仁智企后端
blame | 历史 | 补丁 | 提交 | 提交对比 | ignore whitespace
admin
2025-05-29 ef4c38330371547b66bc0c5b7ebc02d13c81cb2c 
 aiflowy-modules/aiflowy-module-ai/src/main/java/tech/aiflowy/ai/controller/AiBotController.java


@@ -2,6 +2,8 @@




import cn.dev33.satoken.annotation.SaIgnore;


import cn.hutool.core.util.ObjectUtil;


import cn.hutool.json.JSONArray;


import cn.hutool.json.JSONObject;


import com.agentsflex.core.llm.ChatContext;


import com.agentsflex.core.llm.Llm;


import com.agentsflex.core.llm.StreamResponseListener;


@@ -16,24 +18,31 @@


import com.agentsflex.core.util.CollectionUtil;


import com.alibaba.fastjson.JSON;


import com.alibaba.fastjson.serializer.SerializeConfig;


import com.fasterxml.jackson.databind.ObjectMapper;


import com.fasterxml.jackson.databind.node.ArrayNode;


import com.fasterxml.jackson.databind.node.ObjectNode;


import com.jfinal.template.stat.ast.Break;


import com.mybatisflex.core.paginate.Page;


import com.mybatisflex.core.query.QueryWrapper;


import com.mybatisflex.core.table.TableInfo;


import com.mybatisflex.core.table.TableInfoFactory;


import io.milvus.param.R;


import org.slf4j.Logger;


import org.slf4j.LoggerFactory;


import org.springframework.http.HttpStatus;


import org.springframework.http.ResponseEntity;


import org.springframework.beans.factory.annotation.Autowired;


import org.springframework.http.*;


import org.springframework.web.bind.annotation.PostMapping;


import org.springframework.web.bind.annotation.RequestMapping;


import org.springframework.web.bind.annotation.RequestParam;


import org.springframework.web.bind.annotation.RestController;


import org.springframework.web.client.RestTemplate;


import org.springframework.web.context.request.RequestContextHolder;


import org.springframework.web.context.request.ServletRequestAttributes;


import org.springframework.web.servlet.mvc.method.annotation.SseEmitter;


import tech.aiflowy.ai.entity.*;


import tech.aiflowy.ai.mapper.AiBotConversationMessageMapper;


import tech.aiflowy.ai.service.*;


import tech.aiflowy.ai.vo.ModelConfig;


import tech.aiflowy.common.ai.ChatManager;


import tech.aiflowy.common.ai.MySseEmitter;


import tech.aiflowy.common.domain.Result;


@@ -48,7 +57,9 @@


import javax.annotation.Resource;


import javax.servlet.http.HttpServletRequest;


import javax.servlet.http.HttpServletResponse;


import java.io.IOException;


import java.math.BigInteger;


import java.time.Duration;


import java.util.*;


import java.util.concurrent.CompletableFuture;


import java.util.concurrent.atomic.AtomicBoolean;


@@ -77,6 +88,8 @@


    private AiBotConversationMessageMapper aiBotConversationMessageMapper;




    private static final Logger logger = LoggerFactory.getLogger(AiBotController.class);


    @Autowired


    private RestTemplate restTemplate;




    public AiBotController(AiBotService service, AiLlmService aiLlmService, AiBotWorkflowService aiBotWorkflowService, AiBotKnowledgeService aiBotKnowledgeService, AiBotMessageService aiBotMessageService) {


        super(service);


@@ -132,6 +145,10 @@


     * @param response


     * @return


     */




    @Autowired


    private ObjectMapper objectMapper;




    @PostMapping("chat")


    public SseEmitter chat(@JsonBody(value = "prompt", required = true) String prompt,


                           @JsonBody(value = "botId", required = true) BigInteger botId,


@@ -143,86 +160,77 @@


        if (aiBot == null) {


            return ChatManager.getInstance().sseEmitterForContent("机器人不存在");


        }


        if (StringUtil.hasText(aiBot.getApiEndpoint())){


            // 情况1：aiBot自带大模型信息


            try {


                // 从aiBot构建自定义LLM实现


                Llm llm = null;


                if (llm == null) {


                    return ChatManager.getInstance().sseEmitterForContent("LLM获取为空");


                }


        if (StringUtil.hasText(aiBot.getModelAPI())){


            // 使用Dify模型的逻辑


            String apiUrl = aiBot.getModelAPI();


            String bearerToken = aiBot.getModelKey();




                AiBotMessageMemory memory = new AiBotMessageMemory(botId, SaTokenUtil.getLoginAccount().getId(),


                        sessionId, isExternalMsg, aiBotMessageService, aiBotConversationMessageMapper,


                        aiBotConversationMessageService);


