FAQ

Is Clusy just Jupyter with a chatbot?

No, though the notebook surface is familiar on purpose. The differences that matter are an agent that runs the work loop (writes cells, runs them, reads results, iterates), branching that keeps your kernel state when you fork, and dual-face cells that hold both code and a plain-language description.

Do I have to let the agent write everything?

No. You can take over any time: write cells, edit the agent's, run things yourself. It picks up from whatever you leave. Plenty of people use it as a smart pair rather than a driver.

Can I talk to the agent while it's working?

Yes, and you don't have to wait for it to finish. What happens to your message depends on how you send it: it can be folded into the running turn, queued until the turn ends, or sent after stopping the run outright. Stepping in draws all three side by side.

Where does my code run?

In an isolated sandbox with its own filesystem and kernel, not on your machine. CPU is free. GPUs are on paid plans and metered by the minute.

What's free?

On a paid plan, the Auto model and CPU runtime are never metered: run Auto on CPU all day and your allowance doesn't move.

On the free tier, the rule is different, and it's the thing people get wrong. Auto and CPU are what the free allowance buys, so every run counts against it. Nothing is ever blocked, because there's no premium model or GPU to cut off, but the meter does move. See the free floor.

What actually costs money?

Three things: calls to the metered models (everything above Auto), time on a metered runtime like a GPU, and the agent's web-tool calls. Editing, branching, running cells, and reading output are free.

Can I avoid a surprise bill?

Yes. By default, when your allowance runs out, Clusy drops you back to the free Auto-on-CPU loop instead of charging you. If you opt into pay-as-you-go, you set a monthly cap and spending stops there.

There is no per-run limit, and that's deliberate: within your allowance a run is uncapped, so an hour-long agentic GPU run isn't killed halfway through. The monthly cap is the only guardrail, and it's the one you set.

Which models can I use?

Auto on every plan; each paid tier unlocks more. Choosing a model has the current plan-by-plan mapping, and it's the page that gets updated when a model launches.

Can I use Claude or GPT on the free tier?

Yes, by two routes that skip Clusy's metering entirely. Bring your own key: paste an Anthropic or OpenAI key and those models unlock on every plan, billed by your provider. Or pair a local runner: the turn runs on your own machine against your existing Claude or ChatGPT CLI subscription. Neither spends your Clusy allowance.

Do I lose my variables when the sandbox shuts down?

No. When a sandbox goes idle and is torn down, the kernel's state (your DataFrames, fitted models, and other variables) is checkpointed and restored when you come back, so you don't re-run the notebook from the top. The one caveat: a very large namespace is skipped, and some objects can't be captured. See Runtimes.

How is a Clusy branch different from a Git branch?

A Clusy branch is a live experiment inside a notebook. It carries your kernel state and exists only in Clusy. A Git branch versions the notebook files. Use Clusy branches to explore, and Git to keep history.

Can I use my own data?

Yes. Upload datasets to the project workspace and the agent or your code reads them from disk. There's a per-plan storage limit.

Can I show my work to someone?

Publish a share link, a read-only view anyone can open with no account needed. They can read it, fork it to their own workspace, download the .ipynb, or open it in Colab.

Can I get my notebooks out of Clusy?

Always. Notebooks are .ipynb-compatible. Download them, commit them to GitHub as clean Python diffs, or publish to Hugging Face or Kaggle.

Can the agent reach my data warehouse?

Yes. Connect Databricks or Snowflake under Settings → Integrations and the agent can list your tables and run SQL against your warehouse, pulling results straight into the notebook.

How do I extend what the agent can do?

Connect MCP servers under Settings → MCP servers to give it extra tools, like querying your own systems or hitting private APIs, with per-tool approval so you stay in control.

How do I sign in?

With an email one-time code, Google, or GitHub. No password.

How do I delete my account or a project?

Account deletion is in Settings → Account, and project deletion is on the project itself. Both are permanent and both ask you to confirm.

Something's broken. How do I tell you?

Settings → Help & feedback goes straight to the team. Concrete reports (what you did, what you expected, what happened) get fixed fastest.