1、前言
2025年5月,SpringAI 1.0.0终于正式发布。这不仅是另一个普通的库,更是将Java和Spring推向AI革命前沿的战略性举措。给Java生态带来了强大且全面的AI工程解决方案。众多企业级应用在SpringBoot上运行关键业务,而SpringAI 1.0.0的发布,将赋予开发者将应用程序与前沿AI模型无缝连接的能力!
官方文档中已提供了众多能力的说明,旨在简化大模型的应用程序的开发。
另外,开源大模型的选择(如deepseekR1(0528版)),不同蒸馏模型的选择,可参考github上的开源大模型排行榜
2、运行环境
SpringAI基于spingboot3.x版本,需要JDK17以上。
<parent>
<groupId>org.springframework.boot</groupId>
<artifactId>spring-boot-starter-parent</artifactId>
<version>3.4.5</version>
<relativePath/> <!-- lookup parent from repository -->
</parent>
<dependencyManagement>
<dependencies>
<dependency>
<groupId>org.springframework.ai</groupId>
<artifactId>spring-ai-bom</artifactId>
<version>1.0.0</version>
<type>pom</type>
<scope>import</scope>
</dependency>
</dependencies>
</dependencyManagement>
3、Api-Key申请
SpringAI提供多种AI提供商的便携式Model,包括各类多模态:图像识别、语音识别、视频识别,以及最基本的LLM文本对话,例如:Claude、OpenAI、DeepSeek、ZhiPu等。
本文使用智谱AI的大模型演示,新用户可获得有期限的免费次数
也可以使用本地安装大模型:https://ollama.com/。通过ollama,就不再需要环境(大模型很多都是依赖Python环境)
4、完整pom
<?xml version="1.0" encoding="UTF-8"?>
<project xmlns="http://maven.apache.org/POM/4.0.0" xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance"
xsi:schemaLocation="http://maven.apache.org/POM/4.0.0 https://maven.apache.org/xsd/maven-4.0.0.xsd">
<modelVersion>4.0.0</modelVersion>
<parent>
<groupId>org.springframework.boot</groupId>
<artifactId>spring-boot-starter-parent</artifactId>
<version>3.4.5</version>
<relativePath/> <!-- lookup parent from repository -->
</parent>
<groupId>com.example</groupId>
<artifactId>spring-ai-demo</artifactId>
<version>0.0.1-SNAPSHOT</version>
<name>spring-ai-demo</name>
<description>Demo project for Spring Boot</description>
<properties>
<java.version>17</java.version>
<fastjson.version>2.0.53</fastjson.version>
</properties>
<dependencies>
<dependency>
<groupId>org.springframework.boot</groupId>
<artifactId>spring-boot-starter-web</artifactId>
</dependency>
<!-- 智谱 -->
<dependency>
<groupId>org.springframework.ai</groupId>
<artifactId>spring-ai-starter-model-zhipuai</artifactId>
</dependency>
<!-- deepseek -->
<!--<dependency>
<groupId>org.springframework.ai</groupId>
<artifactId>spring-ai-starter-model-deepseek</artifactId>
</dependency>-->
<dependency>
<groupId>org.springframework.boot</groupId>
<artifactId>spring-boot-starter-data-redis</artifactId>
</dependency>
<dependency>
<groupId>org.projectlombok</groupId>
<artifactId>lombok</artifactId>
<optional>true</optional>
</dependency>
<dependency>
<groupId>com.alibaba.fastjson2</groupId>
<artifactId>fastjson2</artifactId>
<version>${fastjson.version}</version>
</dependency>
<!--mcp server-->
<dependency>
<groupId>org.springframework.ai</groupId>
<artifactId>spring-ai-starter-mcp-server</artifactId>
</dependency>
<!--即支持sse,也支持stdio-->
<dependency>
<groupId>org.springframework.ai</groupId>
<artifactId>spring-ai-starter-mcp-client-webflux</artifactId>
</dependency>
</dependencies>
<dependencyManagement>
<dependencies>
<dependency>
<groupId>org.springframework.ai</groupId>
<artifactId>spring-ai-bom</artifactId>
<version>1.0.0</version>
<type>pom</type>
<scope>import</scope>
</dependency>
</dependencies>
</dependencyManagement>
<build>
<plugins>
<plugin>
<groupId>org.apache.maven.plugins</groupId>
<artifactId>maven-compiler-plugin</artifactId>
<configuration>
<source>16</source>
<target>16</target>
</configuration>
</plugin>
</plugins>
</build>
</project>
阿里的大模型pom=sping-ai-alibaba-starter-dashscope
5、application.properties文件
server:
servlet:
context-path: /ai
spring:
application:
name: spring-ai-demo
ai:
chat:
client:
# 禁用默认chat client
enabled: false
zhipuai:
# 从环境变量取
api-key: ${API_KEY_ZHIPUAI}
chat:
options:
model: glm-4-plus
temperature: 0.7
data:
redis:
host: localhost
port: 6379
password: 123123!
