当前位置： 首页 > news >正文

Spring AI Alibaba 10分钟快速入门

news 2025/10/25 12:29:47

Spring AI Alibaba 快速入门指南

本文案例源码笔记见：https://gitee.com/zhengqingya/java-workspace

1. 环境准备

1.1 API Key 配置

访问 阿里云百炼平台 获取API Key，并在Windows环境变量中配置：

AI_DASHSCOPE_API_KEY=your_api_key_here

配置完成后重启电脑使环境变量生效，可通过命令行验证：

echo %AI_DASHSCOPE_API_KEY%

1.2 Ollama本地模型安装（可选）

Ollama是一个用于在本地运行大型语言模型（LLM）的工具和平台。

Windows安装步骤：
  1. 下载安装包：https://github.com/ollama/ollama/releases/download/v0.12.3/OllamaSetup.exe
  2. 配置环境变量（可选）：
    OLLAMA_MODELS=D:\zhengqingya\soft\soft-dev\AI\LLM\Ollama\models
  3. 自定义安装路径：
    OllamaSetup.exe /DIR="D:\zhengqingya\soft\soft-dev\AI\LLM\Ollama"
  4. 验证安装：
    ollama --version
常用Ollama命令：
# 启动 Ollama 服务（默认端口：11434）
ollama serve# 列出已下载的模型
ollama list# 下载指定模型
ollama pull <model_name># 运行指定模型
ollama run <model_name># 停止运行中的模型
ollama stop <model_name># 删除已下载的模型
ollama rm <model_name># 查看当前运行的模型实例
ollama ps

2. 项目依赖配置

2.1 Maven依赖

<properties><!-- Spring Boot --><spring-boot.version>3.4.0</spring-boot.version><!-- Spring AI --><spring-ai.version>1.0.0</spring-ai.version><!-- Spring AI Alibaba --><spring-ai-alibaba.version>1.0.0.3</spring-ai-alibaba.version>
</properties><dependencyManagement><dependencies><dependency><groupId>org.springframework.boot</groupId><artifactId>spring-boot-dependencies</artifactId><version>${spring-boot.version}</version><type>pom</type><scope>import</scope></dependency><!-- 用于为所有组件做统一版本管理 --><dependency><groupId>org.springframework.ai</groupId><artifactId>spring-ai-bom</artifactId><version>${spring-ai.version}</version><type>pom</type><scope>import</scope></dependency><dependency><groupId>com.alibaba.cloud.ai</groupId><artifactId>spring-ai-alibaba-bom</artifactId><version>${spring-ai-alibaba.version}</version><type>pom</type><scope>import</scope></dependency></dependencies>
</dependencyManagement><!-- 核心依赖 -->
<dependency><groupId>com.alibaba.cloud.ai</groupId><artifactId>spring-ai-alibaba-starter-dashscope</artifactId>
</dependency><!-- Ollama本地模型支持（可选） -->
<dependency><groupId>org.springframework.ai</groupId><artifactId>spring-ai-starter-model-ollama</artifactId><version>1.0.0</version>
</dependency>

2.2 基础配置

application.yml 配置
server:port: 888servlet:encoding:enabled: truecharset: utf-8force: truespring:ai:dashscope:api-key: ${AI_DASHSCOPE_API_KEY}# Ollama配置（可选）ollama:base-url: http://localhost:11434chat:model: deepseek-r1 # qwen3:8b | deepseek-r1

3. 基础使用

3.1 简单对话示例

import lombok.RequiredArgsConstructor;
import org.springframework.ai.chat.model.ChatModel;
import org.springframework.web.bind.annotation.GetMapping;
import org.springframework.web.bind.annotation.RequestParam;
import org.springframework.web.bind.annotation.RestController;
import reactor.core.publisher.Flux;@RequiredArgsConstructor
@RestController
public class HelloWorldController {private final ChatModel chatModel;/*** 简单调用* http://localhost:888/simple/chat?msg=你是谁？*/@GetMapping("/simple/chat")public String simpleChat(@RequestParam String msg) {return chatModel.call(msg);}/*** 流式调用* http://localhost:888/stream/chat?msg=你是谁？*/@GetMapping("/stream/chat")public Flux<String> streamChat(@RequestParam String msg) {return chatModel.stream(msg);}
}

3.2 ChatClient使用

ChatClient是更高层次的抽象，提供了简化的聊天接口和流式处理能力。

import com.alibaba.cloud.ai.dashscope.chat.DashScopeChatModel;
import org.springframework.ai.chat.client.ChatClient;
import org.springframework.web.bind.annotation.GetMapping;
import org.springframework.web.bind.annotation.RequestParam;
import org.springframework.web.bind.annotation.RestController;
import reactor.core.publisher.Flux;@RestController
public class ChatClientController {private final ChatClient chatClient;public ChatClientController(DashScopeChatModel dashScopeChatModel) {chatClient = ChatClient.builder(dashScopeChatModel).build();}/*** 简单调用* http://localhost:888/chat-client/simple/chat?msg=你是谁？*/@GetMapping("/chat-client/simple/chat")public String simpleChat(@RequestParam String msg) {return chatClient.prompt().user(msg).call().content();}/*** 流式调用* http://localhost:888/chat-client/stream/chat?msg=你是谁？*/@GetMapping("/chat-client/stream/chat")public Flux<String> streamChat(@RequestParam String msg) {return chatClient.prompt().user(msg).stream().content();}
}

