Composite Components

Composite components are reusable, prebuilt subgraphs in MASFactory. They are implemented with standard Graph/Loop/Agent primitives, and can be created directly as nodes inside your RootGraph.

Common built-in composite components include:

• VerticalGraph
• VerticalDecisionGraph
• VerticalSolverFirstDecisionGraph
• HorizontalGraph
• AdjacencyListGraph
• AdjacencyMatrixGraph
• BrainstormingGraph
• HubGraph
• MeshGraph
• InstructorAssistantGraph
• PingPongGraph

Import them from masfactory directly and use them like any other graph node.

Most composite components take keyword-style configuration. In declarative graphs, use NodeTemplate(...) to pass those kwargs.

---

Minimal example: embed a HorizontalGraph

HorizontalGraph is a prebuilt sequential pipeline: ENTRY -> node[0] -> node[1] -> ... -> EXIT.

Declarative (recommended)

from masfactory import CustomNode, HorizontalGraph, NodeTemplate, RootGraph

Pipeline = NodeTemplate(
    HorizontalGraph,
    node_args_list=[
        {"cls": CustomNode, "name": "a", "forward": lambda d: {"x": int(d["n"]) + 1}},
        {"cls": CustomNode, "name": "b", "forward": lambda d: {"y": int(d["x"]) * 2}},
    ],
    edge_keys_list=[{"x": "a output"}],
)

g = RootGraph(
    name="composite_demo",
    nodes=[("pipeline", Pipeline)],
    edges=[
        ("entry", "pipeline", {"n": "input number"}),
        ("pipeline", "exit", {"y": "final output"}),
    ],
)

g.build()
out, _attrs = g.invoke({"n": 3})
print(out)  # {'y': 8}

Imperative (alternative)

from masfactory import CustomNode, HorizontalGraph, RootGraph

g = RootGraph(name="composite_demo")

pipeline = g.create_node(
    HorizontalGraph,
    name="pipeline",
    node_args_list=[
        {"cls": CustomNode, "name": "a", "forward": lambda d: {"x": int(d["n"]) + 1}},
        {"cls": CustomNode, "name": "b", "forward": lambda d: {"y": int(d["x"]) * 2}},
    ],
    edge_keys_list=[{"x": "a output"}],
)

g.edge_from_entry(pipeline, {"n": "input number"})
g.edge_to_exit(pipeline, {"y": "final output"})

g.build()
out, _attrs = g.invoke({"n": 3})
print(out)  # {'y': 8}