            // 构建请求JSON


            JSONObject jsonBody = new JSONObject();


            jsonBody.put("inputs", new JSONObject());


            jsonBody.put("query", prompt);


            jsonBody.put("response_mode", "blocking");


            jsonBody.put("conversation_id", "");


            jsonBody.put("user", userId.toString());




                final HistoriesPrompt historiesPrompt = new HistoriesPrompt();


            JSONArray files = new JSONArray();


            JSONObject fileObj = new JSONObject();


            fileObj.put("type", "");


            fileObj.put("transfer_method", "");


            fileObj.put("url", "");


            files.put(fileObj);


            jsonBody.put("files", files);




                historiesPrompt.setMemory(memory);


            // 发送HTTP请求


            HttpClient client = HttpClient.newBuilder()


                    .connectTimeout(Duration.ofSeconds(30))


                    .build();




                HumanMessage humanMessage = new HumanMessage(prompt);


            HttpRequest request = HttpRequest.newBuilder()


                    .uri(URI.create(apiUrl))


                    .header("Content-Type", "application/json")


                    .header("Authorization", "Bearer " + bearerToken)


                    .POST(HttpRequest.BodyPublishers.ofString(jsonBody.toString()))


                    .build();




                // 添加插件相关的function calling


                appendPluginToolFunction(botId, humanMessage);


            HttpResponse<String> response = client.send(request, HttpResponse.BodyHandlers.ofString());




                //添加工作流相关的 Function Calling


                appendWorkflowFunctions(botId, humanMessage);


            // 处理成功响应


            JSONObject responseJson = new JSONObject(response.body());




                //添加知识库相关的 Function Calling


                appendKnowledgeFunctions(botId, humanMessage);


            // 保存用户消息到历史记录


            ChatHistory userChat = new ChatHistory();


            userChat.setContent(prompt);


            userChat.setRole("user");


            userChat.setUserId(userId);


            userChat.setBotId(botId.intValue());


            userChat.setModel("dify");


            userChat.setChatId(responseJson.optString("conversation_id", ""));


            Integer userHistoryId = chatHistoryService.saveChatHistory(userChat);




                historiesPrompt.addMessage(humanMessage);


            // 保存AI回复到历史记录


            ChatHistory assistantChat = new ChatHistory();


            assistantChat.setContent(responseJson.getString("answer"));


            assistantChat.setRole("assistant");


            assistantChat.setUserId(userId);


            assistantChat.setBotId(botId.intValue());


            assistantChat.setModel("dify");


            assistantChat.setChatId(userChat.getChatId());


            Integer assistantHistoryId = chatHistoryService.saveChatHistory(assistantChat);




                MySseEmitter emitter = new MySseEmitter((long) (1000 * 60 * 2));


            // 构建响应数据


            int count = chatHistoryService.getChatHistoryCount(userId, botId.intValue());


            JSONObject result = new JSONObject();


            result.put("Count", count);


            result.put("Data", responseJson.getString("answer"));


            result.put("UserId", userHistoryId);


            result.put("AssistantId", assistantHistoryId);




                final Boolean[] needClose = {true};




                ServletRequestAttributes sra = (ServletRequestAttributes) RequestContextHolder.getRequestAttributes();


                // 统一使用流式处理，无论是否有 Function Calling


                llm.chatStream(historiesPrompt, new StreamResponseListener() {


                    @Override


                    public void onMessage(ChatContext context, AiMessageResponse response) {


                        try {


                            RequestContextHolder.setRequestAttributes(sra, true);


                            if (response != null) {


                                // 检查是否需要触发 Function Calling


                                if (response.getFunctionCallers() != null && CollectionUtil.hasItems(response.getFunctionCallers())) {


                                    needClose[0] = false;


                                    function_call(response, emitter, needClose, historiesPrompt, llm, prompt, false);


                                } else {


                                    // 强制流式返回，即使有 Function Calling 也先返回部分结果


                                    if (response.getMessage() != null) {


                                        String content = response.getMessage().getContent();


                                        if (StringUtil.hasText(content)) {


                                            emitter.send(JSON.toJSONString(response.getMessage()));


                                        }


                                    }


                                }


                            }






                        } catch (Exception e) {


                            emitter.completeWithError(e);


                        }


                    }




                    @Override


                    public void onStop(ChatContext context) {


                        if (needClose[0]) {


                            emitter.complete();


                        }


                    }




                    @Override


                    public void onFailure(ChatContext context, Throwable throwable) {


                        emitter.completeWithError(throwable);


                    }


                });




                return emitter;


            } catch (Exception e) {


                return ChatManager.getInstance().sseEmitterForContent("自定义LLM配置错误");


            }


            // 发送SSE响应


            emitter.send(SseEmitter.event()


                    .name("message")


                    .data(result.toString()));


            emitter.complete();


        }else{


            Map<String, Object> llmOptions = aiBot.getLlmOptions();


            String systemPrompt = llmOptions != null ? (String) llmOptions.get("systemPrompt") : null;