Notebooks

Open interactive Python notebooks in a tab next to your chat, so you can explore and visualize data without leaving the conversation. A notebook is a shared document you and the agent build together: you edit and run cells directly, while the agent reads, writes, and runs cells alongside you as you talk through the analysis. Notebooks run on a dedicated notebook runtime.

Open a notebook

A notebook opens as its own tab beside the chat. There are a few ways to get there:

• Ask the agent to list the files in your workspace, then select an .ipynb file in its reply to open it.
• From the Files tab, search or browse the workspace and select an .ipynb file.
• Ask the agent to open it by name — the agent can open and list notebooks for you, so the one you're discussing stays in view. You can also create a new notebook by asking the agent.

Work with cells

The toolbar at the top of the notebook has controls to add a code cell, add a Markdown cell, run cells, restart, restart and run all cells, and interrupt a running cell. Run a cell to see its output inline, and view the status (such as Idle) on the right of the toolbar.

Each individual cell includes controls to run it, switch it between code and Markdown, hide its input or output, and delete it.

Edit with the agent

Work on notebooks together with the agent. Ask it to make a change — fix a typo, add analysis, rework a chart — and it reads the notebook, edits or adds cells, and saves. When the notebook is open beside the chat, the results appear in its tab, and a dot marks each cell the agent changes so you can see at a glance what it edited. The notebook doesn't need to be open, though — the agent can edit any notebook in your workspace.

Environment and packages

Each notebook uses a Python environment. Select the kernel button in the toolbar to see the details of that environment. New notebooks can get a shared environment.yml in their workspace; toggle this under Settings → Notebooks.

The agent can work with conda environments too — ask it to find an appropriate conda environment for this notebook and point the notebook at it, and it will. A helpful pattern is to have the agent write an environment.yml into the workspace: then every notebook there, existing or new, picks up that environment automatically.

Runtime status

The notebook runtime starts on demand. While it's starting you'll see a brief status; if it can't be reached, the tab shows a Retry button.