Install the legacy Python CLI for AI Runtime
This documentation has been retired and might not be updated.
The Python-based air CLI, installed with the databricks-air package, is now deprecated and is no longer actively maintained.
Use the Databricks CLI for new workloads. See Use the Databricks CLI with AI Runtime.
Install the air CLI with uv and authenticate it against a Databricks workspace using a Databricks CLI profile. The CLI requires Python 3.10 or newer.
Requirements
- Python 3.10 or newer.
- A Databricks workspace with AI Runtime enabled. See Requirements.
- The Databricks CLI, which manages authentication profiles in
~/.databrickscfg.
Install the CLI
Databricks recommends installing the CLI with uv:
uv tool install --force databricks-air --python 3.12
uv tool install puts air in its own isolated environment and exposes it on your PATH, so it doesn't conflict with the Python interpreter you use for your training code.
--python 3.12 is recommended but optional. If you do not specify the Python version, uv uses the latest available version that satisfies the package's Python constraint.
If you don't already have uv, install it first:
curl -LsSf https://astral.sh/uv/install.sh | sh
Verify the installation:
air --version
air --help
Authenticate
The AI Runtime CLI reuses Databricks CLI authentication profiles. Log in to your workspace and name the profile when prompted:
databricks auth login --host https://<your-workspace>.cloud.databricks.com
Pass the profile name to any air command with -p. For example:
air list runs -p my-workspace
Alternatively, set DATABRICKS_CONFIG_PROFILE in your shell to make a profile the default:
export DATABRICKS_CONFIG_PROFILE=my-workspace
For all authentication options, see Authentication for the Databricks CLI.
Additional resources
After installing, define workloads in a train.yaml config with inline dependencies. Start with the quickstart, then use the YAML reference as you build out your config: