feat: chat quote

src/api/chat.ts

@@ -1,10 +1,11 @@
 import { GET, POST, DELETE } from './request';
-import type { ChatItemType, HistoryItemType } from '@/types/chat';
+import type { HistoryItemType } from '@/types/chat';
 import type { InitChatResponse, InitShareChatResponse } from './response/chat';
 import { RequestPaging } from '../types/index';
 import type { ShareChatSchema } from '@/types/mongoSchema';
 import type { ShareChatEditType } from '@/types/model';
 import { Obj2Query } from '@/utils/tools';
+import { Response as LastChatResultResponseType } from '@/pages/api/openapi/chat/lastChatResult';
 
 /**
  * 获取初始化聊天内容
@@ -24,15 +25,10 @@ export const getChatHistory = (data: RequestPaging) =>
 export const delChatHistoryById = (id: string) => GET(`/chat/removeHistory?id=${id}`);
 
 /**
- * 存储一轮对话
+ * get latest chat result by chatId
  */
-export const postSaveChat = (data: {
+export const getChatResult = (chatId: string) =>
-  modelId: string;
+  GET<LastChatResultResponseType>('/openapi/chat/lastChatResult', { chatId });
-  newChatId: '' | string;
-  chatId: '' | string;
-  prompts: [ChatItemType, ChatItemType];
-}) => POST<string>('/chat/saveChat', data);
-
 /**
  * 删除一句对话
  */

src/api/fetch.ts

@@ -1,4 +1,4 @@
-import { SYSTEM_PROMPT_HEADER, NEW_CHATID_HEADER } from '@/constants/chat';
+import { NEW_CHATID_HEADER } from '@/constants/chat';
 
 interface StreamFetchProps {
   url: string;
@@ -7,55 +7,52 @@ interface StreamFetchProps {
   abortSignal: AbortController;
 }
 export const streamFetch = ({ url, data, onMessage, abortSignal }: StreamFetchProps) =>
-  new Promise<{ responseText: string; systemPrompt: string; newChatId: string }>(
+  new Promise<{ responseText: string; newChatId: string }>(async (resolve, reject) => {
-    async (resolve, reject) => {
+    try {
-      try {
+      const res = await fetch(url, {
-        const res = await fetch(url, {
+        method: 'POST',
-          method: 'POST',
+        headers: {
-          headers: {
+          'Content-Type': 'application/json'
-            'Content-Type': 'application/json'
+        },
-          },
+        body: JSON.stringify(data),
-          body: JSON.stringify(data),
+        signal: abortSignal.signal
-          signal: abortSignal.signal
+      });
-        });
+      const reader = res.body?.getReader();
-        const reader = res.body?.getReader();
+      if (!reader) return;
-        if (!reader) return;
 
-        const decoder = new TextDecoder();
+      const decoder = new TextDecoder();
 
-        const systemPrompt = decodeURIComponent(res.headers.get(SYSTEM_PROMPT_HEADER) || '').trim();
+      const newChatId = decodeURIComponent(res.headers.get(NEW_CHATID_HEADER) || '');
-        const newChatId = decodeURIComponent(res.headers.get(NEW_CHATID_HEADER) || '');
 
-        let responseText = '';
+      let responseText = '';
 
-        const read = async () => {
+      const read = async () => {
-          try {
+        try {
-            const { done, value } = await reader?.read();
+          const { done, value } = await reader?.read();
-            if (done) {
+          if (done) {
-              if (res.status === 200) {
+            if (res.status === 200) {
-                resolve({ responseText, systemPrompt, newChatId });
+              resolve({ responseText, newChatId });
-              } else {
+            } else {
-                const parseError = JSON.parse(responseText);
+              const parseError = JSON.parse(responseText);
-                reject(parseError?.message || '请求异常');
+              reject(parseError?.message || '请求异常');
-              }
-
-              return;
             }
-            const text = decoder.decode(value);
+
-            responseText += text;
+            return;
-            onMessage(text);
-            read();
-          } catch (err: any) {
-            if (err?.message === 'The user aborted a request.') {
-              return resolve({ responseText, systemPrompt, newChatId });
-            }
-            reject(typeof err === 'string' ? err : err?.message || '请求异常');
           }
-        };
+          const text = decoder.decode(value);
-        read();
+          responseText += text;
-      } catch (err: any) {
+          onMessage(text);
-        console.log(err, '====');
+          read();
-        reject(typeof err === 'string' ? err : err?.message || '请求异常');
+        } catch (err: any) {
-      }
+          if (err?.message === 'The user aborted a request.') {
+            return resolve({ responseText, newChatId });
+          }
+          reject(typeof err === 'string' ? err : err?.message || '请求异常');
+        }
+      };
+      read();
+    } catch (err: any) {
+      console.log(err, '====');
+      reject(typeof err === 'string' ? err : err?.message || '请求异常');
     }
-  );
+  });

src/constants/chat.ts

@@ -1,5 +1,4 @@
-export const SYSTEM_PROMPT_HEADER = 'System-Prompt-Header';
+export const NEW_CHATID_HEADER = 'response-new-chat-id';
-export const NEW_CHATID_HEADER = 'Chat-Id-Header';
 
 export enum ChatRoleEnum {
   System = 'System',

src/pages/_app.tsx

@@ -55,7 +55,7 @@ export default function App({ Component, pageProps }: AppProps) {
       <Script src="/js/qrcode.min.js" strategy="lazyOnload"></Script>
       <Script src="/js/pdf.js" strategy="lazyOnload"></Script>
       <Script src="/js/html2pdf.bundle.min.js" strategy="lazyOnload"></Script>
-      <Script src="/js/particles.js" strategy="lazyOnload"></Script>
+      <Script src="/js/particles.js"></Script>
       <QueryClientProvider client={queryClient}>
         <ChakraProvider theme={theme}>
           <ColorModeScript initialColorMode={theme.config.initialColorMode} />

src/pages/api/chat/chat.ts

@@ -2,19 +2,24 @@ import type { NextApiRequest, NextApiResponse } from 'next';
 import { connectToDatabase } from '@/service/mongo';
 import { authChat } from '@/service/utils/auth';
 import { modelServiceToolMap } from '@/service/utils/chat';
-import { ChatItemSimpleType } from '@/types/chat';
+import { ChatItemType } from '@/types/chat';
 import { jsonRes } from '@/service/response';
 import { ChatModelMap, ModelVectorSearchModeMap } from '@/constants/model';
 import { pushChatBill } from '@/service/events/pushBill';
 import { resStreamResponse } from '@/service/utils/chat';
-import { searchKb } from '@/service/plugins/searchKb';
+import { appKbSearch } from '../openapi/kb/appKbSearch';
 import { ChatRoleEnum } from '@/constants/chat';
 import { BillTypeEnum } from '@/constants/user';
 import { sensitiveCheck } from '@/service/api/text';
+import { NEW_CHATID_HEADER } from '@/constants/chat';
+import { saveChat } from './saveChat';
+import { Types } from 'mongoose';
 
 /* 发送提示词 */
 export default async function handler(req: NextApiRequest, res: NextApiResponse) {
-  let step = 0; // step=1时，表示开始了流响应
+  res.on('close', () => {
+    res.end();
+  });
   res.on('error', () => {
     console.log('error: ', 'request error');
     res.end();
@@ -22,9 +27,9 @@ export default async function handler(req: NextApiRequest, res: NextApiResponse)
 
   try {
     const { chatId, prompt, modelId } = req.body as {
-      prompt: ChatItemSimpleType;
+      prompt: [ChatItemType, ChatItemType];
       modelId: string;
-      chatId: '' | string;
+      chatId?: string;
     };
 
     if (!modelId || !prompt) {
@@ -44,42 +49,69 @@ export default async function handler(req: NextApiRequest, res: NextApiResponse)
     const modelConstantsData = ChatModelMap[model.chat.chatModel];
 
     // 读取对话内容
-    const prompts = [...content, prompt];
+    const prompts = [...content, prompt[0]];
-    let systemPrompts: {
+    const {
-      obj: ChatRoleEnum;
+      code = 200,
-      value: string;
+      systemPrompts = [],
-    }[] = [];
+      quote = []
+    } = await (async () => {
+      // 使用了知识库搜索
+      if (model.chat.relatedKbs.length > 0) {
+        const { code, searchPrompts, rawSearch } = await appKbSearch({
+          model,
+          userId,
+          prompts,
+          similarity: ModelVectorSearchModeMap[model.chat.searchMode]?.similarity
+        });
 
-    // 使用了知识库搜索
+        return {
-    if (model.chat.relatedKbs.length > 0) {
+          code,
-      const { code, searchPrompts } = await searchKb({
+          quote: rawSearch,
-        userOpenAiKey,
+          systemPrompts: searchPrompts
-        prompts,
+        };
-        similarity: ModelVectorSearchModeMap[model.chat.searchMode]?.similarity,
+      }
-        model,
+      if (model.chat.systemPrompt) {
+        return {
+          systemPrompts: [
+            {
+              obj: ChatRoleEnum.System,
+              value: model.chat.systemPrompt
+            }
+          ]
+        };
+      }
+      return {};
+    })();
+
+    // get conversationId. create a newId if it is null
+    const conversationId = chatId || String(new Types.ObjectId());
+    !chatId && res?.setHeader(NEW_CHATID_HEADER, conversationId);
+
+    // search result is empty
+    if (code === 201) {
+      const response = systemPrompts[0]?.value;
+      await saveChat({
+        chatId,
+        newChatId: conversationId,
+        modelId,
+        prompts: [
+          prompt[0],
+          {
+            ...prompt[1],
+            quote: [],
+            value: response
+          }
+        ],
         userId
       });
-
+      return res.end(response);
-      // search result is empty
-      if (code === 201) {
-        return res.send(searchPrompts[0]?.value);
-      }
-
-      systemPrompts = searchPrompts;
-    } else if (model.chat.systemPrompt) {
-      systemPrompts = [
-        {
-          obj: ChatRoleEnum.System,
-          value: model.chat.systemPrompt
-        }
-      ];
     }
 
     prompts.splice(prompts.length - 3, 0, ...systemPrompts);
 
     // content check
     await sensitiveCheck({
-      input: [...systemPrompts, prompt].map((item) => item.value).join('')
+      input: [...systemPrompts, prompt[0]].map((item) => item.value).join('')
     });
 
     // 计算温度
@@ -87,54 +119,65 @@ export default async function handler(req: NextApiRequest, res: NextApiResponse)
       2
     );
 
-    // 发出请求
+    // 发出 chat 请求
     const { streamResponse } = await modelServiceToolMap[model.chat.chatModel].chatCompletion({
       apiKey: userOpenAiKey || systemAuthKey,
       temperature: +temperature,
       messages: prompts,
       stream: true,
       res,
-      chatId
+      chatId: conversationId
     });
 
     console.log('api response time:', `${(Date.now() - startTime) / 1000}s`);
 
-    step = 1;
+    if (res.closed) return res.end();
 
-    const { totalTokens, finishMessages } = await resStreamResponse({
+    try {
-      model: model.chat.chatModel,
+      const { totalTokens, finishMessages, responseContent } = await resStreamResponse({
-      res,
+        model: model.chat.chatModel,
-      chatResponse: streamResponse,
+        res,
-      prompts,
+        chatResponse: streamResponse,
-      systemPrompt: showModelDetail
+        prompts
-        ? prompts
-            .filter((item) => item.obj === ChatRoleEnum.System)
-            .map((item) => item.value)
-            .join('\n')
-        : ''
-    });
-
-    // 只有使用平台的 key 才计费
-    pushChatBill({
-      isPay: !userOpenAiKey,
-      chatModel: model.chat.chatModel,
-      userId,
-      chatId,
-      textLen: finishMessages.map((item) => item.value).join('').length,
-      tokens: totalTokens,
-      type: BillTypeEnum.chat
-    });
-  } catch (err: any) {
-    if (step === 1) {
-      // 直接结束流
-      res.end();
-      console.log('error，结束');
-    } else {
-      res.status(500);
-      jsonRes(res, {
-        code: 500,
-        error: err
       });
+
+      // save chat
+      await saveChat({
+        chatId,
+        newChatId: conversationId,
+        modelId,
+        prompts: [
+          prompt[0],
+          {
+            ...prompt[1],
+            quote: showModelDetail ? quote : [],
+            value: responseContent
+          }
+        ],
+        userId
+      });
+
+      res.end();
+
+      // 只有使用平台的 key 才计费
+      pushChatBill({
+        isPay: !userOpenAiKey,
+        chatModel: model.chat.chatModel,
+        userId,
+        chatId: conversationId,
+        textLen: finishMessages.map((item) => item.value).join('').length,
+        tokens: totalTokens,
+        type: BillTypeEnum.chat
+      });
+    } catch (error) {
+      res.end();
+      console.log('error，结束', error);
     }
+  } catch (err: any) {
+    res.status(500);
+    jsonRes(res, {
+      code: 500,
+      error: err
+    });
   }
 }

src/pages/api/chat/init.ts

@@ -73,7 +73,7 @@ export default async function handler(req: NextApiRequest, res: NextApiResponse)
             _id: '$content._id',
             obj: '$content.obj',
             value: '$content.value',
-            systemPrompt: '$content.systemPrompt'
+            quote: '$content.quote'
           }
         }
       ]);

src/pages/api/chat/saveChat.ts

@@ -6,15 +6,17 @@ import { authModel } from '@/service/utils/auth';
 import { authUser } from '@/service/utils/auth';
 import mongoose from 'mongoose';
 
+type Props = {
+  newChatId?: string;
+  chatId?: string;
+  modelId: string;
+  prompts: [ChatItemType, ChatItemType];
+};
+
 /* 聊天内容存存储 */
 export default async function handler(req: NextApiRequest, res: NextApiResponse) {
   try {
-    const { chatId, modelId, prompts, newChatId } = req.body as {
+    const { chatId, modelId, prompts, newChatId } = req.body as Props;
-      newChatId: '' | string;
-      chatId: '' | string;
-      modelId: string;
-      prompts: [ChatItemType, ChatItemType];
-    };
 
     if (!prompts) {
       throw new Error('缺少参数');
@@ -22,44 +24,17 @@ export default async function handler(req: NextApiRequest, res: NextApiResponse)
 
     const { userId } = await authUser({ req, authToken: true });
 
-    await connectToDatabase();
+    const nId = await saveChat({
+      chatId,
+      modelId,
+      prompts,
+      newChatId,
+      userId
+    });
 
-    const content = prompts.map((item) => ({
+    jsonRes(res, {
-      _id: new mongoose.Types.ObjectId(item._id),
+      data: nId
-      obj: item.obj,
+    });
-      value: item.value,
-      systemPrompt: item.systemPrompt
-    }));
-
-    await authModel({ modelId, userId, authOwner: false });
-
-    // 没有 chatId, 创建一个对话
-    if (!chatId) {
-      const { _id } = await Chat.create({
-        _id: newChatId ? new mongoose.Types.ObjectId(newChatId) : undefined,
-        userId,
-        modelId,
-        content,
-        title: content[0].value.slice(0, 20),
-        latestChat: content[1].value
-      });
-      return jsonRes(res, {
-        data: _id
-      });
-    } else {
-      // 已经有记录，追加入库
-      await Chat.findByIdAndUpdate(chatId, {
-        $push: {
-          content: {
-            $each: content
-          }
-        },
-        title: content[0].value.slice(0, 20),
-        latestChat: content[1].value,
-        updateTime: new Date()
-      });
-    }
-    jsonRes(res);
   } catch (err) {
     jsonRes(res, {
       code: 500,
@@ -67,3 +42,46 @@ export default async function handler(req: NextApiRequest, res: NextApiResponse)
     });
   }
 }
+
+export async function saveChat({
+  chatId,
+  newChatId,
+  modelId,
+  prompts,
+  userId
+}: Props & { userId: string }) {
+  await connectToDatabase();
+  await authModel({ modelId, userId, authOwner: false });
+
+  const content = prompts.map((item) => ({
+    _id: item._id ? new mongoose.Types.ObjectId(item._id) : undefined,
+    obj: item.obj,
+    value: item.value,
+    quote: item.quote
+  }));
+
+  // 没有 chatId, 创建一个对话
+  if (!chatId) {
+    const { _id } = await Chat.create({
+      _id: newChatId ? new mongoose.Types.ObjectId(newChatId) : undefined,
+      userId,
+      modelId,
+      content,
+      title: content[0].value.slice(0, 20),
+      latestChat: content[1].value
+    });
+    return _id;
+  } else {
+    // 已经有记录，追加入库
+    await Chat.findByIdAndUpdate(chatId, {
+      $push: {
+        content: {
+          $each: content
+        }
+      },
+      title: content[0].value.slice(0, 20),
+      latestChat: content[1].value,
+      updateTime: new Date()
+    });
+  }
+}

src/pages/api/chat/shareChat/chat.ts

@@ -7,14 +7,13 @@ import { jsonRes } from '@/service/response';
 import { ChatModelMap, ModelVectorSearchModeMap } from '@/constants/model';
 import { pushChatBill, updateShareChatBill } from '@/service/events/pushBill';
 import { resStreamResponse } from '@/service/utils/chat';
-import { searchKb } from '@/service/plugins/searchKb';
 import { ChatRoleEnum } from '@/constants/chat';
 import { BillTypeEnum } from '@/constants/user';
 import { sensitiveCheck } from '@/service/api/text';
+import { appKbSearch } from '../../openapi/kb/appKbSearch';
 
 /* 发送提示词 */
 export default async function handler(req: NextApiRequest, res: NextApiResponse) {
-  let step = 0; // step=1 时，表示开始了流响应
   res.on('error', () => {
     console.log('error: ', 'request error');
     res.end();
@@ -42,34 +41,37 @@ export default async function handler(req: NextApiRequest, res: NextApiResponse)
 
     const modelConstantsData = ChatModelMap[model.chat.chatModel];
 
-    let systemPrompts: {
+    const { code = 200, systemPrompts = [] } = await (async () => {
-      obj: ChatRoleEnum;
+      // 使用了知识库搜索
-      value: string;
+      if (model.chat.relatedKbs.length > 0) {
-    }[] = [];
+        const { code, searchPrompts } = await appKbSearch({
+          model,
+          userId,
+          prompts,
+          similarity: ModelVectorSearchModeMap[model.chat.searchMode]?.similarity
+        });
 
