Notebooks

Open any .ipynb and you get a Jupyter-style notebook UI inline — cells, run, outputs, markdown — without leaving the workspace.

Opening a notebook

Click any .ipynb file in the file tree. The standard code-preview is hidden and the notebook UI mounts in its place.

If you don't have a notebook yet, create one:

echo '{"cells":[],"metadata":{},"nbformat":4,"nbformat_minor":5}' > scratch.ipynb

Then click it in the tree.

The notebook UI

A sticky header sits above the cells with four controls:

+ Code — append a new code cell.
+ Markdown — append a new markdown cell.
Run All — execute every code cell top to bottom.
Clear — remove all outputs and reset execution counts.

Each cell has its own toolbar:

[n] — execution count (shows after a run).
▶ — run this cell.
↑ / ↓ — move up / down.
+ — insert a new cell after this one.
× — delete this cell.

Running cells

Shift+Enter — run the focused cell.
Tab inserts 4 spaces (because notebooks).
Blur — when you click out of a cell, it auto-saves.

Output rendering

The notebook UI handles all four nbformat output shapes:

Shape	Rendered as
`stream` (stdout)	Plain text.
`stream` (stderr)	Plain text in a muted-red panel.
`error`	Traceback with `ename`, `evalue`, frames.
`execute_result` / `display_data`	Priority: `image/png` → `text/html` → `text/plain`.

Output regions cap at ~360px tall and scroll. Long outputs don't blow up the layout.

Saving

Saves are atomic (.tmp → replace). The notebook is written as plain nbformat v4.5 JSON. You can open the same file in JupyterLab and everything is byte-compatible.

Execution backend

The notebook runtime tries two execution paths:

jupyter_client if installed in the workspace — gives you a full kernel with state preservation across cells.
Subprocess fallback — each cell is run as python -I -c <source> with the workspace as CWD. State does not persist across cells in this mode.

Check which is active:

curl -b cookies.txt https://your-workspace.krowforge.com/api/notebook/capabilities

To enable kernel mode:

pip install jupyter_client
# restart the workspace

Cell timeout

Each cell run has a server-side timeout (default 20s, max 300s). Long-running cells raise a TimeoutError and are killed cleanly.

Limitations

No widgets — ipywidgets isn't supported in the inline UI. Use JupyterLab for widget-heavy notebooks.
No magics other than %matplotlib inline in subprocess mode. Kernel mode supports any magic the kernel supports.
One notebook open at a time in the workspace's notebook host. Open another .ipynb and it replaces the current one.