fix: v1 api

README.md

@@ -44,9 +44,10 @@ Fast GPT 允许你使用自己的 openai API KEY 来快速的调用 openai 接
 
 ## Powered by
 
-- [TuShan 5 分钟搭建后台管理系统](https://github.com/msgbyte/tushan)
-- [Laf 3 分钟快速接入三方应用](https://github.com/labring/laf)
-- [Sealos 快速部署集群应用](https://github.com/labring/sealos)
+- [TuShan: 5 分钟搭建后台管理系统](https://github.com/msgbyte/tushan)
+- [Laf: 3 分钟快速接入三方应用](https://github.com/labring/laf)
+- [Sealos: 快速部署集群应用](https://github.com/labring/sealos)
+- [One API: 令牌管理 & 二次分发，支持 Azure](https://github.com/songquanpeng/one-api)
 
 ## 🌟 Star History
 

client/.env.template

@@ -17,10 +17,12 @@ aliTemplateCode=xxxx
 TOKEN_KEY=dfdasfdas
 # root key, 最高权限
 ROOT_KEY=fdafasd
-# openai
-# OPENAI_BASE_URL=http://ai.openai.com/v1
-# OPENAI_BASE_URL_AUTH=可选安全凭证，会放到 header.auth 里
-OPENAIKEY=sk-xxx
+# 使用 oneapi 
+# ONEAPI_URL=https://xxxx.cloud.sealos.io/v1
+# ONEAPI_KEY=sk-xxxx
+# openai 的基本地址（国外的可以忽略，默认走 api.openai.com）。不用 oneapi 的话需要下面 2 个参数，用户的 key 也会走下面的参数
+OPENAI_BASE_URL=https://xxxx.cloud.sealos.io/openai/v1
+OPENAIKEY=sk-xxxx
 # db
 MONGODB_URI=mongodb://username:password@0.0.0.0:27017/?authSource=admin
 MONGODB_NAME=fastgpt

client/package.json

@@ -26,7 +26,6 @@
     "crypto": "^1.0.1",
     "date-fns": "^2.30.0",
     "dayjs": "^1.11.7",
-    "eventsource-parser": "^0.1.0",
     "formidable": "^2.1.1",
     "framer-motion": "^9.0.6",
     "hyperdown": "^2.4.29",

client/pnpm-lock.yaml

@@ -56,9 +56,6 @@ dependencies:
   dayjs:
     specifier: ^1.11.7
     version: registry.npmmirror.com/dayjs@1.11.7
-  eventsource-parser:
-    specifier: ^0.1.0
-    version: registry.npmmirror.com/eventsource-parser@0.1.0
   formidable:
     specifier: ^2.1.1
     version: registry.npmmirror.com/formidable@2.1.1
@@ -7510,13 +7507,6 @@ packages:
     version: 2.0.3
     engines: {node: '>=0.10.0'}
 
-  registry.npmmirror.com/eventsource-parser@0.1.0:
-    resolution: {integrity: sha512-M9QjFtEIkwytUarnx113HGmgtk52LSn3jNAtnWKi3V+b9rqSfQeVdLsaD5AG/O4IrGQwmAAHBIsqbmURPTd2rA==, registry: https://registry.npm.taobao.org/, tarball: https://registry.npmmirror.com/eventsource-parser/-/eventsource-parser-0.1.0.tgz}
-    name: eventsource-parser
-    version: 0.1.0
-    engines: {node: '>=14.18'}
-    dev: false
-
   registry.npmmirror.com/execa@5.1.1:
     resolution: {integrity: sha512-8uSpZZocAZRBAPIEINJj3Lo9HyGitllczc27Eh5YYojjMFMn8yHMDMaUHE2Jqfq05D/wucwI4JGURyXt1vchyg==, registry: https://registry.npm.taobao.org/, tarball: https://registry.npmmirror.com/execa/-/execa-5.1.1.tgz}
     name: execa

client/src/api/fetch.ts

@@ -1,6 +1,7 @@
 import { Props, ChatResponseType } from '@/pages/api/openapi/v1/chat/completions';
 import { sseResponseEventEnum } from '@/constants/chat';
 import { getErrText } from '@/utils/tools';
+import { parseStreamChunk } from '@/utils/adapt';
 
