marimo
marimo is a Python notebook environment for data exploration, SQL queries and model work. It stores notebooks as Python files and combines an interactive editor with reactive execution. The publisher describes the notebook as free and open source, with options to run the same work as a script or share it as an app.
Keep dependent cells in sync
Running a cell causes marimo to rerun affected cells that depend on it. Deleting a cell removes its variables and updates dependent work. For expensive computations, you can configure marimo to mark outputs as stale instead of rerunning them automatically.
Built-in sliders, dropdowns, dataframe tables and selectable plots let you adjust inputs while exploring data. SQL cells work with dataframes and databases, including Pandas, Polars, DuckDB, SQLite and Postgres. This combination suits an analysis where changing an input should update related calculations and figures without manually choosing an execution order.
Version, test and share Python files
marimo notebooks use .py files. The Git feature page explains that these files omit cell outputs and notebook metadata, making small code changes easier to inspect in a diff. You can export to HTML or IPYNB when you want to preserve outputs.
The publisher also describes optional pytest integration for cells containing test code. That gives teams a way to keep checks alongside an analysis and run them in CI; it is a product feature, rather than a claim that the notebook will validate an analysis on its own. Notebooks can run as scripts or modules, become CLI-served web apps, or export to WebAssembly-powered HTML for web sharing.
Add AI help when useful
The AI feature page lists code completion, error-fix suggestions with diffs and an integrated chat. You can give the assistant context by referring to variables and database tables, and add preferences in user settings.
marimo supports several hosted providers, Ollama and OpenAI-compatible APIs. Its marimo pair skill connects coding agents to a running notebook, while the Agent sidebar supports ACP-compatible agents. Those agents can inspect state and modify cells, so choose the access and provider that fit your data and workflow. The captured pages do not promise that hosted model usage is included in the free notebook. They separately describe molab as a free hosted notebook with built-in AI assistance.