Skip to content

Installing Python packages

Databricks serverless compute comes with a standard set of pre-installed packages including pandas, numpy, scipy, matplotlib, seaborn, and scikit-learn. If you need something beyond that, this page explains how to install additional packages.

No direct PyPI access

Serverless compute does not have internet access. You cannot install packages directly from PyPI using %pip install <package>. Packages are instead installed from pre-downloaded wheel files stored in a Unity Catalog volume.

This is a security control — it prevents data from leaving the TRE and ensures supply chain consistency across the workspace.


1. Check if the package is already installed

Before requesting a new package, check whether it is already available:

# List all installed packages
%pip list
# Check for a specific package
import tabulate
print(tabulate.__version__)

2. Browse available packages in the volume

%sh ls /Volumes/<workspace_catalog_name>/default/packages/

3. Install from the packages volume

# Install a specific wheel
%pip install /Volumes/<workspace_catalog_name>/default/packages/<package_name>.whl

# Install all available wheels
%pip install /Volumes/<workspace_catalog_name>/default/packages/*.whl

4. Request a new package

If the package you need is not in the volume, contact the TVS SDE team. Provide the package name and version (if specific) and we will add it to the volume.


Quick reference

Task How
Check installed packages %pip list
Check a specific package import package_name
Browse available wheels %sh ls /Volumes/<workspace_catalog_name>/default/packages/
Install from volume %pip install /Volumes/<workspace_catalog_name>/default/packages/<file>.whl
Install all available wheels %pip install /Volumes/<workspace_catalog_name>/default/packages/*.whl
Request a new package Contact the TVS SDE team

For R packages

R packages on the classic cluster install differently — see Using R in Databricks for details.