feat: Add Subgraph Visualization Support

[Chajf]

Description

Currently, when a compiled graph is created from other graphs (subgraphs), the visualization renders each subgraph as a single node labeled with the subgraph’s name.

While this keeps the top-level graph clean, it limits visibility into the internal structure of nested graphs. For more complex systems, especially multi-agent or hierarchical workflows, this makes debugging and analysis significantly harder.

Proposed Feature

Add support for subgraph expansion / nested visualization, allowing users to view the internal nodes and edges of subgraphs.

[ArtyomVancyan]

@Chajf, tested with DeepAgents and it shows all sub-steps. Can you please share a simplified version of your graph workflow with subgraphs, so I can test and validate the results?

[Chajf]

Thanks for checking.
Here is a modified version of the basic agent example with an added subgraph pattern similar to what I’m using in my project.

basic_agent_subgraph.py

Below is the subgraph implementation logic:

def create_subgraph():
    builder = StateGraph(State)

    builder.add_node("node1", example_node)
    builder.add_node("node2", example_node)
    builder.add_node("node3", example_node)

    builder.add_edge(START, "node1")
    builder.add_edge("node1", "node2")
    builder.add_edge("node2", "node3")
    builder.add_edge("node3", END)
    

    return builder.compile()

def create_main_graph():
    builder = StateGraph(State)

    subgraph = create_subgraph()
    builder.add_node("subgraph", subgraph)
    builder.add_node("summariser_runner", summariser_runner)
    builder.add_node("responder", responder)
    builder.add_node("tools", tool_node)

    builder.add_edge(START, "summariser_runner")
    builder.add_edge("summariser_runner", "subgraph")
    builder.add_edge("subgraph", "responder")

    builder.add_conditional_edges(
        "responder",
        route_after_respond,
        {"tools": "tools", "__end__": END},
    )

    builder.add_edge("tools", "responder")

    return builder.compile()


def main() -> None:
    graph = create_main_graph()
    graph = watch(graph)

    graph.invoke({"messages": [HumanMessage(content="What is the meaning of life?")]})

When visualizing graph, the subgraph is shown as a single node rather than exposing its structure.

[ArtyomVancyan]

This only applies to the graph visualization, as the library implements it for the main graph nodes, where a node can be a subgraph. If you open the inspect panel in full mode, you can see all the sub-setps of the subgraph, so you can see what's going on underneath, but because of the structure of the library, it's impossible to get the topology of the subgraph in advance. It would be possible if we showed the visualization at the end of the execution, which contradicts the real-time visualization feature of the library. So, you can still see the child nodes of your subgraph in the inspect panel, but they cannot dynamically change the graph of the main agent workflow.

[Chajf]

That's a shame it's not possible to visualize the nested structure of subgraphs. When working on more complex flows, composing them with subgraphs is my main way of structuring things in LangGraph.

Anyway, thanks for the explanation and the quick response!