-    // 使用了知识库搜索
+        return {
-    if (model.chat.relatedKbs.length > 0) {
+          code,
-      const { code, searchPrompts } = await searchKb({
+          systemPrompts: searchPrompts
-        userOpenAiKey,
+        };
-        prompts,
-        similarity: ModelVectorSearchModeMap[model.chat.searchMode]?.similarity,
-        model,
-        userId
-      });
-
-      // search result is empty
-      if (code === 201) {
-        return res.send(searchPrompts[0]?.value);
       }
+      if (model.chat.systemPrompt) {
+        return {
+          systemPrompts: [
+            {
+              obj: ChatRoleEnum.System,
+              value: model.chat.systemPrompt
+            }
+          ]
+        };
+      }
+      return {};
+    })();
 
-      systemPrompts = searchPrompts;
+    // search result is empty
-    } else if (model.chat.systemPrompt) {
+    if (code === 201) {
-      systemPrompts = [
+      return res.send(systemPrompts[0]?.value);
-        {
-          obj: ChatRoleEnum.System,
-          value: model.chat.systemPrompt
-        }
-      ];
     }
 
     prompts.splice(prompts.length - 3, 0, ...systemPrompts);
@@ -96,40 +98,40 @@ export default async function handler(req: NextApiRequest, res: NextApiResponse)
 
     console.log('api response time:', `${(Date.now() - startTime) / 1000}s`);
 
-    step = 1;
+    if (res.closed) return res.end();
 
-    const { totalTokens, finishMessages } = await resStreamResponse({
+    try {
-      model: model.chat.chatModel,
+      const { totalTokens, finishMessages } = await resStreamResponse({
-      res,
+        model: model.chat.chatModel,
-      chatResponse: streamResponse,
+        res,
-      prompts,
+        chatResponse: streamResponse,
-      systemPrompt: ''
+        prompts
-    });
-
-    /* bill */
-    pushChatBill({
-      isPay: !userOpenAiKey,
-      chatModel: model.chat.chatModel,
-      userId,
-      textLen: finishMessages.map((item) => item.value).join('').length,
-      tokens: totalTokens,
-      type: BillTypeEnum.chat
-    });
-    updateShareChatBill({
-      shareId,
-      tokens: totalTokens
-    });
-  } catch (err: any) {
-    if (step === 1) {
-      // 直接结束流
-      res.end();
-      console.log('error，结束');
-    } else {
-      res.status(500);
-      jsonRes(res, {
-        code: 500,
-        error: err
       });
+
+      res.end();
+
+      /* bill */
+      pushChatBill({
+        isPay: !userOpenAiKey,
+        chatModel: model.chat.chatModel,
+        userId,
+        textLen: finishMessages.map((item) => item.value).join('').length,
+        tokens: totalTokens,
+        type: BillTypeEnum.chat
+      });
+      updateShareChatBill({
+        shareId,
+        tokens: totalTokens
+      });
+    } catch (error) {
+      res.end();
+      console.log('error，结束', error);
     }
+  } catch (err: any) {
+    res.status(500);
+    jsonRes(res, {
+      code: 500,
+      error: err
+    });
   }
 }

src/pages/api/openapi/chat/chat.ts

@@ -6,15 +6,19 @@ import { ChatItemSimpleType } from '@/types/chat';
 import { jsonRes } from '@/service/response';
 import { ChatModelMap, ModelVectorSearchModeMap } from '@/constants/model';
 import { pushChatBill } from '@/service/events/pushBill';
-import { searchKb } from '@/service/plugins/searchKb';
 import { ChatRoleEnum } from '@/constants/chat';
 import { withNextCors } from '@/service/utils/tools';
 import { BillTypeEnum } from '@/constants/user';
 import { sensitiveCheck } from '@/service/api/text';
+import { NEW_CHATID_HEADER } from '@/constants/chat';
+import { Types } from 'mongoose';
+import { appKbSearch } from '../kb/appKbSearch';
 
 /* 发送提示词 */
 export default withNextCors(async function handler(req: NextApiRequest, res: NextApiResponse) {
-  let step = 0; // step=1时，表示开始了流响应
+  res.on('close', () => {
+    res.end();
+  });
   res.on('error', () => {
     console.log('error: ', 'request error');
     res.end();
@@ -70,7 +74,7 @@ export default withNextCors(async function handler(req: NextApiRequest, res: Nex
 
     // 使用了知识库搜索
     if (model.chat.relatedKbs.length > 0) {
-      const { code, searchPrompts } = await searchKb({
+      const { code, searchPrompts } = await appKbSearch({
         prompts,
         similarity: ModelVectorSearchModeMap[model.chat.searchMode]?.similarity,
         model,
@@ -109,6 +113,10 @@ export default withNextCors(async function handler(req: NextApiRequest, res: Nex
       2
     );
 
+    // get conversationId. create a newId if it is null
+    const conversationId = chatId || String(new Types.ObjectId());
+    !chatId && res?.setHeader(NEW_CHATID_HEADER, conversationId);
+
     // 发出请求
     const { streamResponse, responseMessages, responseText, totalTokens } =
       await modelServiceToolMap[model.chat.chatModel].chatCompletion({
@@ -117,30 +125,41 @@ export default withNextCors(async function handler(req: NextApiRequest, res: Nex
         messages: prompts,
         stream: isStream,
         res,
-        chatId
+        chatId: conversationId
       });
 
     console.log('api response time:', `${(Date.now() - startTime) / 1000}s`);
 
-    let textLen = 0;
+    if (res.closed) return res.end();
-    let tokens = totalTokens;
 
-    if (isStream) {
+    const { textLen = 0, tokens = totalTokens } = await (async () => {
-      step = 1;
+      if (isStream) {
-      const { finishMessages, totalTokens } = await resStreamResponse({
+        try {
-        model: model.chat.chatModel,
+          const { finishMessages, totalTokens } = await resStreamResponse({
-        res,
+            model: model.chat.chatModel,
-        chatResponse: streamResponse,
+            res,
-        prompts
+            chatResponse: streamResponse,
-      });
+            prompts
-      textLen = finishMessages.map((item) => item.value).join('').length;
+          });
-      tokens = totalTokens;
+          res.end();
-    } else {
+          return {
-      textLen = responseMessages.map((item) => item.value).join('').length;
+            textLen: finishMessages.map((item) => item.value).join('').length,
-      jsonRes(res, {
+            tokens: totalTokens
-        data: responseText
+          };
-      });
+        } catch (error) {
-    }
+          res.end();
+          console.log('error，结束', error);
+        }
+      } else {
+        jsonRes(res, {
+          data: responseText
+        });
+        return {
+          textLen: responseMessages.map((item) => item.value).join('').length
+        };
+      }
+      return {};
+    })();
 
     pushChatBill({
       isPay: true,
@@ -151,16 +170,10 @@ export default withNextCors(async function handler(req: NextApiRequest, res: Nex
       type: BillTypeEnum.openapiChat
     });
   } catch (err: any) {
-    if (step === 1) {
+    res.status(500);
-      // 直接结束流
+    jsonRes(res, {
-      res.end();
+      code: 500,
-      console.log('error，结束');
+      error: err
-    } else {
+    });
-      res.status(500);
-      jsonRes(res, {
-        code: 500,
-        error: err
-      });
-    }
   }
 });

src/pages/api/openapi/chat/lastChatResult.ts

@@ -0,0 +1,38 @@
+import type { NextApiRequest, NextApiResponse } from 'next';
+import { jsonRes } from '@/service/response';
+import { Chat } from '@/service/mongo';
+import { authUser } from '@/service/utils/auth';
+import { QuoteItemType } from '../kb/appKbSearch';
+
+type Props = {
+  chatId: string;
+};
+export type Response = {
+  quote: QuoteItemType[];
+};
+
+/* 聊天内容存存储 */
+export default async function handler(req: NextApiRequest, res: NextApiResponse) {
+  try {
+    const { chatId } = req.query as Props;
+
+    if (!chatId) {
+      throw new Error('缺少参数');
+    }
+
+    const { userId } = await authUser({ req });
+
+    const chatItem = await Chat.findOne({ _id: chatId, userId }, { content: { $slice: -1 } });
+
+    jsonRes<Response>(res, {
+      data: {
+        quote: chatItem?.content[0]?.quote || []
+      }
+    });
+  } catch (err) {
+    jsonRes(res, {
+      code: 500,
+      error: err
+    });
+  }
+}

