Jovyan 1.11: Python Interactive Windows and Smart Interruption
We are releasing Jovyan 1.11 today. This update isn’t about adding more generative power; it’s about fixing the friction between how AI agents typically work and how data scientists actually work.
Most coding agents are built with a software engineering mindset: read a spec, plan the architecture, write the code. That works for building apps, but data science is different. Data science isn’t just plan → implement. It is Explore & Understand Data → Plan → Implement → Execute and reiterate.
You cannot plan a transformation pipeline until you know what the data looks like. And more often than not, you will execute the code to train model or extract insights. In this release, we updated Jovyan to better support that workflow.
Interactive Windows Integration
Exploration is not optional
In traditional software engineering, you can often write code “blind” if you know the requirements. In data science, working in memory is the norm. You need to inspect the dataframe head, check the shape, and visualize distributions before you write the next line of code.
Until now, using an AI agent often meant sacrificing that interactivity. You would ask a question, the agent would write a script, and you would have to trust it, or copy-paste it to verify.
Bringing the REPL to the agent
Jovyan already provides a great notebook experience, but sometimes you don’t want to spin up a full notebook context just to test a hypothesis or debug a specific slice of data.
With Interactive Windows Integration, Jovyan sends code directly to a Python interactive session in VS Code, executes the code, and returns the results. This enables faster iteration without worrying about keeping a notebook clean.
Our core philosophy is that Jovyan should not operate as a black box. All actions and results are visible to the user, and user-approved.
Unlocking new workflows with Jovyan
With Interactive Windows Integration, you can now choose different workflows for exploration and implementation:
Do your main exploration in a notebook. If a side question comes up, prompt Jovyan to explore it in an Interactive Window, without making your notebook heavy.
Do your exploration in Interactive Windows, then ask Jovyan to format the work into a clean notebook or a script.
Try it out and let us know your preferred workflow.
Interrupt with Feedback
Catching unoptimized loops
Data science tasks differ from general coding tasks in another major way: execution time. Running unit tests takes milliseconds. Training a model or processing a large parquet file can take hours.
We have all been there: you launch a process, and 10 seconds in, you spot a mistake.
Maybe the agent is trying to use .apply() with a complex function on a massive dataframe, instead of vectorizing.
Maybe it started a GridSearch with verbose=0, so you can’t tell whether it’s working or stuck.
Maybe it is training on the full dataset when it should have sanity-checked on a small sample first.
Fix it in flight
Previously, you had to interrupt the run, explain the issue, and restart from scratch.
With Jovyan’s Smart Interuption, you can stop the agent the moment you spot an inefficiency. Hit interrupt and say: “You’re running this on CPU, switch to CUDA,” or “Add a progress bar so we can see the ETA.” Jovyan captures the feedback and fixes the issues before retrying the run.
If user don’t say anything and interrupt, Jovyan will analyze the logs and resources usage and propose some improvement proactively !
This is a first step in this direction. Long training runs are central to data science and machine learning workflows. We will keep making Jovyan smarter and more efficient, both in pre-run checks and in monitoring during execution.
What’s next?
Jovyan 1.11 is available now. These features are designed to make the agent a more pragmatic partner for day-to-day data science work: exploring data interactively, and managing long-running executions more efficiently.
Try it out: open Jovyan, point it to a dataset, and start accelerating your data science workflow !