3.3 多模型动态配置

import com.alibaba.cloud.ai.dashscope.chat.DashScopeChatModel;
import com.google.common.collect.Maps;
import io.swagger.v3.oas.annotations.Operation;
import io.swagger.v3.oas.annotations.tags.Tag;
import org.springframework.ai.chat.client.ChatClient;
import org.springframework.ai.chat.model.ChatModel;
import org.springframework.ai.chat.prompt.ChatOptions;
import org.springframework.ai.ollama.OllamaChatModel;
import org.springframework.web.bind.annotation.GetMapping;
import org.springframework.web.bind.annotation.ModelAttribute;
import org.springframework.web.bind.annotation.RequestMapping;
import org.springframework.web.bind.annotation.RestController;
import reactor.core.publisher.Flux;import java.util.Map;@RestController
@RequestMapping("/mutli-platform-model")
@Tag(name = "多模型动态配置")
public class MutliPlatformModelController {private Map<String, ChatModel> STRATEGY = Maps.newHashMap();public MutliPlatformModelController(DashScopeChatModel dashScopeChatModel,OllamaChatModel ollamaChatModel) {STRATEGY.put("dashscope", dashScopeChatModel);STRATEGY.put("ollama", ollamaChatModel);}/*** http://localhost:888/mutli-platform-model/stream/chat?platform=dashscope&model=qwen3-max&temperature=0.8&msg=你是谁？* http://localhost:888/mutli-platform-model/stream/chat?platform=ollama&model=deepseek-r1&temperature=0.8&msg=你是谁？*/@Operation(summary = "动态切换模型")@GetMapping("/stream/chat")public Flux<String> streamChat(@ModelAttribute MutliPlatformModelOptions options) {ChatModel chatModel = STRATEGY.get(options.platform());ChatClient chatClient = ChatClient.builder(chatModel)// 模型配置参数.defaultOptions(ChatOptions.builder().model(options.model()).temperature(options.temperature()).build()).build();return chatClient.prompt().user(options.msg()).stream().content();}
}

4. 提示词（Prompt）使用

4.1 四大角色

public enum MessageType {USER("user"),      // 用户发起的消息或请求ASSISTANT("assistant"), // AI助手生成的回复或响应SYSTEM("system"),  // 系统级别的指令或信息TOOL("tool");      // 工具调用产生的消息
}

4.2 提示词模板

@RestController
@RequestMapping("/prompt")
@Tag(name = "提示词")
public class PromptController {private ChatClient chatClient;private ChatModel chatModel;@Value("classpath:/prompt-template/tell-joke.txt")private org.springframework.core.io.Resource promptTemplateRes;public PromptController(DashScopeChatModel dashScopeChatModel) {chatClient = ChatClient.builder(dashScopeChatModel).build();chatModel = dashScopeChatModel;}private static final String DEFAULT_PROMPT = """你是一位资深Java架构师，专注于企业级Java后端开发。请严格按照以下规则回答：1. 只回答Java及相关技术栈的问题；2. 提供准确、专业的技术解答；3. 对于非Java后端相关问题，请礼貌说明超出了专业范围。""";/*** http://localhost:888/prompt/chat-client?msg=今天吃什么？*/@GetMapping("/chat-client")public String chatClient(@RequestParam String msg) {return chatClient.prompt().system(DEFAULT_PROMPT).user(msg).call().content();}/*** 提示词模板使用* http://localhost:888/prompt/prompt-template?topic=无聊*/@GetMapping("/prompt-template")public Flux<String> promptTemplate(@RequestParam String topic) {// 创建提示词模板PromptTemplate promptTemplate = new PromptTemplate("给我讲一个有关于{topic}的笑话");// 创建 Prompt 实例Prompt prompt = promptTemplate.create(Map.of("topic", topic));// 流式返回结果return chatModel.stream(prompt).map(chatResponse -> chatResponse.getResult().getOutput().getText());}
}

5. Advisor对话拦截

Advisor主要用于拦截ChatClient的对话请求和响应，在对话过程中添加通用处理逻辑。

5.1 日志记录

@RestController
@RequestMapping("/advisor")
@Tag(name = "Advisor对话拦截--日志记录")
public class AdvisorLogController {private ChatClient chatClient;public AdvisorLogController(DashScopeChatModel dashScopeChatModel) {chatClient = ChatClient.builder(dashScopeChatModel).defaultAdvisors(new SimpleLoggerAdvisor()).build();}/*** http://localhost:888/advisor/log?msg=你好*/@GetMapping("/log")public Flux<String> log(@RequestParam String msg) {return chatClient.prompt().user(msg).stream().content();}
}

