本文基于Langchain4j、Milvus、Redis实现AI智能体。Langchain4j作为会话大脑，Redis实现会话记忆(Memory),Milvus构建RAG知识库。

引入依赖

<!--LangChain4j 针对 open ai 的 Spring Boot starter-->
<dependency>
	<groupId>dev.langchain4j</groupId>
	<artifactId>langchain4j-open-ai-spring-boot-starter</artifactId>
	<version>1.10.0-beta18</version>
</dependency>
<!-- AiService的相关依赖 -->
<dependency>
	<groupId>dev.langchain4j</groupId>
	<artifactId>langchain4j-spring-boot-starter</artifactId>
	<version>1.10.0-beta18</version>
</dependency>

<!--LangChain4j 针对 ollama 的 Spring Boot starter-->
<dependency>
    <groupId>dev.langchain4j</groupId>
    <artifactId>langchain4j-ollama-spring-boot-starter</artifactId>
    <version>1.10.0-beta18</version>
</dependency>

<!-- LangChain4j 流式调用依赖 -->
<dependency>
    <groupId>dev.langchain4j</groupId>
    <artifactId>langchain4j-reactor</artifactId>
    <version>1.10.0-beta18</version>
</dependency>

<!-- LangChain4j 向量数据库milvus 依赖 -->
<dependency>
    <groupId>dev.langchain4j</groupId>
    <artifactId>langchain4j-milvus</artifactId>
    <version>1.10.0-beta18</version>
</dependency>
<dependency>
    <groupId>dev.langchain4j</groupId>
    <artifactId>langchain4j-embeddings-all-minilm-l6-v2</artifactId>
    <version>1.10.0-beta18</version>
</dependency>
<!-- 文档解析器 -->
<dependency>
    <groupId>dev.langchain4j</groupId>
    <artifactId>langchain4j-document-parser-apache-pdfbox</artifactId>
    <version>1.10.0-beta18</version>
</dependency>
<dependency>
    <groupId>dev.langchain4j</groupId>
    <artifactId>langchain4j-document-parser-apache-poi</artifactId>
    <version>1.10.0-beta18</version>
</dependency>

<!-- easy rag 知识库 -->
<dependency>
    <groupId>dev.langchain4j</groupId>
    <artifactId>langchain4j-easy-rag</artifactId>
    <version>1.10.0-beta18</version>
</dependency>

<!-- 流式调用 -->
<dependency>
	<groupId>org.springframework.boot</groupId>
    <artifactId>spring-boot-starter-webflux</artifactId>
</dependency>

智能体配置

配置使用Milvus作为向量数据库实现rag知识检索，此处仅实例，实际使用过程中加载知识库的部分需要独立出来根据需要更新即可。

@Configuration
public class AgentsConfig {

	@Autowired
	private RedisChatMemoryStore redisChatMemoryStore;

	// Milvus配置
	@Value("${milvus.host}")
	private String milvusHost;

	@Value("${milvus.port}")
	private Integer milvusPort;

	@Value("${milvus.database-name}")
	private String milvusDatabase;

	@Value("${milvus.collection-name:knowledge_base}")
	private String milvusCollectionName;

	@Value("${milvus.dimension:384}")
	private Integer milvusDimension;

	/**
	 * 构建chatMemoryProvider
	 * 
	 * @return
	 */
	@Bean
	public ChatMemoryProvider chatMemoryProvider() {
		ChatMemoryProvider chatMemoryProvider = new ChatMemoryProvider() {

			@Override
			public ChatMemory get(Object memoryId) {
				ChatMemory m = MessageWindowChatMemory.builder().id(memoryId).maxMessages(30)
						.chatMemoryStore(redisChatMemoryStore).build();
				return m;
			}
		};
		return chatMemoryProvider;
	}

	/**
	 * MiniLM 嵌入模型
	 */
	@Bean
	public EmbeddingModel embeddingModel() {
		return new AllMiniLmL6V2EmbeddingModel();
	}

	/**
	 * Milvus向量数据库配置
	 */
	@Bean
	public EmbeddingStore<TextSegment> milvusEmbeddingStore() {
		return MilvusEmbeddingStore.builder()
				.host(milvusHost)
				.port(milvusPort)
				.databaseName(milvusDatabase)
				.collectionName(milvusCollectionName)
				.dimension(milvusDimension)
				.build();
	}

	/**
	 * 文档注入器
	 */
	@Bean
	public EmbeddingStoreIngestor embeddingStoreIngestor(EmbeddingModel embeddingModel,
			EmbeddingStore<TextSegment> embeddingStore) {