lettuce:
pool:
min-idle: 0
max-idle: 8
max-active: 8
max-wait: -1ms
6、特性与Demo
6.1、最简单的对话
配置ChatClient
/**
* 默认client
*/
@Bean
public ChatClient zhiPuAiChatClient(ZhiPuAiChatModel chatModel) {
return ChatClient.create(chatModel);
}
定义接口
/**
* 最简单的chat
*
* @author stone
* @date 2025/6/26 16:11
*/
@RestController
@RequestMapping("/case1")
@Slf4j
public class Case1Controller {
@Resource
@Qualifier("zhiPuAiChatClient")
private ChatClient chatClient;
/**
* 直接获取结果
*/
@GetMapping("/chat")
public String chat(@RequestParam("input") String input) {
// input=讲个笑话
return this.chatClient.prompt()
.user(input)
.call()
.content();
}
/**
* 转化实体
*/
@GetMapping("/entity")
public List<ActFilm> entity(@RequestParam("input") String input) {
// input=生成刘德华和刘亦菲的10部电影
return this.chatClient.prompt()
.user(input)
.call()
.entity(new ParameterizedTypeReference<List<ActFilm>>() {
});
}
/**
* 流式响应
*/
@GetMapping(value = "/flux", produces = MediaType.TEXT_EVENT_STREAM_VALUE)
public Flux<String> flux(@RequestParam("input") String input) {
// input=讲个笑话
return this.chatClient.prompt()
.user(input)
.stream()
.content();
}
/**
* 动态输入
*/
@GetMapping(value = "/fluxDynamic", produces = MediaType.TEXT_EVENT_STREAM_VALUE)
public Flux<String> fluxDynamic(@RequestParam("input") String input, @RequestParam("name") String name) {
return this.chatClient.prompt()
.user(promptUserSpec -> promptUserSpec
.text("告诉我中国有多少叫{name}的人")
.param("name", name)
)
.stream()
.content();
}
}
.user,也就是用户提示词
.call,同步方式响应,也就是一整个结果返回
.stream,流式响应,调整为sse方式(text/event-stream)
6.2、默认系统文本
预定义chatClient,设置系统提示词
/**
* 参数-占位符的默认系统文本
*/
@Bean
public ChatClient paramTextChatClient(ZhiPuAiChatModel chatModel) {
return ChatClient.builder(chatModel)
.defaultSystem("你是一个智能聊天机器人,用 {role} 的角度回答问题")
.build();
}
/**
* 默认系统文本
*/
@Bean
public ChatClient defaultTextChatClient(ZhiPuAiChatModel chatModel) {
return ChatClient.builder(chatModel)
.defaultSystem("你是一个智能聊天机器人,用邪恶女巫的角度回答问题")
.build();
}
定义接口
/**
* 默认系统文本
*
* @author stone
* @date 2025/6/30 15:11
*/
@RestController
@RequestMapping("/case2")
@Slf4j
public class Case2Controller {
@Resource
@Qualifier("defaultTextChatClient")
private ChatClient defaultTextChatClient;
/**
* 默认系统文本
*/
@GetMapping(value = "/flux", produces = MediaType.TEXT_EVENT_STREAM_VALUE)
public Flux<String> flux(@RequestParam("input") String input) {
// input=讲个笑话
return this.defaultTextChatClient.prompt()
.user(input)
.stream()
.content();
}
@Resource
@Qualifier("paramTextChatClient")
private ChatClient paramTextChatClient;
/**
* 动态系统文本
*/
@GetMapping(value = "/chat", produces = MediaType.TEXT_EVENT_STREAM_VALUE)
public Flux<String> chat(@RequestParam("input") String input, @RequestParam("role") String role) {
// input=聊一聊圆明园的故事吧,500字以内
// role=数学老师/邪恶女巫
return this.paramTextChatClient.prompt()
.system(promptSystemSpec -> promptSystemSpec.param("role", role))
.user(input)
.stream()
.content();
}
}
消息类型
提示词的不同部分,在交互中扮演着独特和定义明确的角色。
6.3、advisors
提供了强大灵活的拦截式AI交互驱动(配置多个advisor时,前一个做出的更改会传递给下一个)
/**
* @author stone
* @date 2025/6/30 15:41
*/
@RestController
@RequestMapping("/case3")
@Slf4j
public class Case3Controller {
@Resource
@Qualifier("paramTextChatClient")
private ChatClient chatClient;
/**
* 最简单的advisor=日志
* <p>
* org.springframework.ai.chat.client.advisor=debug
*/
@GetMapping(value = "/chat", produces = MediaType.TEXT_EVENT_STREAM_VALUE)
public Flux<String> chat(@RequestParam("input") String input, @RequestParam("role") String role) {
// input=聊一聊圆明园的故事吧,500字以内
// role=邪恶女巫
return this.chatClient.prompt()
.system(promptSystemSpec -> promptSystemSpec.param("role", role))
