Cleon

Cleon brings the magic of LLM Agents to Jupyter notebooks. Have interactive conversations about your notebook and even have agents augment and run cells using the Jupyter extension.

Bring your own tokens via Codex, Claude or Gemini subscriptions or standard token apis.

Features

• invocation is as low friction as typing a single configurable prefix in a code cell
• run code at the same time while waiting for agents to respond
• queue agent prompts and approve actions just like cli agents

Default Prefixes

: hi codex
~ hi claude
> hi gemini <- currently slow

Installation

Default (extension included, no Jupyter)

pip install cleon

This installs the Python magics and backend plus the prebuilt Cleon extension package, but does not pull in JupyterLab itself (so it won’t reinstall Jupyter if you already have it).

Launch with the bundled helper (creates/uses ~/.cache/cleon/jupyter-env):

cleon jupyter lab

You can also run cleon jupyter notebook if you prefer the classic UI.

The launcher will install JupyterLab + the Cleon extension into its managed env if they’re missing. If you install the Cleon Jupyter extension while Jupyter is running, restart Jupyter (and refresh the browser) so the extension loads.

Full install into your current env

pip install "cleon[jupyter]"

This pulls in JupyterLab 4+ and the prebuilt Cleon extension into your active environment.

Minimal install

Need to skip Jupyter/extension? Install without deps and add only what you want:

pip install --no-deps cleon

There is a placeholder extra cleon[no-jupyter], but it’s a no-op; use --no-deps for a truly lean install.

Usage

Codex

• Make sure your codex is already authed, or copy your ~/.codex/auth.json from another machine.

Claude Code

• Install npm install -g @mariozechner/pi-coding-agent
• Run cleon.login()
• Complete auth token flow

Sessions

• Currently supports codex sessions, try cleon.stop() and cleon.resume()

Options

• cleon.status()
• cleon.sessions()
• cleon.resume()
• cleon.stop()
• cleon.mode("learn")
• cleon.mode("do")

Bugs or Feedback

• Twitter/X: x.com/madhavajay
• Blog: madhavajay.com