LangGraph4j 学习系列(1)-顺序工作流

一句话秒懂

LangGraph是一个状态管理和工作流编排框架，专注于构建复杂的多步骤应用，LangGraph4j是LangGraph开源生态中的java实现。

既生瑜？何生亮？

Dify VS LangGraph4j (有了dify，还有必要学习langgraph之类的编排框架吗?)

听听deepseek 怎么说？

 

官网

https://github.com/langgraph4j

 

基本概念

温馨提示：初学者只需了解 节点(NodeAction)、边(EdgeAction)、状态(AgentState) 即可快速上手。

文档

https://deepwiki.com/langgraph4j/langgraph4j

 

 

顺序工作流

 

 

定义节点

import org.bsc.langgraph4j.action.NodeAction;
import org.bsc.langgraph4j.state.AgentState;import java.util.Map;/*** @author junmingyang*/
public class Node1Action implements NodeAction<AgentState> {@Overridepublic Map<String, Object> apply(AgentState state) throws Exception {System.out.println("current Node: node-1");return Map.of("myData", "node1-my-value","node1Key", "node1-value");}
}

　　

import org.bsc.langgraph4j.action.NodeAction;
import org.bsc.langgraph4j.state.AgentState;import java.util.Map;/*** @author junmingyang*/
public class Node2Action implements NodeAction<AgentState> {@Overridepublic Map<String, Object> apply(AgentState state) throws Exception {System.out.println("current Node: node-2");return Map.of("myData", "node2-my-value","node2Key", "node2-value");}
}

注：节点上可以携带数据（即：节点的状态），当工作流运行至本节点时，apply()方法将会调用，业务处理逻辑在此处理（如何：包括调用LLM，调用其它接口来更新状态）

 

完整示例

import org.bsc.langgraph4j.GraphDefinition;
import org.bsc.langgraph4j.GraphRepresentation;
import org.bsc.langgraph4j.GraphStateException;
import org.bsc.langgraph4j.StateGraph;
import org.bsc.langgraph4j.state.AgentState;import java.util.Map;import static org.bsc.langgraph4j.action.AsyncNodeAction.node_async;/*** SequenceGraphApplication 类演示了如何使用 LangGraph4j 构建和执行一个简单的序列图。* 该应用程序创建了一个包含两个节点的序列图，节点按顺序执行：* START -> node-1 -> node-2 -> END* * @author 菩提树下的杨过(yjmyzz.cnblogs.com)* @since 1.0*/
public class SequenceGraphApplication {public static void main(String[] args) throws GraphStateException {StateGraph<AgentState> sequenceGraph = getSequenceGraph();//打印图的mermaid代码System.out.println(sequenceGraph.getGraph(GraphRepresentation.Type.MERMAID, "Sequence Graph", true).content());//编译图并运行（初始数据值为 test->test-init-valuesequenceGraph.compile() //编译.invoke(Map.of("test", "test-init-value")) //运行（并赋初始值).ifPresent(c -> {//运行结束后，会执行本段逻辑System.out.println(c.data());});}public static StateGraph<AgentState> getSequenceGraph() throws GraphStateException {return new StateGraph<>(AgentState::new)//向图中添加2个节点.addNode("node-1", node_async(new Node1Action())).addNode("node-2", node_async(new Node2Action()))//加一条边（开始->node1).addEdge(GraphDefinition.START, "node-1")//加1条边(node1->node2).addEdge("node-1", "node-2")//加1条边(node2->结束).addEdge("node-2", GraphDefinition.END);}}

　　

运行结果

current Node: node-1
current Node: node-2
{node1Key=node1-value, test=test-init-value, node2Key=node2-value, myData=node2-my-value}

　 

解释

第3行是运行到END节点时，图的最终状态数据，可以看到myData被node2中的值覆盖成node2-my-value。而其它key的值，每经过1个节点，都会被merge进AgentState的data map中。 

 

文中源码

https://github.com/yjmyzz/langgraph4j-study/tree/main/src/main/java/org/bsc/langgraph4j/agent/_05_sequence