Spring AI + LLM

dependencies {
    implementation("org.springframework.ai:spring-ai-starter-model-openai-spring-boot-starter:1.0.0-M6")
}

⚠️ Spring AI ещё в milestone — версия может отличаться.

spring:
  ai:
    openai:
      api-key: ${OPENAI_API_KEY}
      chat:
        options:
          model: gpt-4o-mini     # дешевле, чем gpt-4
          temperature: 0.7

taskflow:
  ai:
    enabled: true

⚠️ OPENAI_API_KEY — переменная окружения. НЕ коммитить в Git!

package com.taskflow.service;

import org.springframework.ai.chat.client.ChatClient;
import org.springframework.ai.chat.prompt.Prompt;
import org.springframework.ai.chat.prompt.SystemPromptTemplate;
import org.springframework.ai.openai.OpenAiChatModel;
import org.springframework.beans.factory.annotation.Value;
import org.springframework.stereotype.Service;

import java.util.List;
import java.util.Map;

@Service
public class AiService {
    private final ChatClient chatClient;
    private final boolean enabled;

    public AiService(OpenAiChatModel chatModel,
                     @Value("${taskflow.ai.enabled:true}") boolean enabled) {
        this.chatClient = ChatClient.create(chatModel);
        this.enabled = enabled;
    }

    public String chat(String userMessage) {
        if (!enabled) return "AI is disabled";

        // System prompt — задаёт роль
        String systemPrompt = """
            You are a helpful assistant for TaskFlow, a task management service.
            You help users create, organize, and prioritize their tasks.
            Be concise. Use bullet points for lists.
            """;

        return chatClient.prompt()
            .system(systemPrompt)
            .user(userMessage)
            .call()
            .content();
    }

    // Streaming response
    public Flux streamChat(String userMessage) {
        return chatClient.prompt()
            .user(userMessage)
            .stream()
            .content();
    }
}

package com.taskflow.controller;

import com.taskflow.dto.ChatRequest;
import com.taskflow.dto.ChatResponse;
import com.taskflow.service.AiService;
import jakarta.validation.Valid;
import org.springframework.http.MediaType;
import org.springframework.web.bind.annotation.*;
import reactor.core.publisher.Flux;

@RestController
@RequestMapping("/api/chat")
public class AiController {
    private final AiService aiService;

    public AiController(AiService aiService) { this.aiService = aiService; }

    @PostMapping
    public ChatResponse chat(@Valid @RequestBody ChatRequest req) {
        return new ChatResponse(aiService.chat(req.message()));
    }

    @PostMapping(value = "/stream", produces = MediaType.TEXT_EVENT_STREAM_VALUE)
    public Flux stream(@Valid @RequestBody ChatRequest req) {
        return aiService.streamChat(req.message());
    }
}

public record ChatRequest(@NotBlank String message) {}
public record ChatResponse(String reply) {}

@Bean
@Description("Get task by ID")
public Function getTaskFunction(TaskService taskService) {
    return request -> {
        return taskService.findById(request.id())
            .orElseThrow(() -> new RuntimeException("Task not found: " + request.id()));
    };
}

public record GetTaskRequest(@JsonProperty(required = true) Long id) {}

В AiService:

public String chatWithFunctions(String userMessage) {
    return chatClient.prompt()
        .user(userMessage)
        .functions("getTaskFunction")  // Spring AI найдёт бин
        .call()
        .content();
}

Теперь LLM может вызвать наш getTaskFunction для ответа на вопрос «какая задача #42?»

@Service
public class EmbeddingService {
    private final EmbeddingClient embeddingClient;

    public EmbeddingService(EmbeddingClient embeddingClient) {
        this.embeddingClient = embeddingClient;
    }

    public float[] embed(String text) {
        return embeddingClient.embed(text);
    }
}

Хранить embeddings в pgvector (расширение PostgreSQL) или специализированной БД (Pinecone, Weaviate, Milvus).

public String ragChat(String userQuestion) {
    // 1. Embed user question
    float[] questionVec = embeddingClient.embed(userQuestion);

    // 2. Find similar tasks in DB (cosine similarity)
    List similarTasks = taskRepository.findSimilar(questionVec, 5);

    // 3. Build context from similar tasks
    String context = similarTasks.stream()
        .map(t -> "- " + t.getTitle() + ": " + t.getDescription())
        .collect(Collectors.joining("\n"));

    // 4. Ask LLM with context
    String prompt = """
        Based on these tasks:
        %s

        Answer the question: %s
        """.formatted(context, userQuestion);

    return chatClient.prompt()
        .user(prompt)
        .call()
        .content();
}

# Нужен OPENAI_API_KEY
export OPENAI_API_KEY=sk-...

$ curl -X POST http://localhost:8080/api/chat \
  -H "Authorization: Bearer $TOKEN" \
  -H "Content-Type: application/json" \
  -d '{"message":"What are good practices for task management?"}'

{
  "reply": "Here are 5 best practices for task management:\n1. Set clear priorities..."
}

---