src/pages/api/openapi/kb/appKbSearch.ts

@@ -0,0 +1,224 @@
+import type { NextApiRequest, NextApiResponse } from 'next';
+import { jsonRes } from '@/service/response';
+import { authUser } from '@/service/utils/auth';
+import { PgClient } from '@/service/pg';
+import { withNextCors } from '@/service/utils/tools';
+import type { ChatItemSimpleType } from '@/types/chat';
+import type { ModelSchema } from '@/types/mongoSchema';
+import { ModelVectorSearchModeEnum } from '@/constants/model';
+import { authModel } from '@/service/utils/auth';
+import { ChatModelMap } from '@/constants/model';
+import { ChatRoleEnum } from '@/constants/chat';
+import { openaiEmbedding } from '../plugin/openaiEmbedding';
+import { ModelDataStatusEnum } from '@/constants/model';
+import { modelToolMap } from '@/utils/plugin';
+
+export type QuoteItemType = { id: string; q: string; a: string };
+type Props = {
+  prompts: ChatItemSimpleType[];
+  similarity: number;
+  appId: string;
+};
+type Response = {
+  code: 200 | 201;
+  rawSearch: QuoteItemType[];
+  searchPrompts: {
+    obj: ChatRoleEnum;
+    value: string;
+  }[];
+};
+
+export default withNextCors(async function handler(req: NextApiRequest, res: NextApiResponse<any>) {
+  try {
+    const { userId } = await authUser({ req });
+
+    if (!userId) {
+      throw new Error('userId is empty');
+    }
+
+    const { prompts, similarity, appId } = req.body as Props;
+
+    if (!similarity || !Array.isArray(prompts) || !appId) {
+      throw new Error('params is error');
+    }
+
+    // auth model
+    const { model } = await authModel({
+      modelId: appId,
+      userId
+    });
+
+    const result = await appKbSearch({
+      userId,
+      prompts,
+      similarity,
+      model
+    });
+
+    jsonRes<Response>(res, {
+      data: result
+    });
+  } catch (err) {
+    console.log(err);
+    jsonRes(res, {
+      code: 500,
+      error: err
+    });
+  }
+});
+
+export async function appKbSearch({
+  model,
+  userId,
+  prompts,
+  similarity
+}: {
+  userId: string;
+  prompts: ChatItemSimpleType[];
+  similarity: number;
+  model: ModelSchema;
+}): Promise<Response> {
+  const modelConstantsData = ChatModelMap[model.chat.chatModel];
+
+  // search two times.
+  const userPrompts = prompts.filter((item) => item.obj === 'Human');
+
+  const input: string[] = [
+    userPrompts[userPrompts.length - 1].value,
+    userPrompts[userPrompts.length - 2]?.value
+  ].filter((item) => item);
+
+  // get vector
+  const promptVectors = await openaiEmbedding({
+    userId,
+    input
+  });
+
+  // search kb
+  const searchRes = await Promise.all(
+    promptVectors.map((promptVector) =>
+      PgClient.select<{ id: string; q: string; a: string }>('modelData', {
+        fields: ['id', 'q', 'a'],
+        where: [
+          ['status', ModelDataStatusEnum.ready],
+          'AND',
+          `kb_id IN (${model.chat.relatedKbs.map((item) => `'${item}'`).join(',')})`,
+          'AND',
+          `vector <=> '[${promptVector}]' < ${similarity}`
+        ],
+        order: [{ field: 'vector', mode: `<=> '[${promptVector}]'` }],
+        limit: promptVectors.length === 1 ? 15 : 10
+      }).then((res) => res.rows)
+    )
+  );
+
+  // filter same search result
+  const idSet = new Set<string>();
+  const filterSearch = searchRes.map((search) =>
+    search.filter((item) => {
+      if (idSet.has(item.id)) {
+        return false;
+      }
+      idSet.add(item.id);
+      return true;
+    })
+  );
+
+  // slice search result by rate.
+  const sliceRateMap: Record<number, number[]> = {
+    1: [1],
+    2: [0.7, 0.3]
+  };
+  const sliceRate = sliceRateMap[searchRes.length] || sliceRateMap[0];
+  // 计算固定提示词的 token 数量
+  const fixedPrompts = [
+    // user system prompt
+    ...(model.chat.systemPrompt
+      ? [
+          {
+            obj: ChatRoleEnum.System,
+            value: model.chat.systemPrompt
+          }
+        ]
+      : model.chat.searchMode === ModelVectorSearchModeEnum.noContext
+      ? [
+          {
+            obj: ChatRoleEnum.System,
+            value: `知识库是关于"${model.name}"的内容,根据知识库内容回答问题.`
+          }
+        ]
+      : [
+          {
+            obj: ChatRoleEnum.System,
+            value: `玩一个问答游戏,规则为:
+1.你完全忘记你已有的知识
+2.你只回答关于"${model.name}"的问题
+3.你只从知识库中选择内容进行回答
+4.如果问题不在知识库中,你会回答:"我不知道。"
+请务必遵守规则`
+          }
+        ])
+  ];
+  const fixedSystemTokens = modelToolMap[model.chat.chatModel].countTokens({
+    messages: fixedPrompts
+  });
+  const maxTokens = modelConstantsData.systemMaxToken - fixedSystemTokens;
+  const sliceResult = sliceRate.map((rate, i) =>
+    modelToolMap[model.chat.chatModel]
+      .tokenSlice({
+        maxToken: Math.round(maxTokens * rate),
+        messages: filterSearch[i].map((item) => ({
+          obj: ChatRoleEnum.System,
+          value: `${item.q}\n${item.a}`
+        }))
+      })
+      .map((item) => item.value)
+  );
+
+  // slice filterSearch
+  const sliceSearch = filterSearch.map((item, i) => item.slice(0, sliceResult[i].length)).flat();
+
+  //  system prompt
+  const systemPrompt = sliceResult.flat().join('\n').trim();
+
+  /* 高相似度+不回复 */
+  if (!systemPrompt && model.chat.searchMode === ModelVectorSearchModeEnum.hightSimilarity) {
+    return {
+      code: 201,
+      rawSearch: [],
+      searchPrompts: [
+        {
+          obj: ChatRoleEnum.System,
+          value: '对不起，你的问题不在知识库中。'
+        }
+      ]
+    };
+  }
+  /* 高相似度+无上下文，不添加额外知识,仅用系统提示词 */
+  if (!systemPrompt && model.chat.searchMode === ModelVectorSearchModeEnum.noContext) {
+    return {
+      code: 200,
+      rawSearch: [],
+      searchPrompts: model.chat.systemPrompt
+        ? [
+            {
+              obj: ChatRoleEnum.System,
+              value: model.chat.systemPrompt
+            }
+          ]
+        : []
+    };
+  }
+
+  return {
+    code: 200,
+    rawSearch: sliceSearch,
+    searchPrompts: [
+      {
+        obj: ChatRoleEnum.System,
+        value: `知识库:${systemPrompt}`
+      },
+      ...fixedPrompts
+    ]
+  };
+}

src/pages/api/openapi/plugin/openaiEmbedding.ts

@@ -0,0 +1,77 @@
+import type { NextApiRequest, NextApiResponse } from 'next';
+import { jsonRes } from '@/service/response';
+import { authUser } from '@/service/utils/auth';
+import { PgClient } from '@/service/pg';
+import { withNextCors } from '@/service/utils/tools';
+import { getApiKey } from '@/service/utils/auth';
+import { getOpenAIApi } from '@/service/utils/chat/openai';
+import { embeddingModel } from '@/constants/model';
+import { axiosConfig } from '@/service/utils/tools';
+import { pushGenerateVectorBill } from '@/service/events/pushBill';
+
+type Props = {
+  input: string[];
+};
+type Response = number[][];
+
+export default withNextCors(async function handler(req: NextApiRequest, res: NextApiResponse<any>) {
+  try {
+    const { userId } = await authUser({ req });
+    let { input } = req.query as Props;
+
+    if (!Array.isArray(input)) {
+      throw new Error('缺少参数');
+    }
+
+    jsonRes<Response>(res, {
+      data: await openaiEmbedding({ userId, input, mustPay: true })
+    });
+  } catch (err) {
+    console.log(err);
+    jsonRes(res, {
+      code: 500,
+      error: err
+    });
+  }
+});
+
+export async function openaiEmbedding({
+  userId,
+  input,
+  mustPay = false
+}: { userId: string; mustPay?: boolean } & Props) {
+  const { userOpenAiKey, systemAuthKey } = await getApiKey({
+    model: 'gpt-3.5-turbo',
+    userId,
+    mustPay
+  });
+
+  // 获取 chatAPI
+  const chatAPI = getOpenAIApi();
+
+  // 把输入的内容转成向量
+  const result = await chatAPI
+    .createEmbedding(
+      {
+        model: embeddingModel,
+        input
+      },
+      {
+        timeout: 60000,
+        ...axiosConfig(userOpenAiKey || systemAuthKey)
+      }
+    )
+    .then((res) => ({
+      tokenLen: res.data.usage.total_tokens || 0,
+      vectors: res.data.data.map((item) => item.embedding)
+    }));
+
+  pushGenerateVectorBill({
+    isPay: !userOpenAiKey,
+    userId,
+    text: input.join(''),
+    tokenLen: result.tokenLen
+  });
+
+  return result.vectors;
+}