 interface StreamFetchProps {
   data: Props;
@@ -8,103 +9,95 @@ interface StreamFetchProps {
   abortSignal: AbortController;
 }
 export const streamFetch = ({ data, onMessage, abortSignal }: StreamFetchProps) =>
-  new Promise<ChatResponseType & { responseText: string }>(async (resolve, reject) => {
-    try {
-      const response = await window.fetch('/api/openapi/v1/chat/completions', {
-        method: 'POST',
-        headers: {
-          'Content-Type': 'application/json'
-        },
-        signal: abortSignal.signal,
-        body: JSON.stringify({
-          ...data,
-          stream: true
-        })
-      });
+  new Promise<ChatResponseType & { responseText: string; errMsg: string }>(
+    async (resolve, reject) => {
+      try {
+        const response = await window.fetch('/api/openapi/v1/chat/completions', {
+          method: 'POST',
+          headers: {
+            'Content-Type': 'application/json'
+          },
+          signal: abortSignal.signal,
+          body: JSON.stringify({
+            ...data,
+            stream: true
+          })
+        });
 
-      if (response.status !== 200) {
-        const err = await response.json();
-        return reject(err);
-      }
+        if (response.status !== 200) {
+          const err = await response.json();
+          return reject(err);
+        }
 
-      if (!response?.body) {
-        throw new Error('Request Error');
-      }
+        if (!response?.body) {
+          throw new Error('Request Error');
+        }
 
-      const reader = response.body?.getReader();
-      const decoder = new TextDecoder('utf-8');
+        const reader = response.body?.getReader();
 
-      // response data
-      let responseText = '';
-      let newChatId = '';
-      let quoteLen = 0;
+        // response data
+        let responseText = '';
+        let newChatId = '';
+        let quoteLen = 0;
+        let errMsg = '';
 
-      const read = async () => {
-        try {
-          const { done, value } = await reader.read();
-          if (done) {
-            if (response.status === 200) {
+        const read = async () => {
+          try {
+            const { done, value } = await reader.read();
+            if (done) {
+              if (response.status === 200) {
+                return resolve({
+                  responseText,
+                  newChatId,
+                  quoteLen,
+                  errMsg
+                });
+              } else {
+                return reject('响应过程出现异常~');
+              }
+            }
+            const chunkResponse = parseStreamChunk(value);
+
+            chunkResponse.forEach((item) => {
+              // parse json data
+              const data = (() => {
+                try {
+                  return JSON.parse(item.data);
+                } catch (error) {
+                  return item.data;
+                }
+              })();
+
+              if (item.event === sseResponseEventEnum.answer && data !== '[DONE]') {
+                const answer: string = data?.choices?.[0].delta.content || '';
+                onMessage(answer);
+                responseText += answer;
+              } else if (item.event === sseResponseEventEnum.chatResponse) {
+                const chatResponse = data as ChatResponseType;
+                newChatId = chatResponse.newChatId;
+                quoteLen = chatResponse.quoteLen || 0;
+              } else if (item.event === sseResponseEventEnum.error) {
+                errMsg = getErrText(data, '流响应错误');
+              }
+            });
+            read();
+          } catch (err: any) {
+            if (err?.message === 'The user aborted a request.') {
               return resolve({
                 responseText,
                 newChatId,
-                quoteLen
+                quoteLen,
+                errMsg
               });
-            } else {
-              return reject('响应过程出现异常~');
             }
+            reject(getErrText(err, '请求异常'));
           }
-          const chunk = decoder.decode(value);
-          const chunkLines = chunk.split('\n\n').filter((item) => item);
-          const chunkResponse = chunkLines.map((item) => {
-            const splitEvent = item.split('\n');
-            if (splitEvent.length === 2) {
-              return {
-                event: splitEvent[0].replace('event: ', ''),
-                data: splitEvent[1].replace('data: ', '')
-              };
-            }
-            return {
-              event: '',
-              data: splitEvent[0].replace('data: ', '')
-            };
-          });
+        };
+        read();
+      } catch (err: any) {
+        console.log(err);
 
