May 29, 2026 · Taste Test

Jupyter AI for Research

The story of Jupyter AI, from its origins at SCU to today, followed by a live tutorial on building AI agents for research and general applications. Sanjiv Das (developer of Jupyter AI's quick agent feature and various AI models) and David Qiu (creator and lead maintainer of Jupyter AI) present together, with a live Q&A led by David at the end.

Presenters: Sanjiv Das (Terry Professor of Finance & Data Science, SCU) & David Qiu (Creator & Lead Maintainer, Jupyter AI) · Lucas 306 · 1:30 to 5:00 p.m. · Live tutorial at 2:30

Watch the recording

Watch on YouTube →

What you'll see

Sanjiv covers the history and architecture, then both presenters demo live agents. David leads Q&A at the end.

The history of Jupyter AI, including its origins at Santa Clara University, and how the project grew into a widely used open-source tool.
How Sanjiv's quick agent feature lets you spin up lightweight research agents that compose skills for multi-step tasks.
How these AI agents can be applied to both research (formulating questions, running analysis, drafting papers) and general applications.
Live Q&A with David Qiu, creator of Jupyter AI, on the project's roadmap and how to get involved.

Slides

View presentation slides (live link) →

Repositories & docs

The tools demoed in this session. All are open-source and available to try after the talk.

GitHub

Jupyter AI

The core extension that brings generative AI capabilities to JupyterLab.

Docs

Jupyter AI Documentation (v3)

Installation, configuration, and usage guides for the Jupyter AI extension.

GitHub

Jupyter AI Contrib

Community-contributed extensions and integrations for Jupyter AI.

GitHub

QuickAgents

Lightweight framework for creating research agents inside Jupyter AI. The main tool demoed in this session.

GitHub

Skill Collection

Reusable agent skills that plug into QuickAgents for common research tasks.

Going further

Sanjiv Das's website · Research, publications, and course materials from the Terry Professor of Finance and Data Science at SCU.
Sanjiv Das, candid bio · Background and research interests in Sanjiv's own words.
David Qiu on LinkedIn · Senior Engineer at AWS and creator and lead maintainer of Jupyter AI.
Jupyter AI docs · Getting started guide, model provider setup, and the chat and magic commands reference.

Kai's unofficial tips for lazy installing via Claude Code

AI Kitchen attendees have Claude Pro, which means you can use Claude Code as a free, no-API-key backend for Jupyter AI, and hand the tedious install steps to Claude Code itself. The setup below puts Jupyter AI v3 in an isolated Python environment, with Claude Code wired in as the chat backend over the Agent Client Protocol (ACP). Prerequisites: macOS with Homebrew and Node.js.

Claude Code safe to let the Claude Code agent run for you (one-shot installs) Terminal run yourself in a terminal like Ghostty (servers, global installs, interactive tools)

First: install the Claude Code CLI Terminal

Most of you already installed the Claude Code desktop app, which covers the graphical Code tab. But this Jupyter AI setup talks to the claude terminal command, and the desktop app does not add that to your PATH, so you install the CLI separately (one command). Your Claude Pro sign-in covers it (Claude Code needs Pro, Max, Team, or Enterprise; the free plan does not).

curl -fsSL https://claude.ai/install.sh | bash
claude          # opens a browser to sign in with Claude Pro

Confirm it worked with claude --version. Homebrew users can instead run brew install --cask claude-code.

1. Create an isolated Python environment Claude Code

Keeps everything off your system Python. uv makes a dedicated Python 3.12 virtualenv (3.11+ works).

brew install uv
mkdir jupyter-ai && cd jupyter-ai
uv venv --python 3.12

2. Install Jupyter AI Claude Code

uv pip install jupyter-ai jupyterlab

3. Add the Claude Code chat backend Terminal

The ACP adapter that bridges Jupyter AI to your local Claude Code. It is a global install, so run it yourself.

npm install -g @agentclientprotocol/claude-agent-acp

4. Launch and chat Terminal

A long-running server with a browser UI, so it lives in your own terminal (and the venv stays activated there).

source .venv/bin/activate
jupyter lab

Then in the Chat panel (left sidebar), type @, pick Claude, and ask away, for example @Claude what files are in this directory?

Optional: add QuickAgents

QuickAgents lets you build named agents like @QuickAgent on top of Jupyter AI. It installs into its own separate virtualenv (Python 3.11+), distinct from the one above.

5. Clone the repo Claude Code

git clone https://github.com/srdas/jupyter-ai-quickagent.git

6. Run the installer Terminal

A third-party script that builds its own venv, so run it yourself. chmod a+x makes the script executable first.

cd jupyter-ai-quickagent
chmod a+x install_nondev.sh
./install_nondev.sh

7. Launch and create an agent Terminal

Use the QuickAgents venv this time (separate from step 1).

source .venv/bin/activate
jupyter lab

Then in the Chat panel, send @QuickAgent create and follow the five prompts. Needs a chat model configured (your @Claude works).

Two separate environments: jupyter-ai (base) and jupyter-ai-quickagent (its own venv). Launch jupyter lab from whichever you want; one will not see the other. And do not paste setup commands into the terminal tab that is already running jupyter lab (that tab is busy with the server), open a new tab instead.