About
A JupyterLab extension for agentic AI — Claude Code first, with Copilot, Ollama, and OpenAI-compatible models alongside.
Notebook Intelligence (NBI) is a JupyterLab extension that brings agentic AI coding into the notebook. It runs Claude Code as a first-class provider with sessions, Skills, Plugins, and MCP servers, and ships with GitHub Copilot, Ollama, and any OpenAI- or LiteLLM-compatible endpoint as drop-in alternates.
NBI works the way JupyterLab does. Chat lives in a sidebar with @-mention participants. Inline chat opens on any cell with Ctrl+G or Cmd+G. Auto-complete suggests code as you type. Agent mode goes further: it creates notebooks, edits cells, runs them, and fixes errors without leaving the document.
The extension is built to be extended. Connect Model Context Protocol (MCP) servers to expose your own tools, databases, and APIs to the chat. Manage Claude Skills and Plugins from the UI, or sync an organization-wide manifest so every user in the lab lands with the same set on install. Drop markdown files into ~/.jupyter/nbi/rules/ to inject conventions, style guides, or guardrails into every prompt.
For administrators, NBI ships policy controls that lock specific providers, models, and endpoints via environment variables. A platform team can decide what a managed JupyterHub allows before anyone signs in — each capability has a force-on / force-off / user-choice triad with a matching NBI_*_POLICY env var.
NBI is open source and BSD-licensed. It is maintained by Mehmet Bektaş and a growing group of contributors from the Jupyter community. The codebase lives at github.com/plmbr/notebook-intelligence.