-          chunkResponse.forEach((item) => {
-            // parse json data
-            const data = (() => {
-              try {
-                return JSON.parse(item.data);
-              } catch (error) {
-                return item.data;
-              }
-            })();
-
-            if (item.event === sseResponseEventEnum.answer && data !== '[DONE]') {
-              const answer: string = data?.choices?.[0].delta.content || '';
-              onMessage(answer);
-              responseText += answer;
-            } else if (item.event === sseResponseEventEnum.chatResponse) {
-              const chatResponse = data as ChatResponseType;
-              newChatId = chatResponse.newChatId;
-              quoteLen = chatResponse.quoteLen || 0;
-            }
-          });
-          read();
-        } catch (err: any) {
-          if (err?.message === 'The user aborted a request.') {
-            return resolve({
-              responseText,
-              newChatId,
-              quoteLen
-            });
-          }
-          reject(getErrText(err, '请求异常'));
-        }
-      };
-      read();
-    } catch (err: any) {
-      console.log(err);
-
-      reject(getErrText(err, '请求异常'));
+        reject(getErrText(err, '请求异常'));
+      }
     }
-  });
+  );

client/src/api/response/chat.d.ts

@@ -5,6 +5,7 @@ export interface InitChatResponse {
   chatId: string;
   modelId: string;
   systemPrompt?: string;
+  limitPrompt?: string;
   model: {
     name: string;
     avatar: string;

client/src/components/Select/index.tsx

@@ -26,7 +26,7 @@ const MySelect = ({ placeholder, value, width = 'auto', list, onchange, ...props
 
   return (
     <Menu autoSelect={false} onOpen={onOpen} onClose={onClose}>
-      <MenuButton as={'span'}>
+      <MenuButton style={{ width: '100%' }} as={'span'}>
         <Button
           width={width}
           px={3}

client/src/constants/chat.ts

@@ -1,4 +1,5 @@
 export enum sseResponseEventEnum {
+  error = 'error',
   answer = 'answer',
   chatResponse = 'chatResponse'
 }

client/src/constants/model.ts

@@ -82,6 +82,7 @@ export const defaultModel: ModelSchema = {
     searchLimit: 5,
     searchEmptyText: '',
     systemPrompt: '',
+    limitPrompt: '',
     temperature: 0,
     maxToken: 4000,
     chatModel: OpenAiChatEnum.GPT35

client/src/pages/api/chat/init.ts

@@ -101,6 +101,7 @@ export default async function handler(req: NextApiRequest, res: NextApiResponse)
         },
         chatModel: model.chat.chatModel,
         systemPrompt: isOwner ? model.chat.systemPrompt : '',
+        limitPrompt: isOwner ? model.chat.limitPrompt : '',
         history
       }
     });

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

@@ -65,10 +65,14 @@ export default withNextCors(async function handler(req: NextApiRequest, res: Nex
     const modelConstantsData = ChatModelMap[model.chat.chatModel];
     const prompt = prompts[prompts.length - 1];
 
-    const { userSystemPrompt = [], quotePrompt = [] } = await (async () => {
+    const {
+      userSystemPrompt = [],
+      userLimitPrompt = [],
+      quotePrompt = []
+    } = await (async () => {
       // 使用了知识库搜索
       if (model.chat.relatedKbs?.length > 0) {
-        const { userSystemPrompt, quotePrompt } = await appKbSearch({
+        const { quotePrompt, userSystemPrompt, userLimitPrompt } = await appKbSearch({
           model,
           userId,
           fixedQuote: [],
@@ -78,21 +82,29 @@ export default withNextCors(async function handler(req: NextApiRequest, res: Nex
         });
 
         return {
-          userSystemPrompt: userSystemPrompt ? [userSystemPrompt] : [],
+          userSystemPrompt,
+          userLimitPrompt,
           quotePrompt: [quotePrompt]
         };
       }
-      if (model.chat.systemPrompt) {
-        return {
-          userSystemPrompt: [
-            {
-              obj: ChatRoleEnum.System,
-              value: model.chat.systemPrompt
-            }
-          ]
-        };
-      }
-      return {};
+      return {
+        userSystemPrompt: model.chat.systemPrompt
+          ? [
+              {
+                obj: ChatRoleEnum.System,
+                value: model.chat.systemPrompt
+              }
+            ]
+          : [],
+        userLimitPrompt: model.chat.limitPrompt
+          ? [
+              {
+                obj: ChatRoleEnum.Human,
+                value: model.chat.limitPrompt
+              }
+            ]
+          : []
+      };
     })();
 
     // search result is empty
@@ -102,7 +114,13 @@ export default withNextCors(async function handler(req: NextApiRequest, res: Nex
     }
 
