Browser-based AI workflows in Jupyter
summary
Extension of JupyterLite brings Python into the browser and adds an AI stack: a chat widget with tool calls, an agent discovers 437 Jupyter commands and can create and edit notebooks, a cell-diff UI that shows and accepts suggested code, and an approval prompt before executing arbitrary actions. It runs local models via web llm and transformers.js (examples use Llama 3 with a local LLM download ~1GB), explores MCP remote tools, and prototypes a Jupyter secret manager; demos generally work though large downloads and occasional errors appear.
AI-written summary · about
chapters (11)
- 0:00Browser-based AI in Jupyter
- 3:42JupyterLite and remote models
- 6:57Live AI code completion demo
- 8:01Agent mode and tool calls
- 14:27Model Context Protocol
- 16:57Browser-only language models
- 18:42Built-in AI in browsers
- 21:31Privacy and in-browser AI limits
- 22:37Securing AI workflow secrets
- 24:04JupyterLite AI future steps
- 25:36Q&A on model config
related talks
35mJupyterLite: Jupyter ❤️ WebAssembly ❤️ Python
39mJupyterLite: run all your code in a web browser using WebAssembly
27mBuilding web-based engineering applications with JupyterLab components
30mDistributing AI with Python in the Browser: Edge Inference and Flexibility...
53mAI Didn’t Kill the Web, It Moved in!
28mCollaborative editing in Jupyter
29mJupyterLab+Real Time Collaboration
20mPossible Futures For Jupyter