Kedro starters

A Kedro starter contains code in the form of a Cookiecutter template for a Kedro project. Metaphorically, a starter is similar to using a pre-defined layout when creating a presentation or document.

Kedro starters provide pre-defined example code and configuration that can be reused, for example:

  • As template code for a typical Kedro project

  • To add a docker-compose setup to launch Kedro next to a monitoring stack

  • To add deployment scripts and CI/CD setup for your targeted infrastructure

You can create your own starters for reuse within a project or team, as described in the documentation about how to create a Kedro starter.

How to use Kedro starters

To create a Kedro project using a starter, apply the --starter flag to kedro new:

kedro new --starter=<path-to-starter>


path-to-starter could be a local directory or a VCS repository, as long as Cookiecutter supports it.

To create a project using the PySpark starter:

kedro new --starter=pyspark

Starter aliases

We provide aliases for common starters maintained by the Kedro team so that users don’t have to specify the full path. For example, to use the PySpark starter to create a project:

kedro new --starter=pyspark

To list all the aliases we support:

kedro starter list

List of official starters

The Kedro team maintains the following starters for a range of Kedro projects:

Starter versioning

By default, Kedro will use the latest version available in the repository, but if you want to use a specific version of a starter, you can pass a --checkout argument to the command:

kedro new --starter=pyspark --checkout=0.1.0

The --checkout value points to a branch, tag or commit in the starter repository.

Under the hood, the value will be passed to the --checkout flag in Cookiecutter.

Use a starter with a configuration file

By default, when you create a new project using a starter, kedro new asks you to enter the project_name, which it uses to set the repo_name and python_package name. This is the same behavior as when you create a new empty project

However, Kedro also allows you to specify a configuration file when you create a project using a Kedro starter. Use the --config flag alongside the starter:

kedro new --config=my_kedro_pyspark_project.yml --starter=pyspark

This option is useful when the starter requires more configuration than the default mode requires.

How to create a Kedro starter

Kedro starters are used to create projects that contain code to run as-is, or to adapt and extend. A good example is the Iris dataset example of basic Kedro project layout, configuration and initialisation code. A team may find it useful to build Kedro starters to create reusable projects that bootstrap a common base and can be extended.

A Kedro starter is a Cookiecutter template that contains the boilerplate code for a Kedro project.

To create a Kedro starter, you need a base project to convert to a cookiecutter template, which forms the boilerplate for all projects that use the Kedro starter.

Install cookiecutter as follows:

pip install cookiecutter

You then need to decide which are:

  • the common, boilerplate parts of the project

  • the configurable elements, which need to be replaced by cookiecutter strings

Configuration variables

By default, when you create a new project using a Kedro starter, kedro new launches in interactive mode. The user is then prompted for the variables that have been set in prompts.yml.

The most basic and empty starter triggered by kedro new is set up with the following variable:

  • project_name - A human readable name for the new project

Kedro will then automatically generate the following two variables from the entered project_name:

  • repo_name - A name for the directory that holds the project repository

  • python_package - A Python package name for the project package (see Python package naming conventions)

See the configuration for this basic configuration in the default starter setup.

As the creator of the Kedro starter you can customise the prompts triggered by kedro new by adding your own prompts in prompts.yml. This is an example of a custom prompt:

    title: "Prompt title"
    text: |
      Prompt description that explains to the user what
      information they should provide.

At the very least, the prompt title must be defined for the prompt to be valid. After Kedro gets the user’s input for each prompt, we pass the value to cookiecutter, so every key in your prompts.yml must have a corresponding key in cookiecutter.json.

If the input to the prompts needs to be validated, for example to make sure it only has alphanumeric characters, you can add regex validation rules via the regex_validator key. For more complex validation, have a look at cookiecutter pre/post-generate hooks.

If you want cookiecutter to provide sensible defaults in case a user doesn’t provide any input, you can add those to cookiecutter.json. See the default starter cookiecutter.json as example.

Example Kedro starter

To review an example Kedro starter, check out the spaceflights-pandas starter on GitHub.

When a new spaceflights-pandas project is created with kedro new --starter=spaceflights-pandas, the user is asked to enter a project_name variable, which is then used to generate the repo_name and python_package variables by default.

If you use a configuration file, you must supply all three variables in the file. You can see how these variables are used by inspecting the template:


The human-readable project_name variable is used in the for the new project.


The top-level folder labelled {{ cookiecutter.repo_name }}, which forms the top-level folder to contain the starter project when it is created.


Within the parent folder, inside the src subfolder, is another configurable variable {{ cookiecutter.python_package }} which contains the source code for the example pipelines. The variable is also used within

Here is the layout of the project as a Cookiecutter template:

{{ cookiecutter.repo_name }}     # Parent directory of the template
├── conf                         # Project configuration files
├── data                         # Local project data (not committed to version control)
├── docs                         # Project documentation
├── notebooks                    # Project related Jupyter notebooks (can be used for experimental code before moving the code to src)
├── pyproject.toml               #
├──                    # Project README
├── requirements.txt
└── src                          # Project source code
    └── {{ cookiecutter.python_package }}
       ├── __init.py__
       ├── pipelines
└── tests