Spring AI ChatClient

The flow

1. 1.Validate topic
2. 2.Apply template
3. 3.Call or stream
4. 4.Inspect result
5. 5.Verify

Use the configured builder

A Boot application can inject the auto-configured ChatClient.Builder so registered customization and observation are retained. A plain unit test supplies ChatClient.builder(fakeModel). Keeping the service constructor explicit makes those two compositions visible.

The standalone 1.0.3 pilot remains in examples/spring-ai-chat-client as a historical comparison. This lesson now uses the Phase 2 laboratory: Java 21, Boot 4.0.5 and Spring AI 2.0.2. Do not merge the two Maven configurations.

Templates and response metadata

The example keeps system instructions separate and inserts a validated topic into a user-message template. Parameters avoid hand-built template concatenation; they do not make untrusted text safe or replace authorization.

chatResponse() preserves model response metadata alongside text. Provider support determines which usage and timing fields are available. Record actual usage when supplied; do not infer billable tokens from character count or assume every provider returns the same metadata.

Streaming changes completion handling

stream().content() exposes text chunks; call().entity() needs a complete response for typed conversion. A production SSE endpoint must handle cancellation, disconnects, backpressure and errors after headers are sent. The example returns Flux and does not pretend to implement the full HTTP streaming layer.

Tests assert the actual rendered prompt, response model metadata and ordered chunks from a synthetic streaming model. These are integration contracts, not live model quality or latency benchmarks.

Try it

package com.interviewpreparator.lab;
import org.springframework.ai.chat.client.ChatClient;
import org.springframework.ai.chat.model.ChatResponse;
import reactor.core.publisher.Flux;
public final class ChatClientExample {
    private final ChatClient client;
    public ChatClientExample(ChatClient.Builder builder) {
        this.client = builder.defaultSystem("Explain Java briefly. State uncertainty when evidence is missing.").build();
    }
    public ChatResponse explain(String topic) {
        validate(topic);
        return client.prompt().user(user -> user.text("Explain {topic}").param("topic", topic))
                .call().chatResponse();
    }
    public Flux<String> stream(String topic) {
        validate(topic);
        return client.prompt().user(user -> user.text("Explain {topic}").param("topic", topic))
                .stream().content();
    }
    private static void validate(String topic) {
        if (topic == null || topic.isBlank() || topic.length() > 200) {
            throw new IllegalArgumentException("Topic must contain 1–200 characters");
        }
    }
}

examples/spring-ai-lab/src/main/java/com/interviewpreparator/lab/ChatClientExample.java · mvn -f examples/spring-ai-lab/pom.xml test. Java 21+, Spring Boot 4.0.5, Spring AI 2.0.2 where applicable. The demo uses synthetic models; live inference quality is a separate check.

Interview questions

Official references

• Spring AI — ChatClient

Continue learning

• Spring AI: Getting Started →
• Structured Output in Spring AI →
• Advisors & Conversation Memory →
• Testing & Evaluation for Spring AI →