5.2 敏感词拦截

@RestController
@RequestMapping("/advisor")
@Tag(name = "Advisor对话拦截--敏感词拦截")
public class AdvisorSafeGuardController {private ChatClient chatClient;public AdvisorSafeGuardController(DashScopeChatModel dashScopeChatModel) {chatClient = ChatClient.builder(dashScopeChatModel).defaultAdvisors(new SimpleLoggerAdvisor(),new SafeGuardAdvisor(Lists.newArrayList("敏感词", "WC"),"由于包含敏感内容，我无法对此进行回复。我们可以重新表述或讨论其他话题吗？",1)).build();}/*** http://localhost:888/advisor/safe-guard?msg=你好* http://localhost:888/advisor/safe-guard?msg=WC*/@GetMapping("/safe-guard")public Flux<String> safeGuard(@RequestParam String msg) {return chatClient.prompt().user(msg).stream().content();}
}

5.3 自定义拦截器

/*** <p> 自定义Advisor拦截器 -- 实现重写提示词 </p>*/
public class RewritePromptAdvisor implements BaseAdvisor {private static final String REWRITE_TEXT = """让我们仔细分析这个问题：问题: {input_msg}分析步骤：1. 问题理解: 这个问题在问什么？2. 关键信息: 有哪些重要的细节需要注意？3. 回答构建: 基于以上分析，给出准确详细的回答：回答要求：- 使用通俗易懂的语言- 结合实际情况举例说明- 如涉及技术内容，提供具体的操作步骤- 返回html格式内容""";@Overridepublic ChatClientRequest before(ChatClientRequest chatClientRequest, AdvisorChain advisorChain) {String contents = chatClientRequest.prompt().getContents();String inputMsg = PromptTemplate.builder().template(REWRITE_TEXT).build().render(Map.of("input_msg", contents));return chatClientRequest.mutate().prompt(Prompt.builder().content(inputMsg).build()).build();}@Overridepublic ChatClientResponse after(ChatClientResponse chatClientResponse, AdvisorChain advisorChain) {return chatClientResponse;}@Overridepublic int getOrder() {return 0;}
}

6. 对话记忆（Chat Memory）

用于维护对话上下文状态。

6.1 MySQL存储示例

依赖配置
<dependency><groupId>com.alibaba.cloud.ai</groupId><artifactId>spring-ai-alibaba-starter-memory-jdbc</artifactId><version>1.0.0.2</version>
</dependency>
<dependency><groupId>mysql</groupId><artifactId>mysql-connector-java</artifactId><version>8.0.32</version>
</dependency>
配置文件
spring:datasource:driver-class-name: com.mysql.cj.jdbc.Driverurl: jdbc:mysql://localhost:3306/spring-ai-alibaba-demo?allowMultiQueries=true&useUnicode=true&characterEncoding=UTF8&zeroDateTimeBehavior=convertToNull&useSSL=falseusername: rootpassword: root
代码示例
@RestController
@RequestMapping("/chat-memory")
@Tag(name = "对话记忆")
public class ChatMemoryController {private final ChatClient chatClient;// 注入 JdbcTemplate, ChatClientpublic ChatMemoryController(JdbcTemplate jdbcTemplate, DashScopeChatModel dashScopeChatModel) {// 构造 ChatMemoryRepository 和 ChatMemoryChatMemoryRepository chatMemoryRepository = MysqlChatMemoryRepository.mysqlBuilder().jdbcTemplate(jdbcTemplate).build();ChatMemory chatMemory = MessageWindowChatMemory.builder().chatMemoryRepository(chatMemoryRepository).maxMessages(10) // 消息存储条数.build();this.chatClient = ChatClient.builder(dashScopeChatModel)// 增加聊天记忆能力.defaultAdvisors(MessageChatMemoryAdvisor.builder(chatMemory).build())// 实现 Logger 的 Advisor.defaultAdvisors(new SimpleLoggerAdvisor()).build();}/*** 使用自定义的 Advisor 增加聊天记忆能力* eg:* http://127.0.0.1:888/chat-memory/chat/123?msg=你好，我叫郑清，之后的会话中都带上我的名字* http://127.0.0.1:888/chat-memory/chat/123?msg=我叫什么名字？* http://127.0.0.1:888/chat-memory/chat/111?msg=我叫什么名字？*/@GetMapping("/chat/{id}")public Flux<String> advisorChat(HttpServletResponse response, @PathVariable String id, @RequestParam String msg) {response.setCharacterEncoding("UTF-8");return this.chatClient.prompt(msg).advisors(a -> a.param(ChatMemory.CONVERSATION_ID, id) // 多用户记忆隔离).stream().content();}
}

7. 结构化输出

将模型的输出转换为特定的数据结构。

// 定义数据结构
public record StructuredVO(String category, String question, String answer) {
}@RestController
@RequestMapping("/structured-output")
@Tag(name = "结构化输出")
public class StructuredOutputController {private ChatClient chatClient;public StructuredOutputController(DashScopeChatModel dashScopeChatModel) {chatClient = ChatClient.builder(dashScopeChatModel).build();}/*** http://localhost:888/structured-output/test* {"category":"售后服务","question":"这件商品的质量不好，我可以申请退货吗？","answer":"如果商品存在质量问题，通常可以申请退货。请查看商家的退换货政策或联系客服确认具体流程。"}*/@GetMapping("/test")public StructuredVO structuredOutput() {return chatClient.prompt().system("""对问题分类并简要总结，并给出答案""").user("这件商品的质量不好，我可以申请退货吗？").call().entity(StructuredVO.class);}
}