src/pages/api/openapi/text/gptMessagesSlice.ts

@@ -0,0 +1,119 @@
+// Next.js API route support: https://nextjs.org/docs/api-routes/introduction
+import type { NextApiRequest, NextApiResponse } from 'next';
+import { type Tiktoken } from '@dqbd/tiktoken';
+import { jsonRes } from '@/service/response';
+import { authUser } from '@/service/utils/auth';
+import Graphemer from 'graphemer';
+import type { ChatItemSimpleType } from '@/types/chat';
+import { ChatCompletionRequestMessage } from 'openai';
+import { getOpenAiEncMap } from '@/utils/plugin/openai';
+import { adaptChatItem_openAI } from '@/utils/plugin/openai';
+
+type ModelType = 'gpt-3.5-turbo' | 'gpt-4' | 'gpt-4-32k';
+
+type Props = {
+  messages: ChatItemSimpleType[];
+  model: ModelType;
+  maxLen: number;
+};
+type Response = ChatItemSimpleType[];
+
+export default async function handler(req: NextApiRequest, res: NextApiResponse) {
+  try {
+    await authUser({ req });
+
+    const { messages, model, maxLen } = req.body as Props;
+
+    if (!Array.isArray(messages) || !model || !maxLen) {
+      throw new Error('params is error');
+    }
+
+    return jsonRes<Response>(res, {
+      data: gpt_chatItemTokenSlice({
+        messages,
+        model,
+        maxToken: maxLen
+      })
+    });
+  } catch (err) {
+    jsonRes(res, {
+      code: 500,
+      error: err
+    });
+  }
+}
+
+export function gpt_chatItemTokenSlice({
+  messages,
+  model,
+  maxToken
+}: {
+  messages: ChatItemSimpleType[];
+  model: ModelType;
+  maxToken: number;
+}) {
+  const textDecoder = new TextDecoder();
+  const graphemer = new Graphemer();
+
+  function getChatGPTEncodingText(messages: ChatCompletionRequestMessage[], model: ModelType) {
+    const isGpt3 = model === 'gpt-3.5-turbo';
+
+    const msgSep = isGpt3 ? '\n' : '';
+    const roleSep = isGpt3 ? '\n' : '<|im_sep|>';
+
+    return [
+      messages
+        .map(({ name = '', role, content }) => {
+          return `<|im_start|>${name || role}${roleSep}${content}<|im_end|>`;
+        })
+        .join(msgSep),
+      `<|im_start|>assistant${roleSep}`
+    ].join(msgSep);
+  }
+  function text2TokensLen(encoder: Tiktoken, inputText: string) {
+    const encoding = encoder.encode(inputText, 'all');
+    const segments: { text: string; tokens: { id: number; idx: number }[] }[] = [];
+
+    let byteAcc: number[] = [];
+    let tokenAcc: { id: number; idx: number }[] = [];
+    let inputGraphemes = graphemer.splitGraphemes(inputText);
+
+    for (let idx = 0; idx < encoding.length; idx++) {
+      const token = encoding[idx]!;
+      byteAcc.push(...encoder.decode_single_token_bytes(token));
+      tokenAcc.push({ id: token, idx });
+
+      const segmentText = textDecoder.decode(new Uint8Array(byteAcc));
+      const graphemes = graphemer.splitGraphemes(segmentText);
+
+      if (graphemes.every((item, idx) => inputGraphemes[idx] === item)) {
+        segments.push({ text: segmentText, tokens: tokenAcc });
+
+        byteAcc = [];
+        tokenAcc = [];
+        inputGraphemes = inputGraphemes.slice(graphemes.length);
+      }
+    }
+
+    return segments.reduce((memo, i) => memo + i.tokens.length, 0) ?? 0;
+  }
+  const OpenAiEncMap = getOpenAiEncMap();
+  const enc = OpenAiEncMap[model];
+
+  let result: ChatItemSimpleType[] = [];
+
+  for (let i = 0; i < messages.length; i++) {
+    const msgs = [...result, messages[i]];
+    const tokens = text2TokensLen(
+      enc,
+      getChatGPTEncodingText(adaptChatItem_openAI({ messages }), model)
+    );
+    if (tokens < maxToken) {
+      result = msgs;
+    } else {
+      break;
+    }
+  }
+
+  return result;
+}

src/pages/chat/index.tsx

@@ -3,10 +3,10 @@ import { useRouter } from 'next/router';
 import {
   getInitChatSiteInfo,
   delChatRecordByIndex,
-  postSaveChat,
+  getChatResult,
   delChatHistoryById
 } from '@/api/chat';
-import type { ChatSiteItemType, ExportChatType } from '@/types/chat';
+import type { ChatItemType, ChatSiteItemType, ExportChatType } from '@/types/chat';
 import {
   Textarea,
   Box,
@@ -29,13 +29,14 @@ import {
   Card,
   Tooltip,
   useOutsideClick,
-  useTheme
+  useTheme,
+  ModalHeader
 } from '@chakra-ui/react';
 import { useToast } from '@/hooks/useToast';
 import { useGlobalStore } from '@/store/global';
 import { useQuery } from '@tanstack/react-query';
 import dynamic from 'next/dynamic';
-import { useCopyData, voiceBroadcast, hasVoiceApi } from '@/utils/tools';
+import { useCopyData, voiceBroadcast, hasVoiceApi, delay } from '@/utils/tools';
 import { streamFetch } from '@/api/fetch';
 import MyIcon from '@/components/Icon';
 import { throttle } from 'lodash';
@@ -47,6 +48,7 @@ import { useLoading } from '@/hooks/useLoading';
 import { fileDownload } from '@/utils/file';
 import { htmlTemplate } from '@/constants/common';
 import { useUserStore } from '@/store/user';
+import type { QuoteItemType } from '@/pages/api/openapi/kb/appKbSearch';
 import Loading from '@/components/Loading';
 import Markdown from '@/components/Markdown';
 import SideBar from '@/components/SideBar';
@@ -78,7 +80,7 @@ const Chat = ({ modelId, chatId }: { modelId: string; chatId: string }) => {
   const controller = useRef(new AbortController());
   const isLeavePage = useRef(false);
 
-  const [showSystemPrompt, setShowSystemPrompt] = useState('');
+  const [showQuote, setShowQuote] = useState<QuoteItemType[]>([]);
   const [messageContextMenuData, setMessageContextMenuData] = useState<{
     // message messageContextMenuData
     left: number;
@@ -173,13 +175,14 @@ const Chat = ({ modelId, chatId }: { modelId: string; chatId: string }) => {
       controller.current = abortSignal;
       isLeavePage.current = false;
 
-      const prompt = {
+      const prompt: ChatItemType[] = prompts.map((item) => ({
-        obj: prompts[0].obj,
+        _id: item._id,
-        value: prompts[0].value
+        obj: item.obj,
-      };
+        value: item.value
+      }));
 
       // 流请求，获取数据
-      let { responseText, systemPrompt, newChatId } = await streamFetch({
+      const { newChatId } = await streamFetch({
         url: '/api/chat/chat',
         data: {
           prompt,
@@ -207,39 +210,16 @@ const Chat = ({ modelId, chatId }: { modelId: string; chatId: string }) => {
         return;
       }
 
-      // save chat record
+      if (newChatId) {
-      try {
+        setForbidLoadChatData(true);
-        newChatId = await postSaveChat({
+        router.replace(`/chat?modelId=${modelId}&chatId=${newChatId}`);
-          newChatId, // 如果有newChatId，会自动以这个Id创建对话框
-          modelId,
-          chatId,
-          prompts: [
-            {
-              _id: prompts[0]._id,
-              obj: 'Human',
-              value: prompt.value
-            },
-            {
-              _id: prompts[1]._id,
-              obj: 'AI',
-              value: responseText,
-              systemPrompt
-            }
-          ]
-        });
-        if (newChatId) {
-          setForbidLoadChatData(true);
-          router.replace(`/chat?modelId=${modelId}&chatId=${newChatId}`);
-        }
-      } catch (err) {
-        toast({
-          title: '对话出现异常, 继续对话会导致上下文丢失，请刷新页面',
-          status: 'warning',
-          duration: 3000,
-          isClosable: true
-        });
       }
 
+      abortSignal.signal.aborted && (await delay(600));
+
+      // get chat result
+      const { quote } = await getChatResult(chatId || newChatId);
+
       // 设置聊天内容为完成状态
       setChatData((state) => ({
         ...state,
@@ -249,7 +229,7 @@ const Chat = ({ modelId, chatId }: { modelId: string; chatId: string }) => {
           return {
             ...item,
             status: 'finish',
-            systemPrompt
+            quote
           };
         })
       }));
@@ -260,16 +240,7 @@ const Chat = ({ modelId, chatId }: { modelId: string; chatId: string }) => {
         generatingMessage();
       }, 100);
     },
-    [
+    [chatId, setForbidLoadChatData, generatingMessage, loadHistory, modelId, router, setChatData]
-      chatId,
-      setForbidLoadChatData,
-      generatingMessage,
-      loadHistory,
-      modelId,
-      router,
-      setChatData,
-      toast
-    ]
   );
 
