apache-airflow-providers-pixi

An Apache Airflow 3 provider that runs tasks inside Pixi-managed environments, on the worker or in a Kubernetes pod, so a task can use conda-forge and PyPI packages that are not installed on the worker.

Documentation: https://danplischke.github.io/apache-airflow-providers-pixi/ (for LLMs and coding agents: llms.txt)

Installation

pip install apache-airflow-providers-pixi
pip install "apache-airflow-providers-pixi[cncf.kubernetes]"  # adds PixiKubernetesPodOperator

Requires Apache Airflow 3.1.2 or newer and Python 3.10 to 3.14 (3.14 needs Airflow 3.2). The workers need pixi 0.81.0 or newer on PATH; the provider never installs it. See Requirements and the deployment guide.

Quick start

from airflow.sdk import dag, task


@dag
def my_dag():
    @task.pixi(dependencies={"python": "3.12.*"}, pypi_dependencies=["pandas>=2"])
    def summarize(n: int) -> float:
        import pandas as pd  # imports go inside the function: it runs in the Pixi environment

        return float(pd.Series(range(n)).mean())

    summarize(3)


my_dag()

Pass pixi_project_path or pixi_toml_path instead to use an existing Pixi project. The Pixi Operator guide covers the ways to choose an environment, arguments and return values, the Airflow context, environment variables, errors, timeouts and caches.

Operators, decorators and sensor

Credentials for private conda channels and PyPI indexes go in a pixi connection. Other providers can build on these operators; see Building on the Pixi Operator.

Development

With uv, just and pixi installed:

• just dev installs the package with the dev and docs groups and the pre-commit hooks.
• just test runs the unit tests, with a fake pixi.
• just check runs what CI runs: lint, unit tests, docs build, package build and smoke test.
• just standalone starts a local Airflow at http://localhost:8080 with the DAGs in dev/dags.

just lists the other recipes. Integration, system and compatibility tests are described in System Tests and Installing from sources.

License

MIT