Install

Pip-install into any JupyterLab 4 environment, then configure your provider of choice.

Prerequisites

  • Python 3.10 or newer. NBI follows Jupyter’s supported-Python window.
  • JupyterLab 4.x. NBI is a labextension, so it ships pre-built and loads on the first JupyterLab restart.
  • A provider — at least one of: Claude Code CLI, GitHub Copilot account, a running Ollama server, or an OpenAI-compatible endpoint.

Install the extension

pip install notebook-intelligence

Then restart JupyterLab. The NBI sidebar appears on the left.

For conda or pixi-managed environments, install into the same env that runs jupyter lab. NBI does not need a separate Python kernel.

Pick a provider

NBI supports four model providers; you can configure one or several. Open Settings → NBI and follow the instructions for your chosen path.

Claude Code (recommended) Install the Claude Code CLI and sign in: ```bash npm install -g @anthropic-ai/claude-code claude login ``` If `claude` is not on your `PATH` (for example, in a non-default Node install), set `NBI_CLAUDE_CLI_PATH` to the full path of the binary before starting JupyterLab. In NBI Settings, toggle **Claude mode** on and pick a chat model.
GitHub Copilot In the NBI chat sidebar, click **Sign in with GitHub** and follow the device-flow prompt in your browser. NBI stores the access token at `~/.jupyter/nbi/copilot.json`. Set `NBI_STORE_GITHUB_ACCESS_TOKEN_POLICY=force-off` to keep it in-memory only.
Ollama Install Ollama and pull a chat-capable model: ```bash curl -fsSL https://ollama.com/install.sh | sh ollama pull llama3:latest ``` Then pick **Ollama** as the chat provider in Settings. NBI talks to the default Ollama endpoint at `http://localhost:11434`. If Ollama runs elsewhere, set `OLLAMA_HOST`.
OpenAI-compatible (vLLM, LiteLLM, hosted) In Settings, pick **OpenAI-compatible** and provide: - a **base URL** (e.g. `https://your-gateway.internal/v1`) - a **model ID** (e.g. `gpt-oss-120b`) - an **API key** The base URL must speak the Chat Completions API. vLLM, LiteLLM, TGI, and llama.cpp's HTTP server all qualify; anything in front of a real OpenAI-compatible LLM will work.

First run

  1. Open the chat sidebar. The NBI icon appears on the left activity bar.
  2. Send a prompt. Type a question and hit return. The first response confirms your provider is wired up.
  3. Try inline chat. Click into a code cell and hit Cmd+I (or Ctrl+I on Linux/Windows) to open the inline popover.
  4. Try agent mode. Click the sparkle icon on the active notebook toolbar to scope a generation to that notebook. The agent will plan, write, and run cells.

Troubleshooting

If something doesn’t work, check:

  • jupyter server extension list — notebook_intelligence should appear under “OK”.
  • jupyter labextension list — @notebook-intelligence/notebook-intelligence should be listed as enabled.
  • The JupyterLab terminal output where you ran jupyter lab. NBI logs server-side errors there.

For deeper issues, see the troubleshooting guide in the repo or open an issue.