Python assets

A Python asset runs a script as part of the pipeline — for anything SQL can’t express: calling a library, scoring a model, reshaping data with pandas.

Create one

On the canvas, add an asset and choose Python. The asset starts with a materialize() function — that’s the contract:

import pandas as pd


def materialize():
    return pd.DataFrame(
        [
            {"customer_id": 1, "customer_name": "Ada Lovelace"},
            {"customer_id": 2, "customer_name": "Grace Hopper"},
            {"customer_id": 3, "customer_name": "Katherine Johnson"},
        ]
    )

When the asset materializes, Renart runs the file and writes whatever materialize() returns — a DataFrame or a list of dicts — to the target table, like any other asset.

Query project data

Use the built-in renart SDK to read another asset without putting connection credentials in your Python process:

from renart import context, query


def materialize():
    orders = query(
        "select customer_id, total_amount from staging.orders",
    ).to_pandas()
    return orders.groupby("customer_id", as_index=False)["total_amount"].sum()

query() accepts one read-only SELECT and returns a PyArrow Table. Call .to_pandas() as above, or pass format="pandas", when you want a pandas DataFrame. It uses the asset’s connection by default; connection="warehouse-name" reads through another connection defined in the project. Renart executes the query inside the runner, so the Python process receives the result but never the connection credentials.

The context object exposes the current run window, run ID, pipeline variables, asset name, and full-refresh state.

Dependencies and lineage

Renart can’t infer upstreams from Python the way it does from SQL. Declare them in the properties panel’s Lineage tab; the asset then runs after its upstreams and appears in the lineage like every other node. If a literal query() reads another project asset without a matching dependency, Type check warns about it. During a run, query() waits when a declared upstream is still materializing rather than reading a half-written table.

Packages

Python assets run in the project’s local virtualenv. When you import a package that isn’t declared yet, the editor shows a banner offering to add it to the project’s dependencies with one click — the same flow notebooks use.

Editing support

Python previews execute code

Inspect does not materialize the destination table, but it still runs your Python code. File writes, network calls, and other side effects in that code are not undone. Review an unfamiliar script before previewing it.

The editor gives Python assets diagnostics, completions, hover, and go-to-definition, alongside the usual Inspect/Materialize flow in the results panel.

Related

• SQL assets — the default choice when a query can do the job.
• Notebooks — for exploration before the logic earns a place in the pipeline.