   /**
@@ -717,24 +688,24 @@ const Chat = ({ modelId, chatId }: { modelId: string; chatId: string }) => {
                   {item.obj === 'Human' && <Box flex={1} />}
                   {/* avatar */}
                   <Menu autoSelect={false} isLazy>
-                    <MenuButton
+                    <Tooltip label={item.obj === 'AI' ? '应用详情' : ''}>
-                      as={Box}
+                      <MenuButton
-                      {...(item.obj === 'AI'
+                        as={Box}
-                        ? {
+                        {...(item.obj === 'AI'
-                            order: 1,
+                          ? {
-                            mr: ['6px', 2],
+                              order: 1,
-                            cursor: 'pointer',
+                              mr: ['6px', 2],
-                            onClick: () =>
+                              cursor: 'pointer',
-                              isPc &&
+                              onClick: () =>
-                              chatData.model.canUse &&
+                                isPc &&
-                              router.push(`/model?modelId=${chatData.modelId}`)
+                                chatData.model.canUse &&
-                          }
+                                router.push(`/model?modelId=${chatData.modelId}`)
-                        : {
+                            }
-                            order: 3,
+                          : {
-                            ml: ['6px', 2]
+                              order: 3,
-                          })}
+                              ml: ['6px', 2]
-                    >
+                            })}
-                      <Tooltip label={item.obj === 'AI' ? '应用详情' : ''}>
+                      >
                         <Avatar
                           src={
                             item.obj === 'Human'
@@ -744,8 +715,8 @@ const Chat = ({ modelId, chatId }: { modelId: string; chatId: string }) => {
                           w={['20px', '34px']}
                           h={['20px', '34px']}
                         />
-                      </Tooltip>
+                      </MenuButton>
-                    </MenuButton>
+                    </Tooltip>
                     {!isPc && <RenderContextMenu history={item} index={index} AiDetail />}
                   </Menu>
                   {/* message */}
@@ -764,7 +735,7 @@ const Chat = ({ modelId, chatId }: { modelId: string; chatId: string }) => {
                             isChatting={isChatting && index === chatData.history.length - 1}
                             formatLink
                           />
-                          {item.systemPrompt && (
+                          {item.quote && item.quote.length > 0 && (
                             <Button
                               size={'xs'}
                               mt={2}
@@ -772,9 +743,9 @@ const Chat = ({ modelId, chatId }: { modelId: string; chatId: string }) => {
                               colorScheme={'gray'}
                               variant={'outline'}
                               w={'90px'}
-                              onClick={() => setShowSystemPrompt(item.systemPrompt || '')}
+                              onClick={() => setShowQuote(item.quote || [])}
                             >
-                              查看提示词
+                              查看引用
                             </Button>
                           )}
                         </Card>
@@ -907,12 +878,24 @@ const Chat = ({ modelId, chatId }: { modelId: string; chatId: string }) => {
       )}
       {/* system prompt show modal */}
       {
-        <Modal isOpen={!!showSystemPrompt} onClose={() => setShowSystemPrompt('')}>
+        <Modal isOpen={showQuote.length > 0} onClose={() => setShowQuote([])}>
           <ModalOverlay />
-          <ModalContent pt={5} maxW={'min(90vw, 600px)'} h={'80vh'} overflow={'overlay'}>
+          <ModalContent maxW={'min(90vw, 700px)'} h={'80vh'} overflow={'overlay'}>
+            <ModalHeader>知识库引用({showQuote.length}条)</ModalHeader>
             <ModalCloseButton />
-            <ModalBody pt={5} whiteSpace={'pre-wrap'} textAlign={'justify'}>
+            <ModalBody whiteSpace={'pre-wrap'} textAlign={'justify'} fontSize={'sm'}>
-              {showSystemPrompt}
+              {showQuote.map((item) => (
+                <Box
+                  key={item.id}
+                  p={2}
+                  borderRadius={'sm'}
+                  border={theme.borders.base}
+                  _notLast={{ mb: 2 }}
+                >
+                  <Box>{item.q}</Box>
+                  <Box>{item.a}</Box>
+                </Box>
+              ))}
             </ModalBody>
           </ModalContent>
         </Modal>

src/pages/chat/share.tsx

@@ -73,7 +73,6 @@ const Chat = ({ shareId, historyId }: { shareId: string; historyId: string }) =>
   const isLeavePage = useRef(false);
 
   const [inputVal, setInputVal] = useState(''); // user input prompt
-  const [showSystemPrompt, setShowSystemPrompt] = useState('');
   const [messageContextMenuData, setMessageContextMenuData] = useState<{
     // message messageContextMenuData
     left: number;
@@ -178,7 +177,7 @@ const Chat = ({ shareId, historyId }: { shareId: string; historyId: string }) =>
       }));
 
       // 流请求，获取数据
-      const { responseText, systemPrompt } = await streamFetch({
+      const { responseText } = await streamFetch({
         url: '/api/chat/shareChat/chat',
         data: {
           prompts: formatPrompts.slice(-shareChatData.maxContext - 1, -1),
@@ -215,8 +214,7 @@ const Chat = ({ shareId, historyId }: { shareId: string; historyId: string }) =>
           if (index !== state.history.length - 1) return item;
           return {
             ...item,
-            status: 'finish',
+            status: 'finish'
-            systemPrompt
           };
         });
 
@@ -614,19 +612,19 @@ const Chat = ({ shareId, historyId }: { shareId: string; historyId: string }) =>
                 {item.obj === 'Human' && <Box flex={1} />}
                 {/* avatar */}
                 <Menu autoSelect={false} isLazy>
-                  <MenuButton
+                  <Tooltip label={item.obj === 'AI' ? '应用详情' : ''}>
-                    as={Box}
+                    <MenuButton
-                    {...(item.obj === 'AI'
+                      as={Box}
-                      ? {
+                      {...(item.obj === 'AI'
-                          order: 1,
+                        ? {
-                          mr: ['6px', 2]
+                            order: 1,
-                        }
+                            mr: ['6px', 2]
-                      : {
+                          }
-                          order: 3,
+                        : {
-                          ml: ['6px', 2]
+                            order: 3,
-                        })}
+                            ml: ['6px', 2]
-                  >
+                          })}
-                    <Tooltip label={item.obj === 'AI' ? '应用详情' : ''}>
+                    >
                       <Avatar
                         src={
                           item.obj === 'Human'
@@ -636,8 +634,8 @@ const Chat = ({ shareId, historyId }: { shareId: string; historyId: string }) =>
                         w={['20px', '34px']}
                         h={['20px', '34px']}
                       />
-                    </Tooltip>
+                    </MenuButton>
-                  </MenuButton>
+                  </Tooltip>
                   {!isPc && <RenderContextMenu history={item} index={index} />}
                 </Menu>
                 {/* message */}
@@ -656,19 +654,6 @@ const Chat = ({ shareId, historyId }: { shareId: string; historyId: string }) =>
                           isChatting={isChatting && index === shareChatData.history.length - 1}
                           formatLink
                         />
-                        {item.systemPrompt && (
-                          <Button
-                            size={'xs'}
-                            mt={2}
-                            fontWeight={'normal'}
-                            colorScheme={'gray'}
-                            variant={'outline'}
-                            w={'90px'}
-                            onClick={() => setShowSystemPrompt(item.systemPrompt || '')}
-                          >
-                            查看提示词
-                          </Button>
-                        )}
                       </Card>
                     </Box>
                   ) : (
@@ -796,18 +781,6 @@ const Chat = ({ shareId, historyId }: { shareId: string; historyId: string }) =>
           </DrawerContent>
         </Drawer>
       )}
-      {/* system prompt show modal */}
-      {
-        <Modal isOpen={!!showSystemPrompt} onClose={() => setShowSystemPrompt('')}>
-          <ModalOverlay />
-          <ModalContent maxW={'min(90vw, 600px)'} pr={2} maxH={'80vh'} overflowY={'auto'}>
-            <ModalCloseButton />
-            <ModalBody pt={5} whiteSpace={'pre-wrap'} textAlign={'justify'}>
-              {showSystemPrompt}
-            </ModalBody>
-          </ModalContent>
-        </Modal>
-      }
       {/* context menu */}
       {messageContextMenuData && (
         <Box

src/pages/model/components/detail/components/ModelEditForm.tsx

@@ -358,7 +358,9 @@ ${e.password ? `密码为: ${e.password}` : ''}`;
               rows={8}
               maxLength={-1}
               isDisabled={!isOwner}
-              placeholder={'模型默认的 prompt 词，通过调整该内容，可以引导模型聊天方向。'}
+              placeholder={
+                '模型默认的 prompt 词，通过调整该内容，可以引导模型聊天方向。\n\n如果使用了知识库搜索，没有填写该内容时，系统会自动补充提示词；如果填写了内容，则以填写的内容为准。'
+              }
               {...register('chat.systemPrompt')}
             />
           </Box>

src/service/events/generateVector.ts

@@ -1,8 +1,8 @@
-import { openaiCreateEmbedding } from '../utils/chat/openai';
 import { getApiKey } from '../utils/auth';
 import { openaiError2 } from '../errorCode';
 import { PgClient } from '@/service/pg';
 import { getErrText } from '@/utils/tools';
+import { openaiEmbedding } from '@/pages/api/openapi/plugin/openaiEmbedding';
 
 export async function generateVector(next = false): Promise<any> {
   if (process.env.queueTask !== '1') {
@@ -42,24 +42,20 @@ export async function generateVector(next = false): Promise<any> {
     dataId = dataItem.id;
 
     // 获取 openapi Key
-    let userOpenAiKey;
     try {
-      const res = await getApiKey({ model: 'gpt-3.5-turbo', userId: dataItem.userId });
+      await getApiKey({ model: 'gpt-3.5-turbo', userId: dataItem.userId });
-      userOpenAiKey = res.userOpenAiKey;
     } catch (err: any) {
       await PgClient.delete('modelData', {
         where: [['id', dataId]]
       });
-      generateVector(true);
       getErrText(err, '获取 OpenAi Key 失败');
-      return;
+      return generateVector(true);
     }
 
     // 生成词向量
-    const { vectors } = await openaiCreateEmbedding({
+    const vectors = await openaiEmbedding({
-      textArr: [dataItem.q],
+      input: [dataItem.q],
-      userId: dataItem.userId,
+      userId: dataItem.userId
-      userOpenAiKey
     });
 
