GitHub Copilot

Use any Copilot chat or completion model from inside JupyterLab.

NBI was originally built around GitHub Copilot, and Copilot remains a first-class provider when Claude mode is off. Chat and inline completion both go through Copilot’s chat-completions endpoint.

What you get

  • Device-flow auth. Sign in once from NBI Settings — the same GitHub device-flow you use in VS Code. NBI stores the refresh token in ~/.jupyter/nbi/copilot.json.
  • Model picker. All Copilot chat models the API exposes appear in the model dropdown, including the latest GPT and Anthropic options Copilot routes for you.
  • Inline completion. Tab-completion as you type, with configurable debounce delay and a separate model from chat.
  • Slash commands. /explain, /fix, /troubleshoot, /newNotebook, /newPythonFile work out of the box.

Setup

Open the chat sidebar, click Sign in with GitHub, and follow the device-flow prompt. 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).

Configuration

{
  "chat_model": { "provider": "github-copilot", "model": "gpt-4o" },
  "inline_completion_model": { "provider": "github-copilot", "model": "gpt-4o-mini" }
}

Reference

Install NBI