Spring AI Alibaba 92｜Graph Core · 为图添加持久化 · 完整示例：带工具调用的持久化对话

import com.alibaba.cloud.ai.graph.*;
import org.springframework.ai.chat.client.ChatClient;
import org.springframework.ai.model.function.FunctionCallbackWrapper;

public class PersistenceExample {

    private static final Logger log = LoggerFactory.getLogger(PersistenceExample.class);

    public static void main(String[] args) {
        // 1. 配置状态策略
        KeyStrategyFactory keyStrategyFactory = () -> {
            HashMap<String, KeyStrategy> strategies = new HashMap<>();
            strategies.put("messages", new AppendStrategy());
            strategies.put("user_name", new ReplaceStrategy());
            return strategies;
        };

        // 2. 创建工具和节点
        SearchTool searchTool = new SearchTool();
        ChatClient.Builder builder = ChatClient.builder(chatModel);
        var agentNode = nodeasync(new AgentNode(builder, searchTool));

        // 3. 构建 Graph
        StateGraph workflow = new StateGraph(keyStrategyFactory)
            .addNode("agent", agentNode)
            .addEdge(StateGraph.START, "agent")
            .addEdge("agent", StateGraph.END);

        // 4. 配置持久化
        var checkpointer = new MemorySaver();
        var compileConfig = CompileConfig.builder()
            .checkpointSaver(checkpointer)
            .build();

        CompiledGraph graph = workflow.compile(compileConfig);

        // 5. 测试持久化对话
        var config = RunnableConfig.builder()
            .threadId("demo-session")
            .build();

        // 第一轮对话
        graph.invoke(Map.of("messages", "Hi, I'm Charlie"), config);

        // 第二轮对话 - 能记住名字
        var result = graph.invoke(Map.of("messages", "What's my name?"), config);
        log.info("Response: {}", result.data().get("messages"));

        // 第三轮对话 - 使用工具
        result = graph.invoke(Map.of("messages", "What's the weather like?"), config);
        log.info("Response: {}", result.data().get("messages"));

        // 查看状态历史
        List<StateSnapshot> history = graph.getStateHistory(config);
        log.info("Total conversation steps: {}", history.size());
    }
}

1. 多轮对话系统: 记住用户的上下文和偏好
2. 客服机器人: 跨会话跟踪客户问题
3. 工作流状态恢复: 长时间运行的任务可以中断和恢复
4. A/B 测试: 比较不同会话的处理结果
5. 审计和调试: 追踪完整的对话历史

• 等待用户输入 – 中断和恢复示例
• 时光旅行 – 状态历史导航
• 快速入门 – Graph 基础使用

本文整理自 Spring AI Alibaba 官方文档《为图添加持久化》：https://java2ai.com/docs/frameworks/graph-core/examples/persistence