marimo — reactive notebook for Python (github.com)

Marimo is a new reactive Python notebook that reframes notebooks as reproducible, git-friendly Python programs rather than JSON blobs. Run a cell or interact with a UI widget and marimo automatically re-runs dependent cells (or marks them stale), scrubs deleted-cell state, and enforces a deterministic execution order derived from variable references rather than cell positions. Notebooks are stored as plain .py files (first-class SQL support), can be executed as scripts or deployed as interactive web apps (including WASM in-browser execution), and convert from Jupyter — positioning marimo as an all-in-one replacement for Jupyter, Streamlit, jupytext, ipywidgets, and more. For data and ML workflows marimo offers built-in SQL that mixes with Python values, an interactive dataframe viewer that can page/filter millions of rows, and a lazy runtime mode to avoid accidentally running expensive computations. It includes package-management-on-import with isolated venvs, testability via pytest on notebooks, static analysis to run only needed cells, and AI-native features (data-aware code generation, Copilot integration, customizable prompts, or local models). The result is a more reproducible, performant, and shareable environment for experimentation, storytelling, and deployment — especially useful for teams needing deterministic, auditable notebooks that double as scripts and apps.