8. 工具调用（Tool Calling）

允许AI模型调用外部工具和函数来执行特定任务。

8.1 定义工具

方法工具
import org.springframework.ai.tool.annotation.Tool;
import org.springframework.ai.tool.annotation.ToolParam;
import org.springframework.stereotype.Service;@Service
public class TimeTools {@Tool(description = "获取当前时间，默认时间格式：YYYY-MM-DD HH:mm:ss")public String getCurrentTime(@ToolParam(description = "时间格式") String format) {log.info("获取当前时间格式：{}", format);return DateUtil.now();}
}
函数工具
import com.fasterxml.jackson.annotation.JsonPropertyDescription;
import lombok.extern.slf4j.Slf4j;
import org.springframework.context.annotation.Description;
import org.springframework.stereotype.Service;import java.util.function.Function;@Slf4j
@Service(value = "getWeather")
public class WeatherFunction implements Function<String, String> {@Override@Description("获取天气")public String apply(@JsonPropertyDescription("城市") String city) {log.info("获取{}天气", city);return city + "天气：晴";}
}

8.2 注册工具 & 调用

@RestController
@RequestMapping("/tool-calling")
@Tag(name = "工具调用")
public class ToolCallingController {private ChatClient chatClient;@Autowiredprivate TimeTools timeTools;public ToolCallingController(DashScopeChatModel dashScopeChatModel) {chatClient = ChatClient.builder(dashScopeChatModel).build();}/*** 法1：方法工具* http://localhost:888/tool-calling/time?msg=现在时间?*/@GetMapping("/time")public Flux<String> time(@RequestParam String msg) {return chatClient.prompt().user(msg).tools(timeTools).stream().content();}/*** 法2：函数工具* http://localhost:888/tool-calling/weather?msg=成都天气?*/@GetMapping("/weather")public Flux<String> weather(@RequestParam String msg) {return chatClient.prompt().user(msg).toolNames(new String[]{"getWeather"}).stream().content();}
}

8.3 动态注册工具

@RestController
@RequestMapping("/tool-calling")
@Tag(name = "工具调用-动态注册")
public class ToolCallingDynamicController {private ChatClient chatClient;@Autowiredprivate TimeTools timeTools;public ToolCallingDynamicController(DashScopeChatModel dashScopeChatModel,TimeTools timeTools) {chatClient = ChatClient.builder(dashScopeChatModel).build();}/*** 法一：方法工具* http://localhost:888/tool-calling/dynamic/1?msg=现在时间?* http://localhost:888/tool-calling/dynamic/2?msg=现在时间?*/@GetMapping("/dynamic/{userId}")public Flux<String> dynamic(@PathVariable Long userId, @RequestParam String msg) {return chatClient.prompt().user(msg)// 动态注册工具.toolCallbacks(getToolCallbacks(userId)).stream().content();}private List<ToolCallback> getToolCallbacks(Long userId) {List<ToolCallback> toolCallbacks = Lists.newArrayList();if (userId != 1) {return toolCallbacks;}Method method = ReflectionUtils.findMethod(TimeTools.class, "getCurrentTime", String.class);if (method == null) {throw new RuntimeException("Method not found");}String inputSchema = JsonSchemaGenerator.generateForMethodInput(method);ToolCallback timeToolCallback = MethodToolCallback.builder().toolDefinition(ToolDefinition.builder().name("getCurrentTime").description("获取当前时间，默认时间格式：YYYY-MM-DD HH:mm:ss").inputSchema(inputSchema).build()).toolMethod(method).toolObject(timeTools).build();toolCallbacks.add(timeToolCallback);return toolCallbacks;}/*** 法二：函数工具* http://localhost:888/tool-calling/dynamic2/1?msg=成都天气?* http://localhost:888/tool-calling/dynamic2/2?msg=成都天气?*/@GetMapping("/dynamic2/{userId}")public Flux<String> dynamic2(@PathVariable Long userId, @RequestParam String msg) {String[] toolNames = userId == 1 ? new String[]{"getWeather"} : new String[0];return chatClient.prompt().user(msg)// 动态注册工具.toolNames(toolNames).stream().content();}
}

9. MCP（Model Context Protocol）

MCP是一种模型上下文协议，主要用于AI模型调用外部工具和函数。

9.1 传输方式

  1. STDIO（标准输入输出）：通过标准输入输出流进行数据传输
  2. SSE（Server-Sent Events）：基于HTTP的单向服务器推送技术
  3. Streamable HTTP：基于HTTP协议的流式传输

9.2 自定义MCP服务 - STDIO

服务端实现
// pom.xml依赖
<dependency><groupId>org.springframework.ai</groupId><artifactId>spring-ai-starter-mcp-server-webflux</artifactId>
</dependency>// application.yml配置
spring:main:web-application-type: none # 设置为非 Web 应用类型banner-mode: offai:mcp:server:stdio: truename: my-weather-serverversion: 1.0.0// 实现MCP工具
@Service
public class MyWeatherService {@Tool(description = "获取天气")public String getWeather(@ToolParam(description = "城市") String city) {return city + "天气：雨";}
}// 注册MCP工具
@Bean
public ToolCallbackProvider weatherTools(MyWeatherService myWeatherService) {return MethodToolCallbackProvider.builder().toolObjects(myWeatherService).build();
}
客户端接入
spring:ai:mcp:client:stdio:servers-configuration: classpath:/mcp/mcp-servers-config.json

