Using a Jupyter Notebook with Pixi and VS Code

[Omnistic]

Problem description

I am aware of #411, and I am not a software developer. But I can't get a Jupyter Notebook to work in my VS Code with the Pixi Extension. My workflow is as follows (in VS Code):

• Initialize Pixi
• Add ipykernel and pip (as suggested there: https://stackoverflow.com/questions/79160156/how-to-use-a-pixi-virtual-environment-to-run-a-jupyter-notebook-in-vscode)
• Add Python (3.12) and my modules (such as numpy)
• Create a new Jupyter Notebook

The kernel seems to be right in the Notebook, it is: .pixi\envs\default\python.exe, and if I just have print('test') in my notebook cell it works. But as soon as I import anything the kernel crashes. In the VS Code OUTPUT, I get a warning that seems relevant:
17:48:39.059 [warn] Failed to get activated env vars for ~\Documents\GitHub\my_folder\.pixi\envs\default\python.exe in 30028ms

I am not sure what to report regarding the crash (or even how to report it), and I am using Windows in case it matters.

Has anybody else had this problem? Is here the right place to ask? Please consider that I am not a software developer and I don't understand a lot of the jargon that is used here.

Thanks for any help you can provide.

Omni

[bollwyvl]

For a single environment, several approaches to starting code (or any other
$PATH-aware tool) will generally set up all the environment variables properly, e.g.

Given code is not running (if it is, none of these will have any effect):

• activate the shell

  pixi shell
  code .

• ...or run code directly (returns control immediately):

  pixi run code .

• ... or make task

  # pixi.toml
  [tasks]
  start-code = "code ."

  ...then

  pixi r start-code

The latter is handy if there is a package-under-development involved, as it could do any pre-flight steps.

For multiple envs, things get a bit more challenging. I've at least once done some out-of-band scripting to:

• plan to launch the tool from a specific environment (e.g. dev)
• find all
  • .pixi/envs/{env}/share/jupyter/kernels/{name}/kernel.json
• munge all of the kernels into
  • .pixi/envs/default/share/jupyter/kernels/pixi-{env}-{name}/kernel.json
  • update all of the argv:
    • from

      {
        "argv": [
          "python",
          "-m",
          "ipykernel_launcher",
          "-f",
          "{connection_file}"
        ],
        "display_name": "Python 3 (ipykernel)",
        "language": "python",
        "metadata": {
          "debugger": true
        }
      }

    • to

      {
        "argv": [
          "pixi",
          "run",
          "--environment={env}",
          "python",
          "-m",
          "ipykernel_launcher",
          "-f",
          "{connection_file}"
        ],
        "display_name": "({env}) Python 3 (ipykernel)",
        "language": "python",
        "metadata": {
          "debugger": true
        }
      }

• launch the tool

It's possible this could all be done with some clever xargs | jq invocation,
but i generally fall back to python.

This approach has the (usually beneficial) side-effect of restart-and-run-all potentially
re-locking/rebuilding an environment. However, this can get very annoying if there is any
"deep magic" at play, such as a pixi-managed python package which also sees the
file-being-edited as something that should invalidate the lock file.

[Omnistic]

@bollwyvl thank you for your answer. It is much appreciated.

However, I don't really understand it.

I have a single environment, and still I tried the various commands that you sent in VS Code terminal to no avail.

I am attaching a screenshot of what I am trying to do, in case it might be helpful.

And here is the log from the screenshot (for copy/paste if necessary):

15:05:46.415 [warn] Failed to get activated env vars for ~\Documents\GitHub\trying_pixi_vscode_notebook\.pixi\envs\default\python.exe in 30075ms
15:05:46.420 [info] Process Execution: ~\Documents\GitHub\trying_pixi_vscode_notebook\.pixi\envs\default\python.exe -c "import site;print("USER_BASE_VALUE");print(site.USER_SITE);print("USER_BASE_VALUE");"
15:05:46.420 [warn] Failed to get activated env vars for ~\Documents\GitHub\trying_pixi_vscode_notebook\.pixi\envs\default\python.exe in 30071ms
15:05:46.445 [info] Process Execution: ~\Documents\GitHub\trying_pixi_vscode_notebook\.pixi\envs\default\python.exe -m pip list
15:05:46.447 [info] Process Execution: ~\Documents\GitHub\trying_pixi_vscode_notebook\.pixi\envs\default\python.exe -c "import ipykernel; print(ipykernel.__version__); print("5dc3a68c-e34e-4080-9c3e-2a532b2ccb4d"); print(ipykernel.__file__)"
15:05:46.450 [info] Process Execution: ~\Documents\GitHub\trying_pixi_vscode_notebook\.pixi\envs\default\python.exe ~\.vscode\extensions\ms-toolsai.jupyter-2025.9.1-win32-x64\pythonFiles\vscode_datascience_helpers\kernel_interrupt_daemon.py --ppid 32624
    > cwd: ~\.vscode\extensions\ms-toolsai.jupyter-2025.9.1-win32-x64\pythonFiles\vscode_datascience_helpers
15:05:46.505 [info] Process Execution: ~\Documents\GitHub\trying_pixi_vscode_notebook\.pixi\envs\default\python.exe -m ipykernel_launcher --f=~\AppData\Roaming\jupyter\runtime\kernel-v3c1244e16bf3a65418c9825c6afb2855b4996da10.json
    > cwd: ~\Documents\GitHub\trying_pixi_vscode_notebook
15:05:47.161 [info] Kernel successfully started
15:05:47.229 [error] Disposing session as kernel process died ExitCode: 3228369023, Reason: 
15:06:17.185 [warn] Failed to get activated env vars for ~\Documents\GitHub\trying_pixi_vscode_notebook\.pixi\envs\default\python.exe in 30022ms
15:06:17.189 [info] Process Execution: ~\Documents\GitHub\trying_pixi_vscode_notebook\.pixi\envs\default\python.exe ~\.vscode\extensions\ms-toolsai.jupyter-2025.9.1-win32-x64\pythonFiles\printJupyterDataDir.py

I would like to have my notebook in VS Code, because that's an environment I am familiar with. And ever since I tried it, I fell in love with Pixi, so I'd also like to have that.

Would you mind explaining to me in Layman's term how I am supposed to set this up?

So far, I went back to a Conda environment and a notebook in VS Code, and that works. But it would be better with Pixi of course.

Thank you for your help and take care,

Omni

[bollwyvl]

The simplest windows path:

• close vscode
• start a boring, non-magic, broken-for-the-same-way-for-30-years command prompt
  • from the run menu, start cmd.exe
• navigate to the working directory where pixi.toml is located
  • cd c:/Documents/GitHub/trying_pixi_vscode_notebook
• start code from pixi
  • pixi run code .
  • if this fails to find pixi.exe, it might need to use the concrete location, e.g.
    • c:\Program Files\pixi\pixi.exe run code .

It may also be that vscode_datascience_helpers relies on sniffing magic values, or is otherwise not good. One could try disabling that, and anything else mentioned in the logs, at least initially, and then re-add them to bisect what is causing the error.

[Omnistic]

It worked right away. I had no idea I could start VS Code from PIXI. I guess this was a stupid question, but thank you so much for taking the time to answer. I am very grateful.