hot-notebook-patching

A JupyterLab extension for hot patching notebooks that are edited externally, for example by an AI coding agent like Claude Code.

Instead of triggering a full notebook reload (which restarts the kernel and clears widget state), only the cells that actually changed are patched into the live notebook model.

https://github.com/user-attachments/assets/6e287a55-41ff-48c0-a48c-f7a56d708413

What it does

Claude Code edits notebook.ipynb on disk
          ↓
Extension detects the file change
          ↓
Only modified cells are patched
          ↓
Kernel keeps running · Widget state intact · Scroll position preserved

Installation

Requires JupyterLab 4.

pip install hot-notebook-patching

Then restart JupyterLab (or start it fresh) — the extension is prebuilt, no jupyter labextension build step needed.

To uninstall:

pip uninstall hot-notebook-patching

Usage

Toolbar button

Each notebook gets a ⚡ bolt button in the toolbar. Click it to toggle auto-patch mode — the extension will watch the file for changes and patch automatically.

Commands

What is preserved

• Running kernel
• Widget state (ipywidgets, anywidget, …)
• Cell outputs and execution counts
• Scroll position

What is updated

• Cell source code (with line-level diff indicators)
• Cell metadata
• Added / deleted / reordered cells

Development

# Terminal 1: rebuild on every source change
source .venv/bin/activate
jlpm watch

# Terminal 2: start JupyterLab
source .venv/bin/activate
jupyter lab

After each rebuild, hard-refresh the browser tab (Cmd+Shift+R) — no server restart needed.