{"mcpServers": {"my-weather-server": {"command": "java","args": ["-Dspring.ai.mcp.server.stdio=true","-Dspring.main.web-application-type=none","-Dlogging.pattern.console=","-jar","D:\\path\\to\\your\\mcp-server.jar"]}}
}

9.3 自定义MCP服务 - SSE

服务端实现
// application.yml配置
server:port: 10088spring:main:banner-mode: offai:mcp:server:name: my-sse-serverversion: 1.0.0type: ASYNCsse-endpoint: /ssesse-message-endpoint: /mcpcapabilities:tool: trueresource: trueprompt: truecompletion: true
客户端接入
spring:ai:mcp:client:sse:connections:my-sse-server:url: http://localhost:10088

10. RAG（Retrieval-Augmented Generation）

RAG是一种结合检索和生成的技术架构，通过引入"外接知识库"让模型在回答问题前先查阅权威资料。

10.1 嵌入模型 (Embedding Model)

spring:ai:dashscope:embedding:options:model: text-embedding-v4dimensions: 64

@RestController
@RequestMapping("/rag")
@Tag(name = "RAG")
public class RagController {private EmbeddingModel embeddingModel;public RagController(DashScopeEmbeddingModel dashScopeEmbeddingModel) {embeddingModel = dashScopeEmbeddingModel;}/*** http://localhost:888/rag/embedding?msg=你好*/@GetMapping("/embedding")public Object embedding(@RequestParam String msg) {EmbeddingResponse embeddingResponse = this.embeddingModel.embedForResponse(List.of(msg));return Map.of("embedding", embeddingResponse);}
}

10.2 向量数据库

使用SimpleVectorStore轻量级向量数据库：

@RestController
@RequestMapping("/rag")
@Tag(name = "RAG")
public class RagController {private VectorStore vectorStore;public RagController(DashScopeEmbeddingModel dashScopeEmbeddingModel) {vectorStore = SimpleVectorStore.builder(dashScopeEmbeddingModel).build();}/*** http://localhost:888/rag/vectorStore?topK=3&threshold=0.1&msg=有哪些开源的向量数据库？*/@GetMapping("/vectorStore")public Object vectorStore(@RequestParam String msg, @RequestParam int topK, @RequestParam double threshold) {// 存储向量vectorStore.add(Lists.newArrayList(Document.builder().text("LangChain是一个用于开发由语言模型驱动的应用程序的框架。").build(),Document.builder().text("Milvus是一款开源向量数据库，专为大规模向量相似性搜索而设计。").build()));// 相似性搜索return vectorStore.similaritySearch(SearchRequest.builder().query(msg).topK(topK).similarityThreshold(threshold).build());}
}

10.3 ELT (Extract, Load, Transform)

文本读取器
@Value("classpath:rag/pet.txt")
private Resource textRes;/*** 文本读取器* http://localhost:888/rag/elt/text*/
@GetMapping("/text")
public Object text() {// 1. 创建文本读取器TextReader textReader = new TextReader(textRes);// 2. 读取文档内容List<Document> documents = textReader.read();// 3. 返回结果return documents;
}
Markdown读取器
<dependency><groupId>org.springframework.ai</groupId><artifactId>spring-ai-markdown-document-reader</artifactId>
</dependency>

@Value("classpath:rag/iphone.md")
private Resource MdRes;/*** markdown读取器* http://localhost:888/rag/elt/markdown*/
@GetMapping("/markdown")
public Object markdown() {// 1. 创建Markdown文档读取器MarkdownDocumentReader markdownReader = new MarkdownDocumentReader(MdRes, MarkdownDocumentReaderConfig.builder().withAdditionalMetadata("filename", MdRes.getFilename()).withHorizontalRuleCreateDocument(false).withIncludeCodeBlock(false).withIncludeBlockquote(false).build());// 2. 读取Markdown文档内容List<Document> documents = markdownReader.read();// 3. 返回文档列表return documents;
}

10.4 文档分割器

TokenTextSplitter是RAG应用中关键的文档预处理工具。

/*** 文本读取器* http://localhost:888/rag/elt/split/text*/
@GetMapping("/text")
public Object text() {// 1. 创建文本读取器TextReader textReader = new TextReader(textRes);// 2. 读取文档内容List<Document> documents = textReader.read();// 3. 文档分割处理List<Document> splitDocs = new TokenTextSplitter(800, 350, 5, 10000, true).split(documents);// 4. 返回结果return splitDocs;
}

10.5 自定义文档分割器

/*** 学校规章制度专用文档分割器 - 针对规章制度结构优化* 支持按"第.*条"分割，保留规章制度的结构完整性*/
public class RegulationDocumentSplitter extends TextSplitter {// 实现细节...public static Builder builder() {return new Builder();}// ... 其他方法实现
}

