World's First Undoable Notebook
summary
Claims an automated, lightweight checkpointing layer for Python notebooks that records incremental checkpointing at a co-variable granularity, detects deltas by monkey‑patching the session dict, and stores only changed objects so undo and branchlike checkouts are cheap. Unserializable objects fall back to cell replay while picklable deltas are restored from a compact history. Demo shows a working JupyterLab extension and a pip-installable package; current limits include no tracking of state outside the notebook process.
AI-written summary · about
chapters (12)
- 0:00Introduction
- 1:05Notebook session frustrations
- 3:32Limitations of recovery
- 8:27Ku incremental checkpointing
- 11:13Co-variable granularity
- 15:25Tracking state deltas
- 18:37Managing checkpoints
- 21:00Demo of Ku in Jupyter Lab
- 24:30Object-level deduplication
- 25:13Tissue board visualization
- 25:56Conclusion and future
- 26:16Q&A
related talks
11mUndo all the things!
27mThe Future of Notebooks in a Claude Code World
19mMultiverse Notebook: Shifting Data Scientists to Time Travelers
49mUntitled13.ipynb
36mReactive Notebooks for Python
25mThe First Notebook War
36mRewind: Python Time-Travel Debugger
33mSignificant Roadblocks to Usefulness for Jupyter Notebooks and a Recipe to Fix them