     // 更新 pg 向量和状态数据

src/service/models/chat.ts

@@ -47,10 +47,14 @@ const ChatSchema = new Schema({
           type: String,
           required: true
         },
-        systemPrompt: {
+        quote: {
-          type: String,
+          type: [{ id: String, q: String, a: String }],
-          default: ''
+          default: []
         }
+        // systemPrompt: {
+        //   type: String,
+        //   default: ''
+        // }
       }
     ],
     default: []

src/service/plugins/searchKb.ts

@@ -1,175 +0,0 @@
-import { PgClient } from '@/service/pg';
-import { ModelDataStatusEnum, ModelVectorSearchModeEnum, ChatModelMap } from '@/constants/model';
-import { ModelSchema } from '@/types/mongoSchema';
-import { openaiCreateEmbedding } from '../utils/chat/openai';
-import { ChatRoleEnum } from '@/constants/chat';
-import { modelToolMap } from '@/utils/chat';
-import { ChatItemSimpleType } from '@/types/chat';
-
-/**
- *  use openai embedding search kb
- */
-export const searchKb = async ({
-  userOpenAiKey,
-  prompts,
-  similarity = 0.2,
-  model,
-  userId
-}: {
-  userOpenAiKey?: string;
-  prompts: ChatItemSimpleType[];
-  model: ModelSchema;
-  userId: string;
-  similarity?: number;
-}): Promise<{
-  code: 200 | 201;
-  searchPrompts: {
-    obj: ChatRoleEnum;
-    value: string;
-  }[];
-}> => {
-  async function search(textArr: string[] = []) {
-    const limitMap: Record<ModelVectorSearchModeEnum, number> = {
-      [ModelVectorSearchModeEnum.hightSimilarity]: 15,
-      [ModelVectorSearchModeEnum.noContext]: 15,
-      [ModelVectorSearchModeEnum.lowSimilarity]: 20
-    };
-    // 获取提示词的向量
-    const { vectors: promptVectors } = await openaiCreateEmbedding({
-      userOpenAiKey,
-      userId,
-      textArr
-    });
-
-    const searchRes = await Promise.all(
-      promptVectors.map((promptVector) =>
-        PgClient.select<{ id: string; q: string; a: string }>('modelData', {
-          fields: ['id', 'q', 'a'],
-          where: [
-            ['status', ModelDataStatusEnum.ready],
-            'AND',
-            `kb_id IN (${model.chat.relatedKbs.map((item) => `'${item}'`).join(',')})`,
-            'AND',
-            `vector <=> '[${promptVector}]' < ${similarity}`
-          ],
-          order: [{ field: 'vector', mode: `<=> '[${promptVector}]'` }],
-          limit: limitMap[model.chat.searchMode]
-        }).then((res) => res.rows)
-      )
-    );
-
-    // Remove repeat record
-    const idSet = new Set<string>();
-    const filterSearch = searchRes.map((search) =>
-      search.filter((item) => {
-        if (idSet.has(item.id)) {
-          return false;
-        }
-        idSet.add(item.id);
-        return true;
-      })
-    );
-
-    return filterSearch.map((item) => item.map((item) => `${item.q}\n${item.a}`).join('\n'));
-  }
-  const modelConstantsData = ChatModelMap[model.chat.chatModel];
-
-  // search three times
-  const userPrompts = prompts.filter((item) => item.obj === 'Human');
-
-  const searchArr: string[] = [
-    userPrompts[userPrompts.length - 1].value,
-    userPrompts[userPrompts.length - 2]?.value
-  ].filter((item) => item);
-  const systemPrompts = await search(searchArr);
-
-  // filter system prompts.
-  const filterRateMap: Record<number, number[]> = {
-    1: [1],
-    2: [0.7, 0.3]
-  };
-  const filterRate = filterRateMap[systemPrompts.length] || filterRateMap[0];
-
-  // 计算固定提示词的 token 数量
-  const fixedPrompts = [
-    ...(model.chat.systemPrompt
-      ? [
-          {
-            obj: ChatRoleEnum.System,
-            value: model.chat.systemPrompt
-          }
-        ]
-      : []),
-    ...(model.chat.searchMode === ModelVectorSearchModeEnum.noContext
-      ? [
-          {
-            obj: ChatRoleEnum.System,
-            value: `知识库是关于"${model.name}"的内容,根据知识库内容回答问题.`
-          }
-        ]
-      : [
-          {
-            obj: ChatRoleEnum.System,
-            value: `玩一个问答游戏,规则为:
-1.你完全忘记你已有的知识
-2.你只回答关于"${model.name}"的问题
-3.你只从知识库中选择内容进行回答
-4.如果问题不在知识库中,你会回答:"我不知道。"
-请务必遵守规则`
-          }
-        ])
-  ];
-  const fixedSystemTokens = modelToolMap[model.chat.chatModel].countTokens({
-    messages: fixedPrompts
-  });
-  const maxTokens = modelConstantsData.systemMaxToken - fixedSystemTokens;
-
-  const filterSystemPrompt = filterRate
-    .map((rate, i) =>
-      modelToolMap[model.chat.chatModel].sliceText({
-        text: systemPrompts[i],
-        length: Math.floor(maxTokens * rate)
-      })
-    )
-    .join('\n')
-    .trim();
-
-  /* 高相似度+不回复 */
-  if (!filterSystemPrompt && model.chat.searchMode === ModelVectorSearchModeEnum.hightSimilarity) {
-    return {
-      code: 201,
-      searchPrompts: [
-        {
-          obj: ChatRoleEnum.System,
-          value: '对不起，你的问题不在知识库中。'
-        }
-      ]
-    };
-  }
-  /* 高相似度+无上下文，不添加额外知识,仅用系统提示词 */
-  if (!filterSystemPrompt && model.chat.searchMode === ModelVectorSearchModeEnum.noContext) {
-    return {
-      code: 200,
-      searchPrompts: model.chat.systemPrompt
-        ? [
-            {
-              obj: ChatRoleEnum.System,
-              value: model.chat.systemPrompt
-            }
-          ]
-        : []
-    };
-  }
-
-  /* 有匹配 */
-  return {
-    code: 200,
-    searchPrompts: [
-      {
-        obj: ChatRoleEnum.System,
-        value: `知识库:${filterSystemPrompt}`
-      },
-      ...fixedPrompts
-    ]
-  };
-};

src/service/utils/auth.ts

@@ -38,12 +38,14 @@ export const authUser = async ({
   req,
   authToken = false,
   authOpenApi = false,
-  authRoot = false
+  authRoot = false,
+  authBalance = false
 }: {
   req: NextApiRequest;
   authToken?: boolean;
   authOpenApi?: boolean;
   authRoot?: boolean;
+  authBalance?: boolean;
 }) => {
   const parseOpenApiKey = async (apiKey?: string) => {
     if (!apiKey) {
@@ -99,6 +101,17 @@ export const authUser = async ({
     return Promise.reject(ERROR_ENUM.unAuthorization);
   }
 
+  if (authBalance) {
+    const user = await User.findById(uid);
+    if (!user) {
+      return Promise.reject(ERROR_ENUM.unAuthorization);
+    }
+
+    if (!user.openaiKey && formatPrice(user.balance) <= 0) {
+      return Promise.reject(ERROR_ENUM.insufficientQuota);
+    }
+  }
+
   return {
     userId: uid
   };
@@ -226,7 +239,7 @@ export const authChat = async ({
   req
 }: {
   modelId: string;
-  chatId: '' | string;
+  chatId?: string;
   req: NextApiRequest;
 }) => {
   const { userId } = await authUser({ req, authToken: true });

src/service/utils/chat/claude.ts

@@ -1,17 +1,9 @@
 import { ChatCompletionType, StreamResponseType } from './index';
 import { ChatRoleEnum } from '@/constants/chat';
 import axios from 'axios';
-import mongoose from 'mongoose';
-import { NEW_CHATID_HEADER } from '@/constants/chat';
 
 /* 模型对话 */
-export const claudChat = async ({ apiKey, messages, stream, chatId, res }: ChatCompletionType) => {
+export const claudChat = async ({ apiKey, messages, stream, chatId }: ChatCompletionType) => {
-  const conversationId = chatId || String(new mongoose.Types.ObjectId());
-  // create a new chat
-  !chatId &&
-    messages.filter((item) => item.obj === 'Human').length === 1 &&
-    res?.setHeader(NEW_CHATID_HEADER, conversationId);
-
   // get system prompt
   const systemPrompt = messages
     .filter((item) => item.obj === 'System')
@@ -26,7 +18,7 @@ export const claudChat = async ({ apiKey, messages, stream, chatId, res }: ChatC
     {
       prompt,
       stream,
-      conversationId
+      conversationId: chatId
     },
     {
       headers: {
@@ -55,8 +47,7 @@ export const claudStreamResponse = async ({ res, chatResponse, prompts }: Stream
     try {
       const decoder = new TextDecoder();
       for await (const chunk of chatResponse.data as any) {
-        if (!res.writable) {
+        if (res.closed) {
-          // 流被中断了，直接忽略后面的内容
           break;
         }
         const content = decoder.decode(chunk);

