Begin Chat Memory

backend/src/main/java/ai/dragon/service/ChatMessageService.java

@@ -0,0 +1,62 @@
+package ai.dragon.service;
+
+import java.util.ArrayList;
+import java.util.List;
+import java.util.Map;
+import java.util.Optional;
+
+import org.slf4j.Logger;
+import org.slf4j.LoggerFactory;
+import org.springframework.stereotype.Service;
+
+import ai.dragon.dto.openai.completion.OpenAiCompletionMessage;
+import dev.langchain4j.data.message.AiMessage;
+import dev.langchain4j.data.message.ChatMessage;
+import dev.langchain4j.data.message.Content;
+import dev.langchain4j.data.message.ImageContent;
+import dev.langchain4j.data.message.SystemMessage;
+import dev.langchain4j.data.message.TextContent;
+import dev.langchain4j.data.message.UserMessage;
+
+@Service
+public class ChatMessageService {
+    private final Logger logger = LoggerFactory.getLogger(this.getClass());
+
+    public Optional<ChatMessage> convertToChatMessage(OpenAiCompletionMessage completionMessage) {
+        ChatMessage chatMessage;
+        switch (completionMessage.getRole()) {
+            case "user":
+                if (completionMessage.getContent() instanceof String) {
+                    chatMessage = new UserMessage(completionMessage.getName(), (String) completionMessage.getContent());
+                } else {
+                    List<Map<String, Object>> content = (List<Map<String, Object>>) completionMessage.getContent();
+                    List<Content> contents = new ArrayList<>();
+                    content.forEach(contentItem -> {
+                        if (!contentItem.containsKey("type")) {
+                            logger.error("Content part must have a type field!");
+                            return;
+                        }
+                        String type = (String) contentItem.get("type");
+                        if ("text".equals(type)) {
+                            String text = (String) contentItem.get("text");
+                            contents.add(new TextContent(text));
+                        } else if ("image_url".equals(type)) {
+                            Map<String, Object> imageURL = (Map<String, Object>) contentItem.get("image_url");
+                            String url = (String) imageURL.get("url");
+                            // TODO String detail = (String) imageURL.get("detail");
+                            contents.add(new ImageContent(url));
+                        }
+                    });
+                    chatMessage = new UserMessage(completionMessage.getName(), contents);
+                }
+                break;
+            case "system":
+                chatMessage = new SystemMessage((String) completionMessage.getContent());
+            case "assistant":
+                chatMessage = new AiMessage((String) completionMessage.getContent());
+            default:
+                throw new IllegalArgumentException("Invalid Message Role: " + completionMessage.getRole());
+        }
+        return Optional.ofNullable(chatMessage);
+    }
+}

backend/src/main/java/ai/dragon/service/RaagService.java

@@ -21,6 +21,7 @@
 import ai.dragon.repository.FarmRepository;
 import ai.dragon.util.ai.AiAssistant;
 import dev.langchain4j.data.segment.TextSegment;
+import dev.langchain4j.memory.chat.MessageWindowChatMemory;
 import dev.langchain4j.model.chat.StreamingChatLanguageModel;
 import dev.langchain4j.model.embedding.EmbeddingModel;
 import dev.langchain4j.rag.DefaultRetrievalAugmentor;
@@ -51,6 +52,9 @@ public class RaagService {
     @Autowired
     private KVSettingService kvSettingService;
 
+    @Autowired
+    private ChatMessageService chatMessageService;
+
     public List<OpenAiModel> listAvailableModels() {
         return farmRepository
                 .find()
@@ -71,8 +75,8 @@ public SseEmitter chatResponse(FarmEntity farm, OpenAiChatCompletionRequest requ
         AiAssistant assistant = AiServices.builder(AiAssistant.class)
                 .streamingChatLanguageModel(this.buildStreamingChatLanguageModel(farm))
                 // TODO support of chatLanguageModel in addition of streamingChatLanguageModel
-                // TODO .chatMemory(MessageWindowChatMemory.withMaxMessages(10)) // it should
                 .retrievalAugmentor(this.buildRetrievalAugmentor(farm))
+                .chatMemory(this.buildChatMemory(request))
                 .build();
         // TODO request.getMessages().get(0).getContent())
         TokenStream stream = assistant.chat((String) request.getMessages().get(0).getContent());
@@ -93,6 +97,16 @@ public SseEmitter chatResponse(FarmEntity farm, OpenAiChatCompletionRequest requ
         return sseService.retrieveEmitter(emitterID);
     }
 
+    private MessageWindowChatMemory buildChatMemory(OpenAiChatCompletionRequest request) {
+        MessageWindowChatMemory memory = MessageWindowChatMemory.withMaxMessages(10); // TODO maxMessages
+        // Don't retrieve the last message, it's not for history but for completion!
+        for (int i = 0; i < request.getMessages().size() - 1; i++) {
+            OpenAiCompletionMessage requestMessage = request.getMessages().get(i);
+            chatMessageService.convertToChatMessage(requestMessage).ifPresent(memory::add);
+        }
+        return memory;
+    }
+
     private OpenAiChatCompletionResponse createChatCompletionResponse(
             UUID emitterID,
             OpenAiChatCompletionRequest request,