We're having our first ever in person meetup on October 7th from 6-7 PM in San Francisco as part of SF Tech Week. ​Join the marimo team and community for a casual chat about AI-assisted coding and agents! RSVP: luma.com/pp27h5fo

We've all seen dataframe spaghetti code out there. But the remedy ... is pretty easy: just start using the pipe method! The moment you see the `.pipe()` in action, you'll never want to write pipelines without it. See the reason why here: www.youtube.com/watch?v=J0P...

Release 0.16.0 delivers multi-cell AI code generation and common error detection/reporting before runtime. The "Generate with AI" feature now can create multiple cells at once and run marimo check to identify issues such as multiple definitions, circular dependencies, and syntax errors.

There's *a lot* of feature's that we're proud of. But some are less well known than others, so we made a video! youtu.be/OM7utZeQ1GY

Release 0.15.3 provides real-time feedback when installing packages. You can now see the live installation logs without the terminal. We've also included the ability to reference files, cell outputs, and media to AI for better assistance.

We just upgraded molab. You can now attach a file to a prompt to generate a full notebook from scratch that analyses the file you just uploaded!

Pair with Claude Code to write marimo notebooks in only three lines. Claude can easily understand your code in marimo via our Python file format and dataflow graph, propelling the shift from prompt engineering to context engineering.

Learning is better when it’s interactive. The CodeCut newsletter by Khuyen Tran delivers Python and AI tips, 3 times a week, and now each issue includes a "Run Code" button. With one click, you can launch a molab notebook and experiment with the examples directly in your browser.

The Python Discord community just finished their annual code jam: create a browser-based application that is written almost entirely in Python that incorporates the theme of "wrong tool for the job." One of our favorite submissions is Tetris Bugs. You fit falling code blocks into working programs.

Sasha Rush from Cursor created these Tensor puzzles to help people learn PyTorch. They're fun and interactive with immediate feedback from cute dogs!

What if a matrix is more like a user interface element and less like a static page in a textbook? What if that matrix can immediately update the whole notebook when you interact with it? Well ... you get an amazing experience!

Release 0.15.1 focuses on performance improvements and context engineering. You can now easily debug by asking AI how to fix issues such as missing tables, import errors, or other exceptions. We've also improved configuration for displayed models, making it easier to access your favorite models.

Want to deploy your own MCP server for your LLM? marimo is here to help! youtu.be/dJRJ9ytP9Vs

For many engineers we talk to this is *the* sales pitch.

You can deploy a Python notebook to GitHub Pages. And the experience is great: just host a notebook as a docs page. Not only can your users see the docs, they can interact with your code too! YT tutorial can be found here: www.youtube.com/watch?v=m8b...

We have better LLM configuration now!

We've been trying Claude Code for data analysis and ... well ... ... let's just say that the recording of it has plenty of lessons we've learned. youtu.be/fDiylmtYlWM

If you know you know

Release 0.15.0 includes many improvements to the marimo editor's AI integration such as dramatically simpler configuration (switch between local and hosted models with a single click) and custom model setup for chat, code editing, and autocompletion.

molab Run Mode When sharing, simply add /app to the end of your notebook URL so anyone can interact with the notebook live, even users not in your workspace. You also don't need to worry about overusing compute due to persistent caching and storage.

"I could buy the new iPhone 17 Pro Max with the compute cost from running this code." No worries, we have persistent caching/storage in molab, our cloud-hosted Python notebook workspace, so you can buy all the phones you want!

Many of you asked for VS Code integration so here's a sneak peek! We previewed this along with other exciting features at our community call this past weekend, check out the full call here: youtu.be/3caIs0jWykc

It turns out that the titanic dataset has an interesting twist! You can model the whole thing with just a single column and get on par performance with the rest of the features! It was a fun exercise with Claude that we recorded here: youtu.be/_Y1-CC1jyvA

We're having our community call this Saturday August 16 from 10:30-11:30 AM PST! ​Join us for live demos from community members, AI developer experience discussion, and more. RSVP / Add to your calendar: lu.ma/ifs1nwzk

Our latest release introduces several conveniences for exploring and visualizing in-memory objects, along with bug fixes. We also improved ibis integration (portable dataframe library) with native marimo support, all thanks to Kyrre Wahl Kongsgård!

We hear the talks at @pydataamsterdam are pretty good this year.