10.6 文档转换器

关键字提取
@GetMapping("/keyword")
public Object keyword() {// 1. 创建Markdown文档读取器MarkdownDocumentReader markdownReader = new MarkdownDocumentReader(MdRes, MarkdownDocumentReaderConfig.builder().withAdditionalMetadata("filename", MdRes.getFilename()).build());// 2. 读取文档内容List<Document> documents = markdownReader.read();// 3. 关键字提取KeywordMetadataEnricher enricher = new KeywordMetadataEnricher(chatModel, 3);List<Document> enricherDocs = enricher.apply(documents);return enricherDocs;
}
摘要生成
@GetMapping("/summary")
public Object summary() {// 1. 创建Markdown文档读取器MarkdownDocumentReader markdownReader = new MarkdownDocumentReader(MdRes, MarkdownDocumentReaderConfig.builder().withAdditionalMetadata("filename", MdRes.getFilename()).build());// 2. 读取文档内容List<Document> documents = markdownReader.read();// 3. 摘要生成SummaryMetadataEnricher enricher = new SummaryMetadataEnricher(chatModel,List.of(SummaryMetadataEnricher.SummaryType.CURRENT, SummaryMetadataEnricher.SummaryType.NEXT),"""请为以下文档内容生成摘要:{context_str}要求:1. 摘要长度不超过100字2. 突出关键信息3. 保持原意不变""",MetadataMode.ALL);List<Document> enricherDocs = enricher.apply(documents);return enricherDocs;
}
元数据过滤
@GetMapping("/metadata-filter")
public Object metadataFilter() {// 1. 创建Markdown文档读取器MarkdownDocumentReader markdownReader = new MarkdownDocumentReader(MdRes, MarkdownDocumentReaderConfig.builder().withAdditionalMetadata("filename", MdRes.getFilename()).build());// 2. 读取文档内容List<Document> documents = markdownReader.read();// 3. 关键字提取KeywordMetadataEnricher enricher = new KeywordMetadataEnricher(chatModel, 3);List<Document> enricherDocs = enricher.apply(documents);// 4. 存储向量vectorStore.add(enricherDocs);// 5. 相似性搜索List<Document> result = vectorStore.similaritySearch(SearchRequest.builder().query("手机处理器").topK(3)// 基于元数据的过滤条件.filterExpression(new FilterExpressionBuilder().eq("filename", "phone.md").build()).build());// 6. 返回结果return result;
}

10.7 检索增强生成

@RestController
@RequestMapping("/rag/retrieval-augmentation")
@Tag(name = "RAG-检索增强生成")
public class RagRetrievalAugmentationAdvisorController {@Value("classpath:rag/iphone.md")private Resource MdRes;private ChatModel chatModel;private ChatClient chatClient;private VectorStore vectorStore;public RagRetrievalAugmentationAdvisorController(DashScopeChatModel dashScopeChatModel,DashScopeEmbeddingModel dashScopeEmbeddingModel) {chatModel = dashScopeChatModel;chatClient = ChatClient.builder(dashScopeChatModel).defaultAdvisors(new SimpleLoggerAdvisor()).build();vectorStore = SimpleVectorStore.builder(dashScopeEmbeddingModel).build();}/*** RetrievalAugmentationAdvisor 检索增强生成* http://localhost:888/rag/retrieval-augmentation/chat?msg=下雨天，我不想出门，你能不能帮我了解下手机店里的iphone18价格？*/@GetMapping("/chat")public Flux<String> chat(@RequestParam String msg) {// 1. 创建Markdown文档读取器MarkdownDocumentReader markdownReader = new MarkdownDocumentReader(MdRes, MarkdownDocumentReaderConfig.builder().withAdditionalMetadata("filename", MdRes.getFilename()).build());// 2. 读取文档内容List<Document> documents = markdownReader.read();vectorStore.add(documents);// 3. 检索增强生成Advisor advisor = RetrievalAugmentationAdvisor.builder()// documentRetriever：文档检索器.documentRetriever(VectorStoreDocumentRetriever.builder().vectorStore(vectorStore).similarityThreshold(0.5).topK(3).build())// queryAugmenter：查询增强器.queryAugmenter(ContextualQueryAugmenter.builder().allowEmptyContext(false).emptyContextPromptTemplate(PromptTemplate.builder().template("用户查询位于知识库之外。礼貌地告知用户您无法回答。").build()).build())// QueryTransformer：查询转换器.queryTransformers(// 查询重写RewriteQueryTransformer.builder().chatClientBuilder(ChatClient.builder(chatModel)).targetSearchSystem("知识库").build(),// 翻译转换器TranslationQueryTransformer.builder().chatClientBuilder(ChatClient.builder(chatModel)).targetLanguage("中文").build()).build();return chatClient.prompt().user(msg).advisors(advisor).stream().content();}
}

