jupyter-notebook

Use the jupyter_live_kernel.py script with the execute subcommand, passing --path to a notebook and --code with your Python code. State persists across calls, so variables and imports survive between executions.

Use a live Jupyter kernel when you need state across steps, such as iterative data exploration, ML prototyping, or inspecting DataFrames. Use one-shot execution for stateless scripts that complete in a single run.

Run the variables subcommand with list to see all live variables, or preview with --name to inspect a specific variable's value. Add --compact to reduce verbose JSON output.

The kernel needs time to initialize after the server starts, so the first execution may exceed the 30-second default timeout. Retry the call or pass --timeout 120 for long-running operations.

Force zmq transport by adding --transport zmq to the execute command. This uses jupyter_client directly and sidesteps broken websocket reply channels where the kernel actually ran fine.

POST requests to /api/sessions fail with '_xsrf argument missing' unless the server was started with --ServerApp.disable_check_xsrf=True. Restart JupyterLab with that flag for REST-only access.