Marimo

Overview

Marimo is an open-source notebook for Python, built as an alternative to Jupyter that fixes some of its well-known pain points: hidden state from out-of-order cell execution, notebooks that don't diff or version well in Git, and a JSON file format that's awkward to review in a pull request. A marimo notebook is a single .py file, and it's reactive — when you change a cell, every downstream cell that depends on it automatically re-runs, so the notebook state can never silently drift out of sync with its code.

Native SQL cells

Marimo has first-class SQL cells that don't require magics or separate extensions. By default, a SQL cell runs against an in-memory DuckDB instance; write a SQL cell and marimo assumes it's DuckDB unless you configure a different backend (Postgres, MySQL, SQLite, and others are also supported through standard connectors). Under the hood, a SQL cell compiles to a call to mo.sql(), keeping the notebook file pure Python even though the editor renders it as a SQL editing experience.

Copy code
import marimo as mo

df = mo.sql(
    f"""
    select category, sum(order_total) as revenue
    from orders
    group by category
    order by revenue desc
    """
)

The query result comes back as a dataframe (Polars if installed, otherwise pandas) that other Python cells can reference directly, and because SQL cells are just Python f-strings, queries can be parameterized with Python variables or UI element values, with the whole chain re-executing reactively when an input changes.

Marimo and MotherDuck

Because marimo's SQL engine is DuckDB itself, connecting to MotherDuck is a matter of using DuckDB's own MotherDuck attachment (a connection string with a service token) inside a marimo SQL cell, giving a notebook access to cloud-hosted DuckDB data with the same reactive, git-friendly workflow used for local data.