10.8 重排序

@RestController
@RequestMapping("/rag/rerank")
@Tag(name = "重排序")
public class RerankController {private RerankModel rerankModel;private ChatClient chatClient;private VectorStore vectorStore;public RerankController(DashScopeChatModel dashScopeChatModel,DashScopeEmbeddingModel dashScopeEmbeddingModel,DashScopeRerankModel dashScopeRerankModel) {chatClient = ChatClient.builder(dashScopeChatModel).defaultAdvisors(new SimpleLoggerAdvisor()).build();vectorStore = SimpleVectorStore.builder(dashScopeEmbeddingModel).build();rerankModel = dashScopeRerankModel;}/*** http://localhost:888/rag/rerank/chat?msg=iphone20价格*/@GetMapping("/chat")public Flux<String> chat(@RequestParam String msg) {vectorStore.add(Lists.newArrayList(Document.builder().text("iPhone18 基础价格: 5999元, 税率: 13%, 运费: 0元").build(),Document.builder().text("iPhone20 基础价格: 12999元, 税率: 13%, 运费: 50元").build(),Document.builder().text("iPhone10 基础价格: 1299元, 税率: 13%, 运费: 0元").build()));// 重排序RetrievalRerankAdvisor rerankAdvisor = new RetrievalRerankAdvisor(vectorStore, rerankModel,SearchRequest.builder().topK(10).build());return chatClient.prompt().user(msg).advisors(rerankAdvisor).stream().content();}
}

10.9 幻觉评估

@RestController
@RequestMapping("/rag/fact-check")
@Tag(name = "RAG-幻觉评估")
public class RagFactCheckController {private ChatModel chatModel;public RagFactCheckController(DashScopeChatModel dashScopeChatModel) {chatModel = dashScopeChatModel;}@GetMapping("/test")public Object test() {// 准备评估数据ArrayList<Document> documents = Lists.newArrayList(Document.builder().text("iPhone18 基础价格: 5999元, 税率: 13%, 运费: 0元").build(),Document.builder().text("iPhone20 基础价格: 12999元, 税率: 13%, 运费: 50元").build(),Document.builder().text("iPhone10 基础价格: 1299元, 税率: 13%, 运费: 0元").build());// AI回答String aiRes = "iPhone20 6000元";// 执行评估FactCheckingEvaluator evaluator = new FactCheckingEvaluator(ChatClient.builder(chatModel));EvaluationRequest evaluationRequest = new EvaluationRequest(documents, aiRes);EvaluationResponse evaluationResponse = evaluator.evaluate(evaluationRequest);return evaluationResponse; // {"pass":false,"score":0.0,"feedback":"","metadata":{}}}
}

11. AI Agent

AI Agent是一种能够自主感知环境、做出决策并执行动作以实现特定目标的智能实体。

11.1 三种人机协同模式

模式名称核心关系 (人类 : AI)特点与工作方式典型场景
嵌入模式 (Embedded)我做你看将AI能力作为功能模块嵌入现有产品智能客服中的自动回复、推荐系统
副驾驶模式 (Copilot)我们一起做AI作为实时协作的伙伴，提供建议和辅助GitHub Copilot代码补全
智能体模式 (Agent)你做我看AI拥有高度自主权，可独立规划、决策、执行复杂任务自动交易系统、智能助手

11.2 AI Agent的五大工作模式

  1. 提示链模式：将复杂任务拆解成一系列顺序执行的子任务
  2. 路由模式：根据输入内容特点，动态选择最合适的处理路径
  3. 并行化模式：同时启动多个Agent处理独立的子任务
  4. 协调者-工作者模式：模拟指挥体系，由协调者负责任务分解和调度
  5. 评估者-优化者模式：引入自我反思和优化的闭环

12. 可观测性

软件的可观测性是指通过系统输出（如日志、指标、跟踪等）来推断其内部状态的能力。

12.1 Zipkin部署

参考：https://gitee.com/zhengqingya/docker-compose

12.2 Java项目配置

依赖配置
<dependency><groupId>org.springframework.boot</groupId><artifactId>spring-boot-starter-actuator</artifactId>
</dependency>
<dependency><groupId>com.alibaba.cloud.ai</groupId><artifactId>spring-ai-alibaba-starter-tool-calling-weather</artifactId>
</dependency>
<dependency><groupId>io.micrometer</groupId><artifactId>micrometer-tracing-bridge-brave</artifactId><version>1.5.0-M2</version>
</dependency>
<dependency><groupId>io.zipkin.reporter2</groupId><artifactId>zipkin-reporter-brave</artifactId><version>3.4.3</version>
</dependency>
配置文件
spring:ai:dashscope:api-key: ${AI_DASHSCOPE_API_KEY}observations:log-completion: truelog-prompt: true# Chat config itemschat:client:observations:log-prompt: truelog-completion: trueinclude-error-logging: true# tools config itemstools:observability:include-content: truehttp:client:read-timeout: 60smanagement:endpoints:web:exposure:include: "*"endpoint:health:show-details: alwaystracing:sampling:probability: 1.0zipkin:tracing:endpoint: http://localhost:9411/api/v2/spans

