Execution engines

Graph ships with local engine so you can run the graph locally in Python.

LocalEngine

LocalEngine executes a graph synchronously in-process and records runtime provenance as it goes.

from node_graph import Graph, task
from node_graph.engine.local import LocalEngine

@task()
def add(x: float, y: float) -> float:
    return x + y

ng = Graph()
left = ng.add_task(add, "left", x=1, y=2)
ng.add_task(add, "right", x=3, y=left.outputs.result)

engine = LocalEngine(name="demo")
results = engine.run(ng)
print(results["right"]["result"])  # 6

The engine keeps a ProvenanceRecorder instance that captures process, data, and edge information for every task invocation. Nested subgraphs reuse the same recorder so you obtain a single provenance DAG per run. Provenance can be exported as JSON or GraphViz:

recorder = engine.recorder
recorder.save_json("run.json")
recorder.save_graphviz("run.dot")