		EmbeddingStoreIngestor ingestor = EmbeddingStoreIngestor.builder()
				.documentSplitter(DocumentSplitters.recursive(500, 50))
				.embeddingModel(embeddingModel)
				.embeddingStore(embeddingStore)
				.build();
		// 1、加载知识库
		List<Document> documents = ClassPathDocumentLoader.loadDocuments("content");
		ingestor.ingest(documents);

		return ingestor;
	}

	/**
	 * 构建向量数据库检索对象
	 * 
	 * @return
	 */
	@Bean
	public ContentRetriever contentRetriever(EmbeddingModel embeddingModel,
			EmbeddingStore<TextSegment> embeddingStore) {

		return EmbeddingStoreContentRetriever.builder()
				.embeddingStore(embeddingStore)
				.embeddingModel(embeddingModel)
				.minScore(0.5)
				.maxResults(3)
				.build();
	}
}

会话记忆

使用Redis做对话记忆。

@Component
public class RedisChatMemoryStore implements ChatMemoryStore{

	@Override
	public List<ChatMessage> getMessages(Object memoryId) {
		String json = RedisUtils.getCacheObject(memoryId.toString());
		return ChatMessageDeserializer.messagesFromJson(json);
	}

	@Override
	public void updateMessages(Object memoryId, List<ChatMessage> messages) {
		RedisUtils.setCacheObject(memoryId.toString(), ChatMessageSerializer.messagesToJson(messages), Duration.ofDays(7));
	}

	@Override
	public void deleteMessages(Object memoryId) {
		RedisUtils.deleteObject(memoryId.toString());
	}

}

工具调用

智能体本地工具调用，此处是举例，实际场景根据业务调整即可。这里演示了一个根据商品名称查询库存的案例。

@Component
public class RetailTool {
	
	@DubboReference
	private IChainStockApiService stockApiService;
	
	@Tool("根据门商品名称查询库存")
	public List<ErpChainStock> queryChainByPhone(@P("商品名称") String productName) {
//		LoginUser user =  LoginHelper.getLoginUser();
//		Long chainId = user.getDeptId();
//		Map<String,String> map = new HashMap<>();
//		map.put("productNameLike", productName);
//		PagerInfo pager = new PagerInfo(10, 1);
//		ResultPage<ErpChainStock> resultPage = stockApiService.query(chainId, map, pager);
//		return resultPage.getList();
		return TenantHelper.dynamic(TenantConstants.DEFAULT_TENANT_ID, ()->{
			Long chainId = 1863398543189741570l;
			Map<String,String> map = new HashMap<>();
			map.put("productNameLike", productName);
			PagerInfo pager = new PagerInfo(10, 1);
			ResultPage<ErpChainStock> resultPage = stockApiService.query(chainId, map, pager);
			return resultPage.getList();
		});
		
	}

}

AI Service

使用@AiService注解，注入会话模型（此处使用ollama本地模型），注入本地工具tools、会话记忆提供者chatMemoryProvider、RAG知识库contentRetriever。

@AiService(
		wiringMode = AiServiceWiringMode.EXPLICIT,
		//chatModel = "ollamaChatModel",
		streamingChatModel = "ollamaStreamingChatModel",
		//streamingChatModel = "openAiStreamingChatModel",
		//chatMemory = "chatMemory", //配置会话记忆对象
		tools = "retailTool" ,
		chatMemoryProvider = "chatMemoryProvider",
		contentRetriever = "contentRetriever"
)
public interface IChartAiService {
	
	/**
	 * 和AI聊天
	 * @param message
	 * @return
	 */
	//@SystemMessage("你是黔丰盈通门店管理助手爱钱钱。请以友好、乐于助人且愉快的方式解答用户关于门店零售的相关问题，回答问题请用中文。")
	@SystemMessage(fromResource = "system-message.txt")
	Flux<String> chat(@MemoryId Long memoryId, @UserMessage String message);

}

HTTP对外接口

此处只是举例，getMemoryId需要根据实际业务场景生成。使用流式对话返回Flux<String>。

@RestController("/ai")
public class AiChatController {
	
	@Autowired
	private IChartAiService  chartAiService;
	
	@GetMapping(value = "/getMemoryId")
	public R<Long> getMemoryId() {
		return R.ok(System.currentTimeMillis()) ;
	}

	@GetMapping(value = "/chat", produces = "text/html;charset=utf-8")
    public Flux<String> model(Long memoryId, String message) {
		Flux<String> r = chartAiService.chat(memoryId, message);
        return r;
    }

}

配置

配置文件如下

langchain4j: 
  #open-ai:
  ollama:  
    streaming-chat-model: 
      #ollama http://192.168.2.131:11434 阿里百炼 https://dashscope.aliyuncs.com/compatible-mode/v1
      base-url: http://192.168.2.131:11434
      #api-key: sk-xxxxxxxxx697
      #ollama deepseek-r1-1.5b, qwen3:0.6b 阿里百炼 qwen3-max
      model-name: qwen3:0.6b
      log-requests: true
      log-responses: true

 # 配置 Milvus 向量数据库
milvus:
  host: localhost
  port: 19530
  # 如果 Milvus 配置了用户名密码，请取消注释并填写
  username: root
  # password: Milvus
  database-name: test