marimo

How one developer took a coding model from fine-tuning to live demo without leaving a notebook. Kishan N, a final-year CS student, wanted to build Chitti, a privacy-first coding assistant that runs entirely on local infrastructure. The problem in his original setup was that sessions timed out mid-training, hidden state bugs ate hours of debugging, and T4 GPUs didn't have the VRAM for serious fine-tuning. He shifted the entire pipeline to marimo on molab. An RTX PRO 6000 with 102GB of VRAM handled the training, and reactive execution meant every cell reflected the actual state of his model. No stale variables, no out-of-order mysteries. The result: dataset curation, fine-tuning Qwen2.5-Coder-7B, benchmarking on HumanEval+, and a live Gradio demo with a public URL, all built and deployed from marimo notebooks. Chitti now outperforms both baseline models across all three benchmarks he tested. Full writeup in the comments ⬇️ #python #machinelearning #mlops #coding #AI

How the #1 virtual power plant aggregator in North America builds the tools that keep the grid stable. Voltus manages 8.5 GW of distributed energy resources: industrial HVAC, battery storage, backup generators, all coordinated to respond when the grid is under stress. Pulling that off takes market experts and engineers working on the same problems with the same tools. Notebooks made that collaboration possible, but keeping them auditable at scale meant heavy manual process: version control by filename, hidden state that could silently break a dashboard, and interactive apps only engineers could build. Rolling out a dedicated BI platform would have taken years and created exactly the silo they were trying to avoid. After trialing marimo in early 2025, Staff Software Engineer Will Lachance found the middle ground: a reactive engine that replaced manual verification with architectural guarantees, in pure .py files that plugged straight into the CI/CD and code review pipelines the team already used. Suddenly, analysts could ship interactive apps without touching JavaScript. marimo apps at Voltus grew from 2 to over 35 in six months. The same notebook moves from prototype to human-verified app to scheduled automation, without rewrites. They even embedded a reactive marimo sandbox in their public API docs so partners can test dispatch integrations themselves, eliminating the usual back-and-forth of manual integration testing. In Will's words: marimo didn't force a tradeoff. Voltus got the reproducibility and auditability a regulated, high-stakes industry demands, while their workflows became more accessible and more enjoyable to use. Full writeup in the comments ⬇️ #python #energy #openSource #dataengineering #reproducibility

We had our first ever launch week here at marimo. Five announcements in five days, and if you blinked you probably missed a couple, so here's everything in one place. Monday: let's pair! - marimo-pair, the skill that gives agents direct access to marimo notebooks, has a new landing page, and is now available in the Claude Code and Codex CLI marketplaces. We also released a VS Code/Cursor extension. Tuesday: marimo anywhere + marimo Glance - quarto-marimo, jupyter-book-marimo, and mdx-marimo let you weave live, reactive Python into Quarto sites, Jupyter Books, and anything built with MDX. - marimo Glance, a browser extension, renders raw Python files as a live, interactive WASM notebook. Wednesday: marimo in PyCharm - the marimo plugin brings reactive notebooks to PyCharm and any other JetBrains IDE. Thursday: widgets galore - 8 new widgets, to be specific. Our favorite is WidgetDAG, which turns your entire notebook into a live, reactive graph. Friday: notebooks as slides - marimo notebooks can now be live, interactive slides. Interact with widgets, edit and re-run code mid-talk, and author the whole deck without leaving presentation mode. If our launch week had a theme, it's that your workflow shouldn't be restricted to standalone tools. It should follow you into your IDE, your published site, your talks, and your agent's toolkit. Your notebook should work where you are. Links to everything in the comments ⬇️ #builtwithmarimo #launch #aitools #research #developer

Sliding into your next presentation like 🏄🏻♀️ To wrap our first ever launch week, we're announcing that you can now present marimo notebooks as slides. The slides are the notebook. Widgets stay interactive and you can edit slides live while presenting, with no second window or separate demo setup. You can also author the whole deck without leaving presentation mode: a minimap to reorder slides, speaker notes for your second monitor, fragments for step-by-step reveals. Brand it with CSS, share straight from molab, or export to PDF. Blog and video in the comments ⬇️ #launch #python #opensource #developer #builtwithmarimo

marimo is reactive, which means that it knows in what order variables are declared. So what if you had widgets at your disposal that could leverage that? For launch week, we're releasing some tools that do exactly this.

A JetBrains team member and a marimo team member meet at PyCon. You can probably guess where this is going. Today we're releasing the marimo plugin for PyCharm! Open, edit, and run marimo notebooks directly in your IDE, with reactive cells and interactive UI elements in a dedicated notebook editor. And because marimo notebooks are stored as plain Python files, they stay Git-friendly, run as scripts, and slot into your existing projects with no extra setup. The reactive notebook you know, in the editor you already open every day. Install it from the JetBrains Marketplace and open your first notebook. Learn more below ⬇️ #python #openSource #pycharm #developer #ide

What should a notebook be when an agent is the one using it?

We just wrapped our second notebook competition with alphaXiv! 🎉 Huge thank you to everyone who took part, whether you submitted a notebook, came to the event, or helped spread the word. It was so fun to see all the creative ideas come through. We kicked things off with team demos. If you were on the call, you got a sneak peek at something we're launching this week 👀 alphaXiv also demo'd their tool that turns an arXiv paper into a marimo notebook in one shot. We closed with presentations from the community. Seeing the energy from the call was fantastic, and we're grateful to everyone who attended! Without further ado, the winners of our second notebook competition: 1st place 🥇: Eric Ma - https://lnkd.in/gae6t68f 2nd place 🥈: Konstantin Taletskiy - https://lnkd.in/gg5agiPD 3rd place 🥉: Antonin Peronnet & Thomas Winniger - https://lnkd.in/gE3GV2yJ #python #competition #openSource #machinelearning #builtWithMarimo