Publishing

When a notebook is something you want out in the world (a portfolio piece, a model card, a reproducible writeup), you can publish it beyond Clusy. Two destinations are built in: Hugging Face and Kaggle.

Connect the account first

Both live under Settings → Integrations. Hugging Face connects over OAuth, so you authorize it once. Kaggle uses an API key you add in the same place. Once connected, the publish options open up.

Connecting is only about pushing work out. Pulling work in needs nothing: the agent searches both hubs for public datasets on its own. See Find a dataset.

Hugging Face

With Hugging Face connected, you can push a notebook or a set of files to a model or dataset repo under your account. You name the repo, choose whether it's private (the default), and optionally write a README or model card to go with it. For a model publish, Clusy shows your workspace files with the likely model files pre-selected, and the files upload straight from the sandbox to Hugging Face, so large weights never make a detour through Clusy.

It's a natural home for a trained model and the notebook that produced it, where others can find and build on them.

Kaggle

With Kaggle connected, you can publish three kinds of things:

• a notebook, pushed as a Kaggle kernel, with toggles for keeping it private and for letting it access the internet when Kaggle runs it (needed for pip install);
• a model, with a framework label (PyTorch, Transformers, scikit-learn, and so on) and a variation name;
• a dataset, from files you pick out of the workspace.

Model and dataset publishes use the same workspace file picker as Hugging Face, and Kaggle accepts uploads up to its 200 GB limit.

How this differs from a share link

A share link is a read-only view Clusy hosts. Publishing pushes your work onto an external platform under your account there, with that platform's audience and versioning.