marimo
@marimo
An open-source reactive Python notebook: reproducible, git-friendly, execute as scripts, share as apps! GitHub: Discord:
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.
Ask and you shall receive: marimo now runs natively in PyCharm. Open, edit, and run reactive notebooks in @JetBrains without leaving your IDE. Install from the JetBrains Marketplace 👇
Open any marimo notebook on GitHub. Click one button. The raw Python turns into a live, interactive notebook you can run, right there in your browser. That's marimo Glance. Free extension for Chrome and Firefox, works on GitHub, gists, and GitLab. Details below.
Put a slider in one paragraph and the chart three paragraphs later. The Python cells stay connected. Reactive marimo cells now run inside Quarto, Jupyter Book, and anywhere MDX lives. quarto-marimo, jupyter-book-marimo, mdx-marimo, all open source. More below ⬇️
SciPy was a blast! Missed Trevor Manz's anywidget talk or Dylan Madisetti's caching poster? Not to fear, for our first ever launch week is here...5 new announcements next week. What are you hoping it'll be?
4. pick a niche that interests you, even if it's not on the pre-selected list of papers
What actually makes a notebook win? We just extended our competition with @askalphaxiv to July 8, so here's the breakdown.
FastAPI Cloud hit public beta and since marimo notebooks are just Python files under the hood, you can deploy one there too. Vincent Warmerdam live-codes it: create_app to fastapi deploy to "hello from marimo" live on the web. A fun party trick in 4 minutes ⬇️
double descent is one of the most counterintuitive results in deep learning, and you can't really show it on plain MNIST. MNIST-1D can. huge thanks to @samgreydanus & @hippopedoid for the paper on @askalphaxiv. new video reproduces it from scratch 👇
running 2,000 training runs and reporting the best one isn't better science. it's loading the dice. the curse of optimality from @askalphaxiv explained below:
Day 5 of our "What can you do with a GPU on molab?" series. Does adding reasoning to open source LLMs actually improve accuracy? Video in the comments ⬇️
Day 3 of our "What can you do with a GPU on molab?" series: how long does it take to solve the Traveling Salesman Problem with 2-opt on a GPU? Hint: it's faster than you can say "500 cities in half a millisecond." Video below ⬇️
Day 2 of our "What can you do with a GPU on molab?" series: python widgets 🤝 stable diffusion. Draw, hit play, and the model reimagines it in real time, no matter how many prompt changes, region erasures, or random scatterings of pixels you throw at it. Video below ⬇️
Sharing is also improved across the board: mirror notebooks from GitHub, share as slides or apps, embed in webpages, and fork from our gallery. Notebooks on molab are public but not discoverable by default, like GitHub Gists.
GPUs unlock new use cases. For example, get started with fine-tuning models with @UnslothAI's official molab notebooks: github.com/unslothai/n...
"'The lion eats a man' vs 'the man eats a lion.' Same bag of words, opposite meaning, and modern embeddings can't tell them apart. Sorry to both the lion and the man. Vincent Warmerdam tests open source and frontier embedding models on word order, grammar, and negation ⬇️
If you can vibe code anything these days ... do we still need notebooks? We're going live in 24 hours to find out with Hamel Husain to discuss. 🗓️ May 22, 2026 at 1 PM PST. Link below ⬇️
We're big Polars fans at marimo. The Polars team has been working on something new: it's early, it's fancy, and it ships with a few features that are directly relevant to marimo. Livestream with Thijs Nieuwdorp on May 22, 2026. Link below ⬇️
We invited Isaac Flath to stress-test marimo-pair: can a reactive notebook actually give agents the same live kernel access that makes RLMs work? Agentic coding, live bugs, and how much time people spend playing World of Warcraft. More in the full video ⬇️
Come find us at PyCon! We have a booth all weekend and would love to meet you ☺️ P.S. our very own Myles Scolnick is hosting a talk later today. Join us to learn "How to port a Python kernel to Pyodide for a blazingly fast in-browser coding experience." See you there!
More than 1,000 registrations for Parul Pandey's live O'Reilly workshop on marimo for AI and ML Development. 🗓️ Monday, May 18, 8 am–11 am PT (8:30–11:30 pm IST). A free 3 hour hands-on, interactive workshop. Register below. ⬇️