Spring AI Free code examples in Java

The starter auto-configures a ChatClient.Builder, so there is no client to construct. The API key goes in configuration like any other Spring property.

# Gradle
implementation("org.springframework.ai:spring-ai-openai-spring-boot-starter:2.0.1")

# application.properties
#   spring.ai.openai.api-key=${OPENAI_API_KEY}
#   spring.ai.openai.chat.options.model=gpt-5.4-mini

export OPENAI_API_KEY="sk-your-key"

ChatClient.Builder is injected, not built. That is the Spring idiom and it is what lets configuration, observability and retries be set centrally rather than at each call site.

import org.springframework.ai.chat.client.ChatClient;
import org.springframework.web.bind.annotation.GetMapping;
import org.springframework.web.bind.annotation.RestController;

@RestController
class AskController {

    private final ChatClient chatClient;

    AskController(ChatClient.Builder chatClientBuilder) {
        this.chatClient = chatClientBuilder.build();
    }

    @GetMapping("/ai")
    String generation(String userInput) {
        return this.chatClient.prompt()
            .user(userInput)
            .call()
            .content();
    }
}

A @Bean ChatClient with defaultSystem and defaultAdvisors means every call site inherits the system prompt and the chain. Nothing downstream repeats it, and changing it is one edit.

import org.springframework.ai.chat.memory.ChatMemory;
import org.springframework.ai.chat.memory.MessageWindowChatMemory;
import org.springframework.ai.chat.client.advisor.MessageChatMemoryAdvisor;
import org.springframework.context.annotation.Bean;
import org.springframework.context.annotation.Configuration;

@Configuration
class ChatConfig {

    @Bean
    ChatMemory chatMemory() {
        return MessageWindowChatMemory.builder().maxMessages(20).build();
    }

    @Bean
    ChatClient chatClient(ChatClient.Builder builder, ChatMemory chatMemory) {
        return builder
            .defaultSystem("You are a concise technical assistant.")
            .defaultAdvisors(
                MessageChatMemoryAdvisor.builder(chatMemory).build())
            .build();
    }
}

TWO ADVISORS, AND THE ORDER IS LOAD-BEARING. Memory first, so the retrieval advisor searches with the conversation's context rather than a pronoun. The conversation id is not optional — without it every user shares one history.

import org.springframework.ai.chat.client.advisor.vectorstore.QuestionAnswerAdvisor;
import org.springframework.ai.vectorstore.VectorStore;

String answer = chatClient.prompt()
    .advisors(advisor -> advisor
        .advisors(
            // 1. Rewrites the prompt with the conversation so far.
            MessageChatMemoryAdvisor.builder(chatMemory).build(),
            // 2. Then retrieves against that, not against "it".
            QuestionAnswerAdvisor.builder(vectorStore).build())
        // REQUIRED. Omit it and every user shares one conversation.
        .param(ChatMemory.CONVERSATION_ID, conversationId))
    .user(userText)
    .call()
    .content();

entity() maps the answer onto a record and retries the parse for you. stream() returns a Flux, which is the shape the rest of a reactive Spring application already speaks.

import reactor.core.publisher.Flux;

record StockAnswer(String productCode, int quantity, boolean reorderAdvised) {}

// Spring AI builds the schema from the record and validates the reply.
StockAnswer parsed = chatClient.prompt()
    .user("How many X-17 are in stock and should we reorder?")
    .call()
    .entity(StockAnswer.class);

// Streaming, for a UI that should not wait for the last token.
Flux<String> tokens = chatClient.prompt()
    .user("Explain B-trees in 200 words.")
    .stream()
    .content();