13. Spring AI Alibaba Graph

Spring AI Alibaba Graph是一个强大且直观的框架，能让你像搭积木一样，通过编排节点和流程来构建复杂的AI应用。

13.1 快速入门

依赖配置
<dependency><groupId>com.alibaba.cloud.ai</groupId><artifactId>spring-ai-alibaba-graph-core</artifactId>
</dependency>
创建自定义节点
public class ExpanderNode implements NodeAction {private final ChatClient chatClient;public ExpanderNode(DashScopeChatModel chatModel) {this.chatClient = ChatClient.builder(chatModel).build();}@Overridepublic Map<String, Object> apply(OverAllState state) throws Exception {// 1. 从全局状态中获取输入String query = state.value("query", "");// 2. 构建提示词并调用大模型String promptTemplate = "请为以下问题生成3个不同角度的变体问题：{query}";String result = chatClient.prompt().user(u -> u.text(promptTemplate).param("query", query)).call().content();// 3. 处理结果List<String> queryVariants = Arrays.asList(result.split("\n"));// 4. 将结果放入MapMap<String, Object> resultMap = new HashMap<>();resultMap.put("expander_content", queryVariants);return resultMap;}
}
配置状态图
@Configuration
public class SimpleGraphConfiguration {@SneakyThrows@Beanpublic StateGraph simpleGraph(DashScopeChatModel chatModel) {// 全局变量的替换策略KeyStrategyFactory keyStrategyFactory = () -> {HashMap<String, KeyStrategy> strategies = new HashMap<>();strategies.put("query", new ReplaceStrategy());strategies.put("expander_content", new ReplaceStrategy());return strategies;};// 构建状态图StateGraph stateGraph = new StateGraph("问题扩展工作流", keyStrategyFactory)// 添加节点.addNode("expander", AsyncNodeAction.node_async(new ExpanderNode(chatModel)))// 添加边.addEdge(StateGraph.START, "expander").addEdge("expander", StateGraph.END);return stateGraph;}
}
编译并运行工作流
@RestController
@RequestMapping("/graph")
@Tag(name = "Graph")
public class SimpleGraphController {private final CompiledGraph compiledGraph;// 注入定义好的StateGraph，并编译成CompiledGraphpublic SimpleGraphController(@Qualifier("simpleGraph") StateGraph stateGraph) throws GraphStateException {this.compiledGraph = stateGraph.compile();}/*** http://localhost:888/graph/expand?query=什么是人工智能*/@GetMapping("/expand")public Map<String, Object> expandQuery(@RequestParam String query) throws GraphRunnerException {// 设置初始状态Map<String, Object> initialState = Map.of("query", query);// 执行工作流Optional<OverAllState> result = compiledGraph.invoke(initialState);// 从最终状态中获取结果return result.map(OverAllState::data).orElse(Map.of());}
}

学习资源

技术选型

  1. Spring AI Alibaba
  2. LangChain4j

教学视频

已阅来源
√(1天看完-推荐看)Spring AI 1.0全套教程（入门+实战+原理&源码；大模型+tools+mcp+Agent全流程落地）
√(1h看完-推荐看)LangChain4j实战教程
√(1天看完)Spring AI Alibaba入门到进阶实战
√(1h看完)Coze（扣子）智能体教程
√(1天看完，不建议看，个人觉得讲的太啰嗦)尚硅谷Spring AI Alibaba教程，2025最新springai从基础到实战

文档资源

tips: 书写本项目案例代码的时候，个人建议根据Spring AI和Spring AI Alibaba 2个官方文档以及官方demo来，避免踩坑~

  • Spring AI Alibaba 官方文档
  • Spring AI
  • ReLE中文大模型能力评测
  • MCP Server 工具与插件
  • DeepSeek 提示库
  • Spring AI 中的 DocumentTransformer 与 RAG 深度解析
  • Spring AI 系列博客文章

今日励志语句

每一次代码提交都是进步的足迹，每一行注释都是智慧的积累。编程路上没有捷径，只有不断的学习和实践。遇到bug不要气馁，它们是成长路上的垫脚石。保持好奇心，拥抱变化，让每一行代码都闪耀着创新的光芒。今天也要元气满满地coding哦！

http://www.dtcms.com/a/525411.html

相关文章：

  • 做网站被罚款门户网站建设哪专业
  • 电子电气架构车载网关系列——常见网关芯片特点
  • trae ide 设置 terminal 使用 powershell , 默认加载 用户和系统环境变量
  • 有哪些摄影网站电子商务网站建设结业论文
  • 网站建设 应该考虑什么wordpress 友情链接调用
  • 嵌入式开发 | C语言 | 单精度浮点数4字节可以表示的范围计算过程
  • JMeter测试HTTP POST（附实例）
  • 网站建设 页面网站ip地址查询
  • UM681A相关参数性能介绍
  • 建企业网站多少钱官网cms
  • html制作音乐网站Erphpdown wordpress
  • 最快做网站的语言wordpress 函数api文件
  • php企业网站开发实验总结软件开发工具包sdk
  • java发送SOAP请求
  • Web单页应用（SPA）路由设计（以React为例）
  • 如何把网站放到域名上广告推广平台哪个好
  • 《操作系统真象还原》 第十一章 用户进程
  • maven打jar包，将依赖的jar提取出来
  • 现在做网站用什么工具营销型网站的推广
  • V-Ray for Blender正式上线，新功能概览
  • Blender入门学习06 - 粒子
  • 在什么网站上可以找设计兼职来做电子商务网站html模板
  • 网站索引量白城网络推广
  • mysql 学习网站网站开发公司照片
  • 小迪安全v2023学习笔记（一百三十七讲）—— Win系统权限提升篇UAC绕过DLL劫持未引号路径可控服务全检项目
  • golang使用泛型
  • 自己做的小说网站要交税吗免费的站内推广方式有哪些
  • Anthropic将使用100万个TPU训练大模型
  • 阿里云服务器网站建设找销售的网站
  • 【PPT导出高清tif图】利用PPT将子图组合并导出高清图片