Importing

Load existing Fabric pipelines into typed Python objects, then re-export or generate source.

Parse pipeline JSON

Fabric pipeline-content.json does not store the pipeline name (that lives on the item folder / .platform). Pass name= explicitly:

from fabric_data_pipelines import Pipeline

pipeline = Pipeline.from_json(json_text, name="Daily_Silver_Sales")
# or
pipeline = Pipeline.from_dict(data, name="Daily_Silver_Sales")

Modeled activity types become concrete classes (Notebook, Copy, IfCondition, …). Anything unmodeled becomes RawActivity with the original type and typeProperties preserved.

Load a Fabric Git item folder

from fabric_data_pipelines import Pipeline, load_workspace

pipeline = Pipeline.load_item("out/Daily_Silver_Sales.DataPipeline")
# logicalId from .platform and schedules from .schedules are restored when present

pipelines = load_workspace("out")  # all *.DataPipeline/ folders, including nested
pipelines_top = load_workspace("out", recursive=False)  # immediate children only

load_item also accepts a workspace directory that contains exactly one *.DataPipeline folder. load_workspace discovers item folders recursively by default; pass recursive=False for a shallow scan.

Round-trip

loaded = Pipeline.load_item(item_dir)
loaded.save_item("out_roundtrip")

Fidelity is semantic: the typed graph and re-exported Fabric JSON match for library-emitted content. Policy defaults the models inject may differ from sparse UI exports; unknown connectors/datasets use generic escape hatches (CopySource, Dataset, RawActivity).

Generate Python with the CLI

fabric-data-pipelines codegen path/to/Item.DataPipeline -o pipeline.py
fabric-data-pipelines codegen path/to/pipeline-content.json --name MyPipeline -o pipeline.py

The CLI emits reviewable constructors for modeled types and RawActivity(...) for the rest. See Exporting for the write path.

For the end-to-end UI/Git → Python → re-export workflow, see Migrate from Fabric.