marimo

marimo (Marimo Inc.) develops an open-source reactive Python notebook. Running a cell or interacting with a UI element causes marimo to automatically re-run dependent cells or mark them as stale, keeping code and outputs consistent; deleting a cell removes its variables from program memory, eliminating hidden state. Notebooks execute in a deterministic order derived from variable references rather than cell position, and only cells that need to run are executed, based on static analysis of the code.

Notebooks are stored as pure Python .py files, making them versionable with Git, importable as modules, executable as scripts (parameterized by CLI arguments), testable with PyTest, and deployable as interactive web apps or slides via the marimo CLI or as WebAssembly-powered HTML. The product includes built-in interactive elements (sliders, dropdowns, dataframe tables and transformers, selectable plots, chat interfaces), interactive dataframes with search/filter/sort, dynamic markdown, and first-class SQL cells that query dataframes, databases, warehouses and lakehouses through backends including Polars, Pandas, PyArrow, DuckDB, SQLite, Postgres and MySQL, returning results as Python dataframes. Built-in package management supports major package managers, install-on-import, serialization of package requirements into notebook files with PEP 723 metadata, and isolated venv sandboxes managed with uv.

The editor offers GitHub Copilot-style autocomplete, AI assistants with context on in-memory variables, custom system prompts and bring-your-own API keys or local models, inline edits, context-aware chat, vim keybindings, Ruff code formatting, a variable explorer, debugging panels and hotkeys, plus a CLI, a Python library, a VS Code extension and a PyCharm plugin. The `marimo pair` command connects external AI agents such as Claude Code, Codex or OpenCode to running notebook sessions. The company also runs molab, a free cloud-based notebook service for creating and sharing notebook links, and has a cloud waitlist. marimo is a NumFOCUS affiliated project.

Founding story

Akshay Agrawal built marimo out of frustration with Jupyter notebooks after finishing a PhD at Stanford, where he grew an open-source optimization library to a million monthly downloads, and after working on TensorFlow at Google Brain. He cited hidden state producing hidden bugs (noting that over a third of the 10 million Jupyter notebooks on GitHub fail to reproduce), the JSON file format's poor fit for Python codebases, and a lack of interactivity in finished documents. With co-founder Myles Scolnick (previously Palantir and CloudKitchens) he set out to combine interactive computing with the rigor of traditional software development. The project was incubated for over a year with Stanford's SLAC Laboratory and engineers at tech startups, and cites Pluto.jl, ObservableHQ and Jupyter as inspirations.

Business model

marimo distributes its notebook as free, permissively-licensed open-source software, and the company has stated it will always remain free and open source. At the seed announcement the company said it would eventually offer a paid product that optionally integrates with marimo and other tools it develops; it also operates molab, a free cloud-based notebook service, and maintains a cloud waitlist.

No paid product was in market at the time of the seed announcement; the company stated a paid product optionally integrating with marimo and its other tools would be offered eventually, while the open-source tools remain free.

Traction

marimo was downloaded over one million times in the ten months after its January 2024 launch, and had over 7,800 GitHub stars, 60+ contributors and 1,000 Discord members as of November 2024; a December 2024 podcast description cited almost 9,000 GitHub stars. The company's site listed 103 example notebooks in its gallery. Reported users include Mozilla AI, BlackRock, Stanford, MLB, Bunkerhill Health, DNB Cyber Defense, Sumble and SLAC National Accelerator Laboratory.

Latest developments

On October 30-31, 2025, CoreWeave acquired marimo for an undisclosed sum, as part of CoreWeave's effort to move up the stack from AI hosting toward AI application building; PitchBook estimated marimo had raised about $5 million. Earlier product developments include the `marimo pair` command for connecting AI agents such as Claude Code, Codex and OpenCode to running notebook sessions, and the molab hosted notebook service.