A jupyter raw kernel for nu.
nu-jupyter-kernel
is a Jupyter kernel specifically for
executing Nu pipelines.
Unlike most Jupyter kernels that rely on Python, this raw kernel directly
implements the Jupyter messaging protocol, allowing direct communication without
Python intermediaries.
It's designed to work seamlessly with Nu, the language utilized by
Nushell โ
a modern shell that emphasizes structured data.
The nu-jupyter-kernel
has several features making it a useful kernel for
Jupyter notebooks:
-
Execution of Nu code: Directly run Nu pipeplines within your Jupyter notebook.
-
State sharing across cells: Unlike isolated script execution, the kernel maintains state across different cells using the
nu-engine
. -
Rich Data Rendering: Outputs are dynamically rendered in various data types wherever applicable.
-
Inline Value Printing: Easily print values at any point during cell execution.
-
Controlled External Commands: By default, external commands are disabled for reproducibility. You can enable them as needed, and they will function as they do in Nushell.
-
Kernel Information: Access kernel-specific information via the
$nuju
constant. -
Error representation: Shell errors are beautifully rendered.
-
Nushell Plugin Compatibility: Supports Nushell plugins within notebooks, allowing them to be loaded and utilized as in a typical Nushell environment.
-
Plotting Integration: The kernel directly integrates the
nu_plugin_plotters
, making plots easily accessible.
In the "examples" directory are some notebooks that show how the kernel works. Opening the examples on Github also shows a preview of them.
The design of the nu-jupyter-kernel
focuses on the following goals:
-
Reproducibility: Notebooks should be as reproducible as possible by default.
-
Clarity in dependencies: Make all dependencies clear and obvious to the user.
-
Script-like behavior: The kernel behaves largely like a regular Nu script to ensure familiarity.
-
Clear Feature Distinctions: Clearly indicating any deviations or limitations compared to standard Nu script capabilities to avoid confusion during notebook executions.
To build the kernel you need to have the rust toolchain installed, check the installation guide on rust's official website.
Using cargo install nu-jupyter-kernel
you can install the latest release of
the kernel.
If you want to install the latest version on the git repo, you can install the
kernel via cargo install nu-jupyter-kernel --git https://github.com/cptpiepmatz/nu-jupyter-kernel.git
After installation, you must register the kernel to make it available within Jupyter environments. This can be done through the command:
nu-jupyter-kernel register
You can specify the registration scope using --user
for the current user
(default) or --system
for system-wide availability.
-
Jupyter Notebook: Open Jupyter Notebook, create or open a notebook, and then select "Nushell" from the kernel options in the top right corner.
-
Visual Studio Code: Ensure you have the Jupyter extension by Microsoft installed. Open a
.ipynb
file, click on "Select Kernel", choose "Jupyter Kernel", and you should see "Nushell" listed.
Both options may require a restart after registering the kernel.
Kernel binary updates do not require re-registration unless the binary's
location changes.
For developers, keep in mind that running cargo run register
and
cargo run --release register
will result in different binary locations.
This crate follows the semantic versioning scheme as required by the
Rust documentation.
The version number is represented as x.y.z+a.b.c
, where x.y.z
is the version
of the crate and a.b.c
is the version of the nu-engine
that this crate is
built with.
The +
symbol is used to separate the two version numbers.
Contributions are welcome!
If you're interested in contributing to the nu-jupyter-kernel
, you can start
by opening an issue or a pull request.
If you'd like to discuss potential changes or get more involved, join the
Nushell community on Discord.
Invite links are available when you start Nushell or on their GitHub repository.
This project uses uv
for integration
testing.
Since tools for executing Jupyter notebooks are not currently available in Rust,
the tests are handled via Python.
To run the tests, follow these steps:
- Register the kernel:
cargo run register
- Sync Python dependencies:
uv sync
- Run the tests:
uv run pytest
Make sure uv
is installed before running the commands.