src/service/utils/chat/index.ts

@@ -1,7 +1,7 @@
 import { ChatItemSimpleType } from '@/types/chat';
-import { modelToolMap } from '@/utils/chat';
+import { modelToolMap } from '@/utils/plugin';
 import type { ChatModelType } from '@/constants/model';
-import { ChatRoleEnum, SYSTEM_PROMPT_HEADER } from '@/constants/chat';
+import { ChatRoleEnum } from '@/constants/chat';
 import { OpenAiChatEnum, ClaudeEnum } from '@/constants/model';
 import { chatResponse, openAiStreamResponse } from './openai';
 import { claudChat, claudStreamResponse } from './claude';
@@ -11,6 +11,7 @@ export type ChatCompletionType = {
   apiKey: string;
   temperature: number;
   messages: ChatItemSimpleType[];
+  chatId?: string;
   [key: string]: any;
 };
 export type ChatCompletionResponseType = {
@@ -23,7 +24,6 @@ export type StreamResponseType = {
   chatResponse: any;
   prompts: ChatItemSimpleType[];
   res: NextApiResponse;
-  systemPrompt?: string;
   [key: string]: any;
 };
 export type StreamResponseReturnType = {
@@ -129,7 +129,6 @@ export const resStreamResponse = async ({
   model,
   res,
   chatResponse,
-  systemPrompt,
   prompts
 }: StreamResponseType & {
   model: ChatModelType;
@@ -139,18 +138,14 @@ export const resStreamResponse = async ({
   res.setHeader('Access-Control-Allow-Origin', '*');
   res.setHeader('X-Accel-Buffering', 'no');
   res.setHeader('Cache-Control', 'no-cache, no-transform');
-  systemPrompt && res.setHeader(SYSTEM_PROMPT_HEADER, encodeURIComponent(systemPrompt));
 
   const { responseContent, totalTokens, finishMessages } = await modelServiceToolMap[
     model
   ].streamResponse({
     chatResponse,
     prompts,
-    res,
+    res
-    systemPrompt
   });
 
-  res.end();
-
   return { responseContent, totalTokens, finishMessages };
 };

src/service/utils/chat/openai.ts

@@ -1,13 +1,11 @@
 import { Configuration, OpenAIApi } from 'openai';
 import { createParser, ParsedEvent, ReconnectInterval } from 'eventsource-parser';
 import { axiosConfig } from '../tools';
-import { ChatModelMap, embeddingModel, OpenAiChatEnum } from '@/constants/model';
+import { ChatModelMap, OpenAiChatEnum } from '@/constants/model';
-import { pushGenerateVectorBill } from '../../events/pushBill';
+import { adaptChatItem_openAI } from '@/utils/plugin/openai';
-import { adaptChatItem_openAI } from '@/utils/chat/openai';
+import { modelToolMap } from '@/utils/plugin';
-import { modelToolMap } from '@/utils/chat';
 import { ChatCompletionType, ChatContextFilter, StreamResponseType } from './index';
 import { ChatRoleEnum } from '@/constants/chat';
-import { getSystemOpenAiKey } from '../auth';
 
 export const getOpenAIApi = () =>
   new OpenAIApi(
@@ -16,51 +14,6 @@ export const getOpenAIApi = () =>
     })
   );
 
-/* 获取向量 */
-export const openaiCreateEmbedding = async ({
-  userOpenAiKey,
-  userId,
-  textArr
-}: {
-  userOpenAiKey?: string;
-  userId: string;
-  textArr: string[];
-}) => {
-  const systemAuthKey = getSystemOpenAiKey();
-
-  // 获取 chatAPI
-  const chatAPI = getOpenAIApi();
-
-  // 把输入的内容转成向量
-  const res = await chatAPI
-    .createEmbedding(
-      {
-        model: embeddingModel,
-        input: textArr
-      },
-      {
-        timeout: 60000,
-        ...axiosConfig(userOpenAiKey || systemAuthKey)
-      }
-    )
-    .then((res) => ({
-      tokenLen: res.data.usage.total_tokens || 0,
-      vectors: res.data.data.map((item) => item.embedding)
-    }));
-
-  pushGenerateVectorBill({
-    isPay: !userOpenAiKey,
-    userId,
-    text: textArr.join(''),
-    tokenLen: res.tokenLen
-  });
-
-  return {
-    vectors: res.vectors,
-    chatAPI
-  };
-};
-
 /* 模型对话 */
 export const chatResponse = async ({
   model,
@@ -127,7 +80,7 @@ export const openAiStreamResponse = async ({
         const content: string = json?.choices?.[0].delta.content || '';
         responseContent += content;
 
-        res.writable && content && res.write(content);
+        !res.closed && content && res.write(content);
       } catch (error) {
         error;
       }
@@ -137,8 +90,7 @@ export const openAiStreamResponse = async ({
       const decoder = new TextDecoder();
       const parser = createParser(onParse);
       for await (const chunk of chatResponse.data as any) {
-        if (!res.writable) {
+        if (res.closed) {
-          // 流被中断了，直接忽略后面的内容
           break;
         }
         parser.feed(decoder.decode(chunk, { stream: true }));

src/types/chat.d.ts

@@ -1,12 +1,13 @@
 import { ChatRoleEnum } from '@/constants/chat';
 import type { InitChatResponse, InitShareChatResponse } from '@/api/response/chat';
+import { QuoteItemType } from '@/pages/api/openapi/kb/appKbSearch';
 
 export type ExportChatType = 'md' | 'pdf' | 'html';
 
 export type ChatItemSimpleType = {
   obj: `${ChatRoleEnum}`;
   value: string;
-  systemPrompt?: string;
+  quote?: QuoteItemType[];
 };
 export type ChatItemType = {
   _id: string;

src/utils/chat/claude.ts

@@ -1,3 +0,0 @@
-export const ClaudeSliceTextByToken = ({ text, length }: { text: string; length: number }) => {
-  return text.slice(0, length);
-};

src/utils/file.ts

@@ -1,6 +1,6 @@
 import mammoth from 'mammoth';
 import Papa from 'papaparse';
-import { getOpenAiEncMap } from './chat/openai';
+import { getOpenAiEncMap } from './plugin/openai';
 
 /**
  * 读取 txt 文件内容

src/utils/plugin/index.ts

@@ -2,29 +2,37 @@ import { ClaudeEnum, OpenAiChatEnum } from '@/constants/model';
 import type { ChatModelType } from '@/constants/model';
 import type { ChatItemSimpleType } from '@/types/chat';
 import { countOpenAIToken, openAiSliceTextByToken } from './openai';
-import { ClaudeSliceTextByToken } from './claude';
+import { gpt_chatItemTokenSlice } from '@/pages/api/openapi/text/gptMessagesSlice';
 
 export const modelToolMap: Record<
   ChatModelType,
   {
     countTokens: (data: { messages: ChatItemSimpleType[] }) => number;
     sliceText: (data: { text: string; length: number }) => string;
+    tokenSlice: (data: {
+      messages: ChatItemSimpleType[];
+      maxToken: number;
+    }) => ChatItemSimpleType[];
   }
 > = {
   [OpenAiChatEnum.GPT35]: {
     countTokens: ({ messages }) => countOpenAIToken({ model: OpenAiChatEnum.GPT35, messages }),
-    sliceText: (data) => openAiSliceTextByToken({ model: OpenAiChatEnum.GPT35, ...data })
+    sliceText: (data) => openAiSliceTextByToken({ model: OpenAiChatEnum.GPT35, ...data }),
+    tokenSlice: (data) => gpt_chatItemTokenSlice({ model: OpenAiChatEnum.GPT35, ...data })
   },
   [OpenAiChatEnum.GPT4]: {
     countTokens: ({ messages }) => countOpenAIToken({ model: OpenAiChatEnum.GPT4, messages }),
-    sliceText: (data) => openAiSliceTextByToken({ model: OpenAiChatEnum.GPT4, ...data })
+    sliceText: (data) => openAiSliceTextByToken({ model: OpenAiChatEnum.GPT4, ...data }),
+    tokenSlice: (data) => gpt_chatItemTokenSlice({ model: OpenAiChatEnum.GPT4, ...data })
   },
   [OpenAiChatEnum.GPT432k]: {
     countTokens: ({ messages }) => countOpenAIToken({ model: OpenAiChatEnum.GPT432k, messages }),
-    sliceText: (data) => openAiSliceTextByToken({ model: OpenAiChatEnum.GPT432k, ...data })
+    sliceText: (data) => openAiSliceTextByToken({ model: OpenAiChatEnum.GPT432k, ...data }),
+    tokenSlice: (data) => gpt_chatItemTokenSlice({ model: OpenAiChatEnum.GPT432k, ...data })
   },
   [ClaudeEnum.Claude]: {
     countTokens: ({ messages }) => countOpenAIToken({ model: OpenAiChatEnum.GPT35, messages }),
-    sliceText: (data) => openAiSliceTextByToken({ model: OpenAiChatEnum.GPT35, ...data })
+    sliceText: (data) => openAiSliceTextByToken({ model: OpenAiChatEnum.GPT35, ...data }),
+    tokenSlice: (data) => gpt_chatItemTokenSlice({ model: OpenAiChatEnum.GPT35, ...data })
   }
 };

src/utils/tools.ts

@@ -126,3 +126,10 @@ export const getErrText = (err: any, def = '') => {
   msg && console.log('error =>', msg);
   return msg;
 };
+
+export const delay = (ms: number) =>
+  new Promise((resolve) => {
+    setTimeout(() => {
+      resolve('');
+    }, ms);
+  });