// the find
Kanaries/pygwalker
PyGWalker: Turn your dataframe into an interactive UI for visual analysis
PyGWalker drops a Tableau-style drag-and-drop UI into your Jupyter notebook with one function call. Point it at a pandas, polars, or Spark dataframe and you get an interactive chart builder without leaving the notebook. It's for data scientists who want exploratory visualization without writing matplotlib boilerplate or standing up a separate BI tool.
The zero-config entry point is genuinely good — `pyg.walk(df)` is all you need to start. The DuckDB kernel computation mode (`kernel_computation=True`) is a real win: it lets you work with datasets that don't fit in memory without spinning up anything external. Chart state serializes to a JSON spec file, so you can persist and reload your work across sessions. Support across Jupyter, Streamlit, Databricks, VS Code, marimo, and Panel means it's not locked to one environment.
The underlying Graphic Walker is a React app bundled into the Python package, so the frontend and Python layers are loosely coupled and debugging communication failures is painful — you're dealing with two entirely different codebases. The `spec` autosave is still manual (you have to click Save in the UI), which means unsaved state is lost on kernel restart; the README acknowledges this as a known gap. The Kanaries cloud features and token system create a soft lock-in path toward a paid tier, which is fine until you realize some advanced sharing features quietly require it. Natural language query support is mentioned but thin — it's more of a marketing bullet than a production feature in the open-source build.