jupyter-live-kernel

Use a live Jupyter kernel as a stateful REPL: each execute call runs against the same kernel, so imports, DataFrames, and objects persist. Create a scratch notebook session via the Jupyter REST API, then send code with the execute command.

Use a live kernel for iterative exploration, data science, and ML work where you build state incrementally. Use one-shot execution for stateless scripts that need tool access, and the terminal for shell commands, installs, and process management.

Run the variables command with the notebook path to list all live variables, or preview a specific variable by name. Place global flags like --path before the sub-subcommand, for example: variables --path scratch.ipynb list.

The kernel needs time to initialize after the server starts or after a restart, so the first execution may exceed the default 30-second timeout. Retry once before escalating, and pass an explicit --timeout of 60 or 120 seconds for heavy operations.

Packages must be installed into the JupyterLab tool environment, since the kernel uses JupyterLab's Python interpreter. Installing into the system global environment will not make packages importable inside kernel executions.

Use the restart-run-all command with --save-outputs to restart the kernel and execute every cell in order. If it fails, parse the returned JSON ename and evalue fields to identify the exact exception.