpenCSG’s open-source platform for managing large-model assets โ models, datasets, and application spaces โ in the style of Hugging Face. Published on GitHub under the Apache-2.0 license with about 1,039 stars, it is written in Go and remains under very active development in 2026. Organizations use it to run their own private model hub instead of depending on a public one.
The server exposes a REST API through which users manage the full life cycle of LLM assets: uploading and versioning models and datasets, running model inference, fine-tuning, and hosting application spaces. It handles user and organization management, automatically tags model and dataset labels, and provides search across users, organizations, models, and data. Practical details are covered too, from online preview of dataset files such as Parquet, to text and image content moderation, to downloads of individual files including Git LFS objects. Activity data such as downloads and likes is tracked for every asset. A Docker Compose setup deploys the service quickly on your own infrastructure.
In spirit, CSGHub brings the code-hosting playbook to AI assets: models, datasets, and live AI applications get the same versioned, searchable, permissioned treatment that code gets on a forge. Teams can mirror public assets into their own environment for compliance, govern who may see or download sensitive weights, and keep track of how assets are used. A recorded demo and quick-start documentation make evaluation straightforward, and the server is one part of the wider open-source CSGHub project.
CSGHub Server is free and open source under the Apache-2.0 license. OpenCSG also operates a public hosted portal where the platform can be tried without self-hosting.
Enterprises, research labs, and AI teams that want a private, Hugging Face-style hub for their models and datasets, with full control over storage, access, and compliance.