Quickstart

This gets you through your first real piece of work: sign in, create a project, and have the agent build and run a notebook. A few minutes, start to finish.

Sign in

Open the app and sign in. There's no password. You get a one-time code by email, or you sign in with Google or GitHub. If it's your first time, you'll go through a short setup (your name, what you're here for, a theme) and land in your first project. You can skip it.

Create a project

A project is one body of work, like a dataset you're exploring or a model you're building. From the dashboard, hit New project and name it.

You don't need to set up compute first. New projects run on CPU, which is free and fine for loading data, cleaning, plotting, and most classical ML. Attach a GPU later, per cell, when something needs it.

Say what you want

You start in chat, before there's a notebook. It's usually easier to describe what you want than to set up the first cell yourself:

Load customers.csv, show me the shape and column types, and flag any columns with missing values.

The more specific you are, the less back-and-forth. If you have a dataset, upload it first so the agent can read it.

Watch the notebook build

The agent reads the request, sketches a plan, and starts writing cells. Once there's something to run, the notebook canvas opens on the right. You'll see each cell appear, run, and produce output.

At any point you can click into a cell and edit it, run it again, or write a new cell by hand. The agent picks up from wherever you leave off.

Steer it

Read what came back and react in plain language:

• "The date column is a string, parse it."
• "Plot the missing-value counts as a bar chart."
• "Split into train and test and fit a logistic regression baseline."

Each message is another turn. The agent edits or adds cells and reruns what it needs to.

Try a second idea without losing the first

When you hit a fork (say you want a random forest alongside that logistic baseline), don't overwrite. Branch from the cell where the two diverge. Clusy keeps your loaded data and fitted objects in the new branch, so nothing recomputes, and you can switch between branches to compare.

Keep your work

When you've got something worth keeping, you can commit it to GitHub, share a read-only link, or publish to Hugging Face or Kaggle.

Next