.advisors(new SimpleLoggerAdvisor())
.user(input)
.stream()
.content();
}
/**
* 自定义打印内容
*/
@GetMapping(value = "/chat2",produces = MediaType.TEXT_EVENT_STREAM_VALUE)
public Flux<String> chat2(@RequestParam("input") String input, @RequestParam("role") String role) {
// input=聊一聊圆明园的故事吧,500字以内
// role=邪恶女巫
return this.chatClient.prompt()
.system(promptSystemSpec -> promptSystemSpec.param("role", role))
.advisors(SimpleLoggerAdvisor.builder()
.requestToString(req -> "请求参数:" + req.prompt().getUserMessage().getText())
.responseToString(resp -> "响应参数:" + resp.getResult().getOutput().getText())
.build())
.user(input)
.stream()
.content();
}
/**
* 定义子类,自定义打印的
*/
@GetMapping(value = "/chat3",produces = MediaType.TEXT_EVENT_STREAM_VALUE)
public Flux<String> chat3(@RequestParam("input") String input, @RequestParam("role") String role) {
// input=聊一聊圆明园的故事吧,500字以内
// role=邪恶女巫
return this.chatClient.prompt()
.system(promptSystemSpec -> promptSystemSpec.param("role", role))
.advisors(new SimpleLogAdvisor())
.user(input)
.stream()
.content();
}
/**
* 自定义日志打印
*/
public static class SimpleLogAdvisor extends SimpleLoggerAdvisor {
public ChatClientResponse adviseCall(ChatClientRequest chatClientRequest, CallAdvisorChain callAdvisorChain) {
log.info("自定义日志,请求参数:{}", chatClientRequest.prompt().getUserMessage().getText());
ChatClientResponse chatClientResponse = super.adviseCall(chatClientRequest, callAdvisorChain);
log.info("自定义日志,响应结果:{}", chatClientResponse.chatResponse().getResult().getOutput().getText());
return chatClientResponse;
}
public Flux<ChatClientResponse> adviseStream(ChatClientRequest chatClientRequest, StreamAdvisorChain streamAdvisorChain) {
log.info("自定义日志,请求参数:{}", chatClientRequest.prompt().getUserMessage().getText());
Flux<ChatClientResponse> chatClientResponses = streamAdvisorChain.nextStream(chatClientRequest);
return (new ChatClientMessageAggregator()).aggregateChatClientResponse(chatClientResponses, this::logResponse);
}
private void logResponse(ChatClientResponse chatClientResponse) {
log.info("自定义日志,响应结果:{}", chatClientResponse.chatResponse().getResult().getOutput().getText());
}
}
}
6.4、对话记忆功能
通过多轮对话,实现聊天内存功能,通过实现交互信息的持久化存储与动态检索机制
public Case4Controller(ZhiPuAiChatModel chatModel,
ChatMemory chatMemory) {
this.chatClient = ChatClient
.builder(chatModel)
.defaultAdvisors(
PromptChatMemoryAdvisor.builder(chatMemory).build()
)
.build();
}
private ChatClient chatClient;
基于JVM内存;设置唯一信息,通过常量区分不同的用户对话
/**
* 对话记忆功能-基于内存
* <p>
* MessageWindowChatMemory:默认最大20
* InMemoryChatMemoryRepository:使用map
*/
// @GetMapping(value = "/chat", produces = MediaType.TEXT_EVENT_STREAM_VALUE)
@GetMapping(value = "/chat")
public Flux<String> chat(@RequestParam("input") String input) {
// input=我叫什么
return this.chatClient.prompt()
.user(input)
.stream()
.content();
}
/**
* 对话记忆功能-区分不同用户
*/
@GetMapping(value = "/chat2")
public Flux<String> chat2(@RequestParam("input") String input,
@RequestParam("userId") String userId) {
return this.chatClient.prompt()
.user(input)
.advisors(a -> a.param(ChatMemory.CONVERSATION_ID, userId))
.stream()
.content();
}
基于Redis的多轮对话记忆功能
/**
* @author stone
* @date 2025/7/14 10:20
*/
@Data
public class ChatBO implements Serializable {
/**
* 用户对话唯一标识
*/
private String chatId;
/**
* 对话类型
*/
private String type;
/**
* 对话内容
*/
private String text;
}
/**
* @author stone
* @date 2025/7/14 10:19
*/
@Slf4j
@Component
public class ChatRedisMemory implements ChatMemory {
private static final String KEY_PREFIX = "chat:history:";
