子图¶

子图是一个用作另一个图中的节点的图——这是应用于 LangGraph 的封装概念。子图允许您构建包含多个组件（这些组件本身就是图）的复杂系统。

使用子图的一些原因包括

• 构建多智能体系统
• 当您想在多个图中重用一组节点时
• 当您希望不同团队独立地处理图的不同部分时，您可以将每个部分定义为子图，只要子图接口（输入和输出模式）得到遵守，父图就可以在不了解子图任何细节的情况下构建

添加子图时的主要问题是父图和子图如何通信，即它们在图执行期间如何相互传递状态。有两种情况

• 父图和子图在其状态模式中具有共享状态键。在这种情况下，您可以将子图作为节点包含在父图中

  from langgraph.graph import StateGraph, MessagesState, START
  
  # Subgraph
  
  def call_model(state: MessagesState):
      response = model.invoke(state["messages"])
      return {"messages": response}
  
  subgraph_builder = StateGraph(State)
  subgraph_builder.add_node(call_model)
  ...
  subgraph = subgraph_builder.compile()
  
  # Parent graph
  
  builder = StateGraph(State)
  builder.add_node("subgraph_node", subgraph)
  builder.add_edge(START, "subgraph_node")
  graph = builder.compile()
  ...
  graph.invoke({"messages": [{"role": "user", "content": "hi!"}]})
  

• 父图和子图具有不同的模式（其状态模式中没有共享状态键）。在这种情况下，您必须从父图中的节点内部调用子图：当父图和子图具有不同的状态模式，并且您需要在调用子图之前或之后转换状态时，这很有用

  from typing_extensions import TypedDict, Annotated
  from langchain_core.messages import AnyMessage
  from langgraph.graph import StateGraph, MessagesState, START
  from langgraph.graph.message import add_messages
  
  class SubgraphMessagesState(TypedDict):
      subgraph_messages: Annotated[list[AnyMessage], add_messages]
  
  # Subgraph
  
  def call_model(state: SubgraphMessagesState):
      response = model.invoke(state["subgraph_messages"])
      return {"subgraph_messages": response}
  
  subgraph_builder = StateGraph(SubgraphMessagesState)
  subgraph_builder.add_node("call_model_from_subgraph", call_model)
  subgraph_builder.add_edge(START, "call_model_from_subgraph")
  ...
  subgraph = subgraph_builder.compile()
  
  # Parent graph
  
  def call_subgraph(state: MessagesState):
      response = subgraph.invoke({"subgraph_messages": state["messages"]})
      return {"messages": response["subgraph_messages"]}
  
  builder = StateGraph(State)
  builder.add_node("subgraph_node", call_subgraph)
  builder.add_edge(START, "subgraph_node")
  graph = builder.compile()
  ...
  graph.invoke({"messages": [{"role": "user", "content": "hi!"}]})