This is a known issue/current limitation with at least two open issue pages currently at the JupyterLite repo on GitHub: here and here. I would especially see here and here for more insight into why this is a limitation at the current time.
For more context, read on. And for some ideas for workarounds, see the bottom section.
I know that is a JupyterLab interface you show in your screenshot, and so you are saying “JupyterLab” in the category & writing in the post: “in the jupyter lab platform.” However, that is a very special and experimental offering you are showing. It is limited in ways that JupyterLab linked to a typical, full Python kernel is not.
See warnings here highlighted by shocking bolts around ‘Status’ and here indicated by ‘Experimental’ in bold between two warning danger symbols about the experimental nature of JupyterLite.
You have run into one of the circumstances that merits these warnings being in place.
For more context, the JupyterLite offerings of JupyterLab and Jupyter Notebook run using kernels powered by web assembly deep in your own browser. There is no remote machine. As it is running inside your browser it is subject to constraints imposed by security concerns for your browser and limitations with how it can interact with your system related to that and how is it is implemented. While some of these limitations may be overcome, or changes in how you workaround them arise, it will in the near term be limited in comparison to typical JupyterLab installed in your local system or an an actual remote machine, running in conjunction with a typical, full Python kernel.
ALTERNATIVE OFFERING: You can work with typical JupyterLab in conjunction with a typical Python kernel without installing anything on your computer
In the meantime, there’s a workaround. Via a few offerings from various sources, you can work with typical JupyterLab in conjunction with a typical Python kernel without installing anything on your computer. I’ll detail a couple:
If your needs are limited, you can check out what I say about ‘MyBinder-served temporary Jupyter sessions’ in the top section here. I don’t specify many repos for launching there but the following are two that have more current features than many I could refer you too:
- my gist here launches JupyterLab 4.3 with Python 3.11
- my repo here defaults to older Python and JupyterLab for now but has a lot of data science packages already installed in sessions that come up
There’s also a good list of current offerings here, although most don’t come with many packages pre-installed. If you let me know better your package needs I can point you to a repo that will launch a session more in line with what you need. Because there is no login needed, it is the easiest to get going with but is more limited in that it is temporary, lacks persistence, and the computational power, while surprisingly adequate, is not overwhelming.
You can also use Anaconda Cloud that is a very thorough offering, yet needs you to be familiar with, or willing to learn about, the Anaconda/conda way of environment and package management. The advantage is that it is more persistent than the temporary sessions offered via MyBinder. (I still would caution you to save anything useful back to your local storage.)