Skip to main content
Version: v0.1.73

Code Environments

A code environment is a managed Python virtual environment with a declared package manifest. It lets a Python recipe run against a known, reproducible set of dependencies instead of the platform's base interpreter, and lets two recipes in the same project use different package sets without colliding.

This mirrors Dataiku's code environments: you declare what a recipe needs, the platform builds the environment, and the recipe selects it.

Declaring an environment​

An environment is defined by a package manifest — the list of pip packages (and versions) it should contain. The platform provisions a virtual environment from the manifest and keeps it available for recipes to select. Rebuild the environment when the manifest changes.

Per-recipe selection​

A Python recipe selects which code environment interpreter runs it. The selected environment's interpreter is what executes the recipe body, so the recipe sees exactly the packages the environment declares — nothing more, nothing less. Recipes that don't select an environment run on the platform's base interpreter.

Because selection is per-recipe, one project can have a lightweight environment for most recipes and a heavier one (e.g. with a modeling or geospatial library) for the few recipes that need it.

Where code environments apply​

Code environments back Python recipes on the Flow. SQL and visual recipes compile to warehouse SQL and don't use a Python interpreter, so they are unaffected. Structured/code agents run in their own subprocess-isolated sandbox and are a separate runtime from project code environments.

See also​