For researchers

marimo is an open-source reactive Python notebook designed specifically for reproducible computation, rapid data manipulation, and interactive science communication on the web. It solves long-standing problems with traditional Python notebooks while also giving researchers new capabilities:

• Reproducible execution and packaging. Run a cell and marimo runs affected cells (or marks them as stale), eliminating hidden state and hidden bugs that affect traditional notebooks.
• Interactivity. UI elements like sliders, scatterplots, and infinitely scalable tables are automatically synchronized with Python, letting you experiment with data in new ways.
• Pure Python file format. A Python file format makes it easy to version with Git, run as scripts, or reuse as modules.
• Share as web apps. Every notebook optionally doubles as an interactive web app that can be published on GitHub Pages or deployed on a server.

Minimum-distortion embedding

Akshay Agrawal | Stanford

View paper

Agentic Visual Reporting

Péter Gyarmati | Vienna, CMU

View paper

Signal Decomposition

Bennet Meyers | Stanford

View paper

marimo was originally designed by Stanford scientists, for Stanford scientists. Today, marimo is used at companies, labs, and classes around the world, has been downloaded millions of times, and runs everywhere: locally, in VS Code, in the cloud, and even inside web pages like this one. Try marimo below!

Features

Reproducible by default

Traditional notebooks suffer from a reproducibility crisis. One study found that less than a quarter of Jupyter notebooks were executable, and less than four percent of those were reproducible. The default experience in Jupyter notebooks accumulates hidden state: run a cell and the kernel imperatively mutates memory; the user must remember to explicitly run affected cells, a tedious and error-prone process. While there are steps the practiced scientist can take to make Jupyter-driven work reproducible, they require awareness of Jupyter's many pitfalls as well as the know-how and discipline to steer clear of them.

In contrast, marimo notebooks are"reproducible by default" (a phrase we borrow from Pluto.jl). Run a cell and marimo reactively runs affected cells based on a statically parsed dataflow graph (for expensive notebooks, users can disable automatic execution, with marimo marking affected cells as stale instead of automatically running them). Install a package, and marimo can not only serialize dependencies in your notebook file, but also create ephemeral, isolated virtual environments, making your notebooks reproducible down to the packages.

To learn more, see the marimo feature published in Nature.

Interactivity with sliders, scatter plots, and more

marimo notebooks come packaged with UI elements, which are automatically bound to Python code. Make a selection on a scatterplot, and marimo automatically sends the selected points back to Python as a dataframe, making data tangible in a way that it just isn't in traditional notebooks.

Explore data with infinitely scalable tables

marimo's built-in dataframe viewer lets you search, sort, filter, and page through as much data as fits in your machine's RAM, which on modern machines could easily be terabytes.

Create custom widgets with anywidget

If you know "just enough JavaScript", you can extend marimo with custom UI elements or "widgets", like the matrix widget below.

You can use bespoke widgets to accelerate domain-specific workflows such as data labeling or geospatial data analysis. If you don't know JavaScript, you can still benefit from marimo's ecosystem of third-party widgets, which is compatible with anywidget. Check out the wigglystuff project for many examples of custom widgets (including the Matrix widget).

Easily share on the web

Researchers can share marimo notebooks on the web in one of many ways:

• export to ipynb or HTML;
• embed in web pages;
• link to notebooks hosted on GitHub;
• publish on molab, our free cloud hosted service;
• deploy on a server, using our CLI and your own backend.

Use software engineering best practices

marimo sets you up for success for industry or academia by gently encouraging software engineering best practices: marimo notebooks are stored as pure Python, so they can be version controlled with Git; they are testable with PyTest; and they encourage functional-style programming with minimal mutations.