@jupyter-ai/agent-lite

The building blocks of in-browser AI personas for Jupyter AI, in JupyterLab, Notebook 7 and JupyterLite. The Python package is jupyter-ai-agent-lite.

The package is a JupyterLab extension that provides:

• IToolRegistry: the tools a persona offers to a model, with the built-in discover_commands, execute_command, browser_fetch, discover_skills and load_skill tools
• ISkillRegistry: the agent skills loaded from the skills paths or registered by extensions
• IAgentSettings: the settings shared by the personas (@jupyter-ai/agent-lite:settings): the commands that require an approval, the commands whose outputs render in the chat, the trusted MIME types and the skills paths
• helpers: getToolInputSchema, parseToolInput, extractMimeBundles, requestApiKey and forwardAuthCode (OpenRouter sign-in), getAppAttribution

The package has no dependency on the AI SDK.

Usage

A persona extension consumes @jupyter-ai/agent-lite as a shared singleton:

{
  "dependencies": {
    "@jupyter-ai/agent-lite": "^0.21.0"
  },
  "jupyterlab": {
    "extension": true,
    "sharedPackages": {
      "@jupyter-ai/agent-lite": {
        "bundled": false,
        "singleton": true
      }
    }
  }
}

Its Python package depends on jupyter-ai-agent-lite, so that the extension is installed.

import { IToolRegistry } from '@jupyter-ai/agent-lite';

const plugin: JupyterFrontEndPlugin<void> = {
  id: 'my-persona:tools',
  autoStart: true,
  requires: [IToolRegistry],
  activate: (app: JupyterFrontEnd, tools: IToolRegistry) => {
    for (const { name, tool } of tools.namedTools) {
      console.log(name, tool.description);
    }
  }
};

See Building an in-browser persona for the complete guide.