The 'Prompt'` class holds an ordered list of `Message` objects – System & User roles.
Prompt prompt = new Prompt(List.of(
new SystemMessage,new UserMessage));ChatResponse response = chatModel.call;String answer = response.content;// Prompt类还提供 getUserMessage / getSystemMessage 等便捷方法。
RAG 检索提高
java
@Configuration public
class RagConfig {
@Bean
VectorStore vectorStore(JdbcTemplate jdbcTemplate,EmbeddingModel embeddingModel){
return new PgVectorStore;}
}
@RestController
public class RagController{
private final ChatClient chat;public RagController(ChatClient.Builder builder,VectorStore store){
this.chat = builder.defaultAdvisors(
QuestionAnswerAdvisor.builder.build)
.build;}
@GetMapping
public String rag{
return chat.prompt.user.call.content;}
}
`QuestionAnswerAdvisor` is an interceptor that automatically retrieves relevant chunks from vector store and appends m to Prompt before each LLM call – **zero‑coding RAG**.
对话记忆
java
@Configuration public
class MemoryConfig{
@Bean public
ChatMemoryRepository repo{return new InMemoryChatMemoryRepository;}
@Bean public
ChatMemory memory{
return MessageWindowChatMemory.builder
.chatMemoryRepository
.maxMessages
.build;}
}
@RestController
public class ChatCtrl{
private final ChatClient client;public ChatCtrl{
this.client=b.defaultAdvisors(
MessageChatMemoryAdvisor.builder.build)
.build;}
@PostMapping
public String chat(@RequestParam String session。@RequestParam String msg){
return client.prompt
.advisors(a->a.param(
AbstractChatMemoryAdvisor.CHAT_MEMORY_CONVERSATION_ID_KEY,session))
.user.call.content;}
}
**优势**这方面，无需自行维护会话状态。Spring AI 为每个会话提供独立内存，实现「上下文感知」。
Spring AI 的真正优势
. 零心智负担的 Spring 集成：If your project already runs on Spring Boot,adding Spring AI costs almost nothing – auto‑configuration。health checks and observability come out of box.
. 优秀的 VectorStore 抽象层：You can swap PGVector ↔ Milvus ↔ Qdrant etc. by changing a single Maven dependency and a few YAML entries.
. Structured Outputs：You can map model responses directly to POJOs – ideal for returning JSON or domain objects.
. MCP 支持 ：The Model Context Protocol gives you fine‑grained control over tool calls & external data sources. Spring AI team maintains Java SDK,ensuring long‑term stability.
说明：This section focuses on two most active frameworks for production use. Semantic Kernel is omitted because it is still RC and lacks domestic model support..
建立同一个 RAG 应用的代码量对比.
:
Spring AI 实现
java
@Service
public class KnowledgeService {
private final ChatClient client;private final VectorStore vectorStore;public KnowledgeService(ChatClient.Builder builder,VectorStore vectorStore。InMemoryChatMemoryRepository repo){
// 对话记忆 + RAG Advisor 自动注入检索结果
var memory= MessageWindowChatMemory.builder
.chatMemoryRepository.maxMessages.build;话说回来，this.client=builder.defaultAdvisors(
QuestionAnswerAdvisor.builder.build。MessageChatMemoryAdvisor.builder.build)
.
build;this.vectorStore=vectorStore;}
/** PDF 文档导入 */
public void ingestPdf{
var docs=new PagePdfDocumentReader.get;var splitter=new TokenTextSplitter;话说回来，var segments=splitter.apply;不过，vectorStore.add;// 向量化并存储
