Outputs can contain more than code
A Jupyter notebook can save table rows, errors, file paths, images, HTML and widget state from earlier runs. This cleaner clears code-cell outputs, execution counts and saved widget state without running a cell.
Read the source before sharing
Code, Markdown and raw-cell source stay unchanged. A token written in a code cell or a name in Markdown remains. Optional metadata cleanup keeps basic kernel and language details so editors can still recognize the notebook. Other top-level fields are retained for compatibility and need manual review.
Attachments often hold images used by Markdown cells. They remain by default, and the report counts them for review. Removing them can break those images and does not remove references from the source.
Supported notebooks
The tool reads nbformat 4 JSON notebooks up to 16 MiB and 20,000 cells. Malformed cells are rejected. Files stay in the browser and are never executed, uploaded or rendered as notebook HTML.