Check out our talk at SciPy 2025! www.youtube.com/watch?v=o4J...

We're making widgets for image-related tasks! First up is annotation, so stuff like this:

We're currently improving our AI coding experience and would greatly appreciate feedback from your experience using agents such as Claude Code, ChatGPT, and Cursor: form.jotform.com/25217726795...

Our latest autocompletion feature doesn't just work on dictionaries, it does dataframes too!

Cerebras gives you *really* fast LLMs. We gave their stack a spin and learned that it actually takes more time to type the prompt than it takes to get the response from the LLM. Exciting times! Full demo in YT video: youtu.be/SVQGTG9CPeI

Can your autocompletions do this?

🌟 Spotlight on @timeplus.bsky.social! Watch GitHub push events flow in a live dashboard. Built along with timeplus, it taps a public GitHub Kafka feed into a live table, rolling counts and top-20 repos — by streaming SQL and reactive UI. Try it live → marimo.demo.timeplus.com/

marimo is the canonical example of a "highly-starred recent GitHub repository" — from "Investigating Training Data Detection in AI Coders" , by Yang Liu et al.

Awesome to see our notebooks listed as job qualifications!

Want to try out all the coding agents? Give Qwen3/KimiK2 a spin? There's a lot of ways to configure this but the simplest way is to just get an OpenRouter account. marimo can integrate directly with it and the experience is great! Full YT tutorial here: www.youtube.com/watch?v=L1X...

🌟 Spotlight on Giovanni Giacometti! From @Strava to stats — right in your browser. Authenticate (or mock) to fetch your runs, then dive into Polars‑powered summaries, weekly heatmaps and dynamic charts — 100% client‑side for full privacy. Try it live → giovannigiacometti.it/strava-mari...

Reactive execution sets marimo apart from traditional notebooks. It keeps code and outputs in sync and prevents bugs before they happen. Our latest release includes an interactive minimap that helps you understand how data flows through your notebook.

The ollama models that you can run on a laptop (7b-14b param range) are getting good! But if you want to benefit most and actually use them in a notebook, we do have some advice as well as some techniques to make the most out of these smaller LLms. youtu.be/NIBprn5cEZA

Announcing molab: a cloud-hosted marimo notebook workspace with link-based sharing. Experiment on AI, ML and data using the world’s best Python (and SQL!) notebook. Launching with examples from @huggingface, @weights_biases, and using @PyTorch marimo.io/blog/announ...

We're going live in a few hours! Very curious to see if people can guess what we've been cooking ... youtube.com/live/cYtWzI...

🌟 Spotlight on Sean McLeod! Flight dynamics analyses — from roll rates to ISA This site hosts notebooks on roll rates, climb, trim envelopes, ISA, and more — illustrated with simulation data and plots. Explore → seanmcleod70.github.io/FlightDynam...

Want to run your ML models in SQL? Posit just released a new pkg, orbital, that opens the path towards doing this with scikit-learn. We chouldn't help ourselves and recorded a preview for it. As expected, it works very nicely with our SQL tooling. youtu.be/vcR9aJIz4A8

We're doing a livestream this Friday and ... we're going to announce something! Feel free to join live if you're curious! youtube.com/live/cYtWzI...

We heard that some of you like to claude code ... ... so we think you'll like this feature, --watch this! youtu.be/tOyZM7HswSk

🌟 Spotlight on koji h! Turn RSS URLs into an interactive feed reader! This notebook fetches and parses RSS XML into @pydantic data models, shows raw XML and a table of entries and offers date‑grouped sidebar navigation with search. Try it → kj-9.github.io/marimo-rss-...

We're doing a livestream later today! This time we're explore pydantic's AI library with Marcelo Trylesinski. You can join live below, or rewatch later on the same link: www.youtube.com/watch?v=ujQ...

We're doing

🌟 Spotlight on @harry.vangberg.name‬! SNP to primers pairs — no more BLAST backlogs. This notebook fetches the SNP’s flanking sequence from Ensembl, designs primers via Primer3 and runs local BLAST — all in under ~5 s per tweak. Try it → harry.vangberg.name/posts/autom...

We made a new widget! If you want to do some quick eval work while working on a pipeline, you can totally do that now. The best part? You can use a gamepad or your own voice to do most of the work! Full demo here: youtu.be/fYlsew5PGag