marimo

marimo is an open-source reactive notebook built around Python, with SQL support and routes from analysis to deployable apps. It is best suited to data and software practitioners who want notebooks to behave more like reproducible Python projects than standalone documents. The central trade-off is clear: strong workflow flexibility comes with a Python-first focus, and hosted molab notebooks are public by default.

Overview

marimo’s defining choice is to store notebooks as Python files rather than hide their logic in a notebook-only format. That makes work easier to version with Git, run as scripts, or reuse as modules. It is a practical fit for teams already working in Python; readers whose work centers on another language should look elsewhere.

Key features

Reactive execution runs dependent cells when a cell changes, or marks them stale, keeping code and displayed results aligned. Combined with deterministic execution and package requirements stored in notebook files, this reduces the hidden state that can make notebook results hard to reproduce. It does not make arbitrary code safe: review notebooks before running them.

Notebooks can also become interactive web apps or run as Python scripts, and can execute in browsers through WebAssembly. Sliders, dropdowns, interactive dataframe tables, and selectable plots let people explore results without editing the underlying code. That range makes marimo useful beyond one-off analysis, especially when a notebook needs to serve as an interface or reusable project artifact.

SQL cells connect to dataframes, CSVs, Google Sheets, and databases. Supported backends include Polars, Pandas, PyArrow, DuckDB, SQLite, Postgres, and MySQL. AI notebook features support OpenAI, Anthropic, Gemini, Ollama, and OpenAI-compatible APIs. Developers can work through extensions for VS Code and PyCharm; the VS Code extension also works in Cursor and other VS Code forks.

Package management options include pip, uv, Poetry, Pixi, and Rye, while notebook sandboxes support uv and Pixi. Installation options also include conda. The breadth is useful for fitting existing Python projects, though sandbox package choices are narrower than the general project options.

Security controls address remote access and opening notebooks, but not the risk of executing code. marimo says code does not run without an explicit user action, edit-mode notebooks are statically loaded, and token authentication is enabled by default for remote edit access. Run mode can use token flags or custom middleware. Once run, notebook code can execute arbitrary code, so trusted origins or code review remain essential.

Pricing

marimo open-source notebook — free. The permissively licensed, open-source notebook is the straightforward choice for anyone who wants to run marimo without paying for hosted compute.

molab — free. The free hosted option provides 4 CPUs, 32 GB RAM, and sessions of up to 12 hours, with use expected to remain reasonable. It suits people who want cloud-hosted notebooks without setting up their own environment, but notebooks are public by default. For notebooks created after August 26, 2026, files larger than 1 GB are not saved when the notebook shuts down.

marimo cloud — custom pricing. The closed beta offers on-demand cloud resources for experimentation, collaboration, sharing, and deployment. It is aimed at users who need hosted resources beyond molab, but access is limited to the beta.

Platforms

marimo supports API, extension, Linux, macOS, self-hosted, web, and Windows environments. It supports Python and SQL, with notebook formats including Python (.py), Jupyter Notebook (.ipynb), HTML, PDF, and Markdown. GPU compute is supported; execution is hybrid. Version history is not supported.

Who it's for

The maker names AI engineers, analysts, educators, analytics engineers, data engineers, data scientists, ML engineers, and web developers as users. Its reactive execution, Git-friendly files, SQL integrations, and app deployment are particularly relevant to people who want analysis to remain reproducible and usable in a broader Python workflow. It is less suited to readers seeking a language-neutral notebook environment.

Pros and cons

• Reactive execution and deterministic behavior: dependent cells update or become stale, helping keep outputs aligned with code.
• Python-file storage: notebooks can be versioned in Git, run as scripts, and reused as modules.
• More than static analysis: interactive controls, SQL integrations, and app deployment support exploration and sharing.
• Public-by-default hosted notebooks: molab is a poor fit for private work unless its visibility model suits the notebook.
• Execution still carries code risk: safeguards around loading and access do not prevent arbitrary code from running after a user starts it.
• Free hosted use has bounds: reasonable-use terms, 12-hour sessions, and the post-shutdown file limit may constrain longer or larger projects.

Alternatives

Choose Apache Zeppelin if you want free, open-source notebook software with Linux, macOS, self-hosted, and web platforms. Deepnote is a hosted-oriented alternative with a free plan, a free trial, and a free tier of up to three editors and five projects. Paperspace Notebooks is worth considering if its free tier’s included storage and CPU and GPU machines better match your compute needs.

Pluto.jl is a free, MIT-licensed alternative for readers who prefer its supported platform range. SageMath is another free open-source option, with installation on unlimited computers and permission to download and redistribute copies. Datalore offers a free cloud tier with defined CPU hours, storage, and parallel notebook limits. Dnote is a free, open-source option with an optional self-hosted server. JupyterHub is a free open-source choice for teams that want to host notebooks on cloud infrastructure or their own hardware.

Browse more options in Notebook Software or Coding Playgrounds.

Verdict

Choose marimo if you want a Python notebook whose reactive execution, Git-friendly files, and deployment options support reproducible work beyond a single analysis session. Its free open-source version is compelling for local or self-hosted use; look elsewhere if you need a language-neutral workflow or private hosted notebooks by default.