FAQ

Is Clusy just Jupyter with a chatbot?

No. The notebook surface is familiar, but the agent runs the work loop, branches copy live kernel state for parallel experiments, and dual-face cells hold both Python and a readable description.

Do I have to let the agent write everything?

No. Write, edit, and run cells yourself whenever you want. The agent continues from the notebook you leave behind.

Can I talk to the agent while it is working?

Yes. A message can steer the active turn, wait in the queue, or stop the run and send immediately. See Stepping in.

Where does code run?

In an isolated sandbox, not on your computer. CPU is available on every plan; its time consumes the Free allowance and is unmetered on paid plans. GPUs require a paid plan and are metered while running.

What is unmetered on a paid plan?

Auto, DeepSeek V4.1 Flash, and CPU runtime. Other built-in models, GPUs, and web tools consume the allowance. On Free, Auto and CPU consume the free allowance because there is no lower floor. See Usage.

Can I avoid a surprise bill?

Pay-as-you-go is off by default. On paid plans, exhausted metered work falls back to the unmetered CPU path. If you enable pay-as-you-go, you set a monthly cap.

On a paid plan, crossing the included allowance does not cut off a run already underway; the next metered start applies the fallback. Free instead ends the turn gracefully at its remaining allowance, and a pay-as-you-go cap is the hard paid-overage ceiling.

Which built-in models can I use?

Free has Auto; Plus adds the named DeepSeek V4.1 Flash and V4 Pro, and GPT-6 Luna; Pro adds the Kimi, GLM, and Qwen open models; Max adds first-party Claude and GPT. Choosing a model has the exact current list.

Can I use Claude or GPT on Free?

Yes, outside the built-in plan catalog. Bring your own key bills Anthropic or OpenAI directly, or use a paired local runner against the CLI subscription on your machine. Their model inference does not consume Clusy allowance; sandbox and web-tool usage still can.

Do I lose variables when a sandbox shuts down?

Normally, no. Clusy checkpoints variables when everyone leaves and restores them when the project returns. Each branch checkpoint is capped at 4 GiB, and unpicklable live objects such as open handles, connections, and CUDA contexts are excluded. See Saving & restoring state.

When exactly is kernel state saved?

At the leave boundary and when you choose Transfer state during a sandbox change. It is not checkpointed after every cell or fork, which keeps execution and branching fast. A sudden sandbox loss can lose variables and workspace files written since the last completed boundary; use files to make results reproducible, then complete a boundary or export critical artifacts for immediate durability.

What happens when I switch CPU, GPU, or RAM?

After staging the new sandbox settings and selecting Apply, choose Transfer state to carry variables or Start clean to reset them. Notebooks, files, packages, and visible outputs remain either way. See Runtimes & sandboxes.

Can I change the sandbox type later?

Only until the first successful cell run. The type—Base, Machine learning, Documents & OCR, or Web—then locks permanently. Hardware and RAM remain changeable.

What does the plan storage limit count?

Only saved kernel checkpoint bytes across projects you own. Notebooks, uploads, datasets, workspace files, outputs, and packages do not count. Manage it in Settings → Storage. See Limits.

How many branches can I have?

Three experiment branches in a CPU sandbox or eight in a GPU sandbox, counted across the project. The cap is hard, and deleting a branch frees its slot. See Branches.

How is a Clusy branch different from a Git branch?

A Clusy branch is a live experiment lineage inside a notebook and copies kernel state. A Git branch versions files. Use Clusy branches to compare approaches and GitHub for durable history.

Can I use my own data?

Yes. Upload or attach datasets, then the agent stages them into the sandbox workspace. Files do not count toward saved-kernel-state storage, though upload and workspace capture paths have individual limits. See Data & files.

What are the gray suggestions in a code cell?

They are inline completions. Tab accepts the suggestion, ⌘→/Ctrl+→ accepts one word, and Esc dismisses it. Disable them in Settings → Models.

Can I show my work to someone?

Create a share link. A branch share is linear; a whole-Lab share preserves the captured branch graph. Shares carry cells and optional outputs, not kernel variables, so a recipient who forks reruns cells to rebuild state.

Can I get notebooks out of Clusy?

Yes. Download .ipynb, commit to GitHub, or publish to Hugging Face or Kaggle.

Can the agent reach a data warehouse or private tool?

Connect Databricks or Snowflake for warehouse queries. Add MCP servers for other tools, with per-tool approval.

How do I sign in?

With an email one-time code, Google, or GitHub. There is no password.

How do I delete an account or project?

Account deletion is under Settings → Account; project deletion is on the project. Both are permanent and require confirmation.

How do I report a problem?

Use Settings → Help & feedback. Include what you did, what you expected, and what happened.