Deepnote makes the notebook collaborative.

Here is a small, true, faintly absurd fact about how a lot of important decisions get made: someone runs an analysis in a Jupyter notebook, produces a chart, takes a screenshot of the chart, and pastes the screenshot into a message. The chart is now a picture. Nobody can click it, rerun it, or check the math. This is roughly how it worked in 2010 and roughly how it worked in 2019, which is the year four engineers decided it was ridiculous enough to build a company around.

Those engineers - Jakub Jurovych, Jan Matas, Filip Stollar and Jirka Lhotka - had spent time inside places that take data seriously: Two Sigma, Palantir, Google. The pattern they kept hitting was not that the tools were weak. Jupyter is powerful and ubiquitous. The problem was that the tools were lonely. Everything assumed one person, one machine, one file. "We tried everything out there," Jurovych has said, "the collaboration was always broken, no matter what we tried."

The move: don't replace the notebook, finish it

The obvious temptation, when you dislike a category-defining tool, is to build a shinier replacement and hope the world switches. Deepnote made the less glamorous bet. It kept the notebook - kept Jupyter compatibility, kept Python, added SQL and R as first-class citizens - and wired in the thing the notebook never had: other people. Two analysts can now sit in the same document at the same time, watch each other's cursors, and argue about a query without emailing a screenshot of it.

That sounds like a feature. It is closer to a worldview. If a notebook is a private draft, the analyst is a soloist and the organization sees only the final recital. If a notebook is a shared space, the whole company can wander into the data workflow - which is, more or less, the company's stated mission.

The money believed the workflow

Deepnote raised a $3.8M seed in 2020 and a $20M Series A in January 2022, both co-led by Index Ventures and Accel, bringing the total to about $23.8M. The more telling detail is the angel list: OpenAI's Greg Brockman, Figma's Dylan Field, Naval Ravikant, Elad Gil, Daniel Gross, Lachy Groom. That is a roster of people who spend their days looking at software workflows for a living, and they were writing checks for a notebook. They were not buying a demo; they were buying the insight that collaboration, not compute, was the missing piece.

Then the AI era arrived on schedule

A notebook is, conveniently, one of the best-shaped surfaces for AI: structured cells, code next to prose, an obvious place to put a suggestion. Deepnote leaned in. Its AI agent will now draft an entire notebook from a prompt - code, SQL, and the explanatory text between them - and edit existing notebooks with context about what's already there. Autocomplete runs on Mistral's Codestral. There is an "ask mode" that answers questions without touching your work, for the times you want a rubber duck rather than a co-author.

The interesting restraint is that the agent drafts; the human still steers. In a field crowded with tools that promise to hand you The Answer, Deepnote's pitch is closer to handing you a first draft you can actually read and correct - which, if you have ever received an unexplained number from a black box, you will recognize as the more useful gift.

An open-source turn

In 2025 Deepnote did something a venture-backed SaaS company does not have to do: it went open source, releasing its notebook as a drop-in Jupyter replacement. It also shipped Modules - reusable packages of trusted SQL and cleaning routines, so a team stops re-solving the same problem every Monday - and unlimited charting that renders millions of rows from a Pandas DataFrame without asking you to downsample first. None of these are flashy. All of them are the sort of thing that quietly earns a data team's loyalty, one unglamorous afternoon at a time.

The competition is real - Hex most of all, plus Databricks notebooks, Mode, Observable, Colab, and Jupyter itself. Deepnote is small, roughly 26 people, going up against much larger operations. But the category it is fighting over is enormous and strangely under-served: the daily working surface of every analyst on earth. The notebook is thirty years old and still runs the world's data work. Deepnote's whole thesis is that finishing it beats replacing it.