Vibe Widget

Make analysis interactive.

Vibe Widget generates interactive notebook interfaces from plain English. Explore data with sliders, linked views, filters, and custom controls without building a front end.

Checkout the docs!

What you can do

• Create widgets from a prompt
  Describe the interface you want and get a working widget immediately.

• Iterate safely
  Revise in plain language, use built-in audits, and (optionally) require approval before any generated code runs.

• Share reusable widgets
  Save widgets as .vw bundles and load them elsewhere—with review + audit on load by default.

• Run where your data lives
  Works in Jupyter/JupyterLab, Colab, VS Code notebooks, marimo, and more (via AnyWidget + React).

Quickstart

pip install vibe-widget
export OPENROUTER_API_KEY="your-key"

import pandas as pd
import vibe_widget as vw

df = pd.read_csv("sales.csv")

widget = vw.create("scatter plot with brush selection and a linked histogram", df)
widget()

Acknowledgements

This repo was originally created at the Sundai Weird Data Hack. We thank Angela for her feedback and suggestions on early versions!

Special thanks to Trevor Manz and the Anywidget project for providing the specification and foundation that made this project possible. Be sure to check out and star AnyWidget!