Use scripts and functions in pipelines

Where: a transform's Script or Function node. Connect the upstream data before configuring arguments and outputs.

Script node#

1. Add Script and select its language.
2. Add dependencies only when the code requires them. Standard-library-only scripts have the simplest runtime setup.
3. Declare the output columns in Output columns.
4. Write the code in Script.
5. Preview the node with a small upstream sample.

The Script panel exposes language, packages, output columns, and code.

The Python starter uses main(input) and returns one dictionary per output row. A minimal example preserves every upstream field:

def main(input):
    return [dict(row) for row in input]

When adding a calculated field, update the declared output schema too. A downstream node can only map the columns the pipeline knows are available. Inspect the selected language's starter rather than assuming its entry function matches a standalone Platform function.

Function node#

1. Create and test the reusable function in Functions first.
2. Add Function and select it.
3. Map upstream columns to its argument/input keys.
4. Map the function's output keys back to pipeline columns.
5. Preview a known input and check the result against a direct function test.

Function configuration connects a reusable function to upstream and downstream fields.

A function can need entity, secret, or AI provider bindings in addition to pipeline columns. Configure these in the function's access/settings form; naming an argument does not grant resource access.

Diagnose execution errors#

Check the run/preview error, entry-point contract, declared outputs, dependency availability, and input types. Test a pure calculation before adding an external API call. If a script or function can write outside Platform, inspect that system before retrying an interrupted execution.