private final RedisTemplate<String, Object> redisTemplate;
public ChatRedisMemory(RedisTemplate<String, Object> redisTemplate) {
this.redisTemplate = redisTemplate;
}
@Override
public void add(String conversationId, List<Message> messages) {
String key = KEY_PREFIX + conversationId;
List<String> list = new ArrayList<>();
for (Message msg : messages) {
String[] strs = msg.getText().split("</think>");
String text = strs.length == 2 ? strs[1] : strs[0];
// 转化
ChatBO bo = new ChatBO();
bo.setChatId(conversationId);
bo.setType(msg.getMessageType().getValue());
bo.setText(text);
list.add(JSON.toJSONString(bo));
}
redisTemplate.opsForList().rightPushAll(key, list.toArray());
redisTemplate.expire(key, 30, TimeUnit.MINUTES);
}
@Override
public List<Message> get(String conversationId) {
String key = KEY_PREFIX + conversationId;
Long size = redisTemplate.opsForList().size(key);
if (size == null || size == 0) {
return Collections.emptyList();
}
List<Object> listTmp = redisTemplate.opsForList().range(key, 0, -1);
List<Message> result = new ArrayList<>();
for (Object obj : listTmp) {
ChatBO chat = JSON.parseObject(obj.toString(), ChatBO.class);
if (MessageType.USER.getValue().equals(chat.getType())) {
result.add(new UserMessage(chat.getText()));
} else if (MessageType.ASSISTANT.getValue().equals(chat.getType())) {
result.add(new AssistantMessage(chat.getText()));
} else if (MessageType.SYSTEM.getValue().equals(chat.getType())) {
result.add(new SystemMessage(chat.getText()));
}
}
return result;
}
@Override
public void clear(String conversationId) {
redisTemplate.delete(KEY_PREFIX + conversationId);
}
}
/**
* 基于redis的对话记忆
*/
@Bean
public ChatMemory chatMemory(RedisTemplate<String, Object> redisTemplate) {
return new ChatRedisMemory(redisTemplate);
}
public Case4Controller(ZhiPuAiChatModel chatModel,
ChatMemory chatMemory) {
this.chatClient = ChatClient
.builder(chatModel)
.defaultAdvisors(
PromptChatMemoryAdvisor.builder(chatMemory).build()
)
.build();
}
/**
* 对话记忆功能-基于redis
*/
@GetMapping(value = "/chat3")
public Flux<String> chat3(@RequestParam("input") String input,
@RequestParam("userId") String userId) {
return this.chatClient.prompt()
.user(input)
.advisors(a -> a.param(ChatMemory.CONVERSATION_ID, userId))
.stream()
.content();
}
多层记忆机构,模仿人类,做到近期(清晰),中期(模糊),长期(关键点)。这里就引入了向量数据库和RAG。
6.5、@Tools使用
声明式Function Calling,将方法转化为工具。提前告诉大模型,提供了什么tools。太多的tools,可以放到向量数据库。
第三方提供的tools,比如百度天气、高德位置,不可能各对接系统去做解析。因此MCP(model content protol)协议,通过JSON-rpc2.0方式(json数据格式),统一格式解析。
/**
* tools工具
*
* @author stone
* @date 2025/7/4 10:42
*/
@Component
@Slf4j
public class OrderTools {
/**
* 比如在退订、取消订单
*/
@Tool(description = "退订、取消订单")
public String cancelOrder(@ToolParam(description = "订单号") String orderNum,
@ToolParam(description = "账号") String userAccount) {
log.info("订单号:{},用户账号:{}", orderNum, userAccount);
// 执行业务逻辑
log.info("处理数据库...");
return "操作成功";
}
}
/**
* @author stone
* @date 2025/7/4 10:40
*/
@RestController
@RequestMapping("/case5")
@Slf4j
public class Case5Controller {
public Case5Controller(ZhiPuAiChatModel chatModel,
ChatMemory chatMemory,
OrderTools orderTools) {
this.chatClient = ChatClient
.builder(chatModel)
.defaultAdvisors(
PromptChatMemoryAdvisor.builder(chatMemory).build()
)
.defaultTools(orderTools)
.build();
}
private ChatClient chatClient;
/**
* tools使用
*/
@GetMapping("/chat")
public String chat(@RequestParam("input") String input) {
// input=我要退订
// input=账号是101,订单号是XXX1111
return this.chatClient.prompt()
// 直接方法使用
// .tools()
.user(input)
.call()
.content();
}
}
6.6、调用外部MCP-server
TODO...
6.7、向量数据库与RAG(检索增强生成)
TODO...