Day 7 of Spring AI Series: How Advisors Form a Filter Chain for AI Applications

Advisor definition

Advisor is an interceptor on the large‑model call chain that can insert logic before a request is sent to the model and after the response is returned.

Built‑in Advisors

MessageChatMemoryAdvisor

– injects conversation history before the model call and saves memory after the call (Day 6). QuestionAnswerAdvisor – performs retrieval‑augmented generation by fetching relevant documents and adding them to the prompt before the call (Day 10). SimpleLoggerAdvisor – logs request and response. SafeGuardAdvisor – filters sensitive words and enforces content safety.

Call chain flow

请求 ──► [Logger] ──► [Memory] ──► [RAG] ──► 大模型
                              │
响应 ◄── [Logger] ◄── [Memory] ◄── [RAG] ◄────┘

Each Advisor receives the request, may process it, then calls chain.nextCall(request) to forward to the next element. After the downstream response returns, the Advisor can further process the response.

Example: SimpleLoggerAdvisor

public class SimpleLoggerAdvisor implements CallAdvisor, StreamAdvisor {
    private static final Logger logger = LoggerFactory.getLogger(SimpleLoggerAdvisor.class);

    @Override
    public String getName() {
        return this.getClass().getSimpleName();
    }

    @Override
    public int getOrder() {
        return 0; // smaller numbers execute earlier
    }

    @Override
    public ChatClientResponse adviseCall(ChatClientRequest request, CallAdvisorChain chain) {
        logger.info("请求: {}", request); // before model call
        ChatClientResponse response = chain.nextCall(request);
        logger.info("响应: {}", response); // after model call
        return response;
    }

    @Override
    public Flux<ChatClientResponse> adviseStream(ChatClientRequest request, StreamAdvisorChain chain) {
        logger.info("请求: {}", request);
        return chain.nextStream(request);
    }
}

Register the advisor with a ChatClient builder:

ChatClient chatClient = ChatClient.builder(chatModel)
        .defaultAdvisors(new SimpleLoggerAdvisor())
        .build();

Ordering

The getOrder() method determines execution precedence; a smaller return value runs earlier. This controls whether memory runs before RAG, whether logging wraps the whole chain, and similar ordering decisions.

Key insight

Complex applications compose functionality by attaching Advisors to the chain rather than using conditional logic, keeping memory, retrieval, logging, and safety concerns decoupled.

Reference URLs

Advisors API: https://docs.spring.io/spring-ai/reference/api/advisors.html

ChatClient API: https://docs.spring.io/spring-ai/reference/api/chatclient.html