     // 读取对话内容
-    const completePrompts = [...quotePrompt, ...prompts.slice(0, -1), ...userSystemPrompt, prompt];
+    const completePrompts = [
+      ...quotePrompt,
+      ...userSystemPrompt,
+      ...prompts.slice(0, -1),
+      ...userLimitPrompt,
+      prompt
+    ];
 
     // 计算温度
     const temperature = (modelConstantsData.maxTemperature * (model.chat.temperature / 10)).toFixed(

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

@@ -28,7 +28,11 @@ type Response = {
   userSystemPrompt: {
     obj: ChatRoleEnum;
     value: string;
-  };
+  }[];
+  userLimitPrompt: {
+    obj: ChatRoleEnum;
+    value: string;
+  }[];
   quotePrompt: {
     obj: ChatRoleEnum;
     value: string;
@@ -130,17 +134,24 @@ export async function appKbSearch({
 
   // 计算固定提示词的 token 数量
   const userSystemPrompt = model.chat.systemPrompt // user system prompt
-    ? {
-        obj: ChatRoleEnum.Human,
-        value: model.chat.systemPrompt
-      }
-    : {
-        obj: ChatRoleEnum.Human,
-        value: `知识库是关于 ${model.name} 的内容，参考知识库回答问题。与 "${model.name}" 无关内容，直接回复: "我不知道"。`
-      };
+    ? [
+        {
+          obj: ChatRoleEnum.System,
+          value: model.chat.systemPrompt
+        }
+      ]
+    : [];
+  const userLimitPrompt = [
+    {
+      obj: ChatRoleEnum.Human,
+      value: model.chat.limitPrompt
+        ? model.chat.limitPrompt
+        : `知识库是关于 ${model.name} 的内容，参考知识库回答问题。与 "${model.name}" 无关内容，直接回复: "我不知道"。`
+    }
+  ];
 
   const fixedSystemTokens = modelToolMap[model.chat.chatModel].countTokens({
-    messages: [userSystemPrompt]
+    messages: [...userSystemPrompt, ...userLimitPrompt]
   });
 
   // filter part quote by maxToken
@@ -164,6 +175,7 @@ export async function appKbSearch({
   return {
     rawSearch,
     userSystemPrompt,
+    userLimitPrompt,
     quotePrompt: {
       obj: ChatRoleEnum.System,
       value: quoteText

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

@@ -1,6 +1,6 @@
 import type { NextApiRequest, NextApiResponse } from 'next';
 import { jsonRes } from '@/service/response';
-import { authUser, getApiKey } from '@/service/utils/auth';
+import { authUser, getApiKey, getSystemOpenAiKey } from '@/service/utils/auth';
 import { withNextCors } from '@/service/utils/tools';
 import { getOpenAIApi } from '@/service/utils/chat/openai';
 import { embeddingModel } from '@/constants/model';
@@ -39,14 +39,10 @@ export async function openaiEmbedding({
   input,
   mustPay = false
 }: { userId: string; mustPay?: boolean } & Props) {
-  const { userOpenAiKey, systemAuthKey } = await getApiKey({
-    model: OpenAiChatEnum.GPT35,
-    userId,
-    mustPay
-  });
+  const apiKey = getSystemOpenAiKey();
 
   // 获取 chatAPI
-  const chatAPI = getOpenAIApi();
+  const chatAPI = getOpenAIApi(apiKey);
 
   // 把输入的内容转成向量
   const result = await chatAPI
@@ -57,16 +53,22 @@ export async function openaiEmbedding({
       },
       {
         timeout: 60000,
-        ...axiosConfig(userOpenAiKey || systemAuthKey)
+        ...axiosConfig(apiKey)
       }
     )
-    .then((res) => ({
-      tokenLen: res.data?.usage?.total_tokens || 0,
-      vectors: res.data.data.map((item) => item.embedding)
-    }));
+    .then((res) => {
+      if (!res.data?.usage?.total_tokens) {
+        // @ts-ignore
+        return Promise.reject(res.data?.error?.message || 'Embedding Error');
+      }
+      return {
+        tokenLen: res.data.usage.total_tokens || 0,
+        vectors: res.data.data.map((item) => item.embedding)
+      };
+    });
 
   pushGenerateVectorBill({
-    isPay: !userOpenAiKey,
+    isPay: mustPay,
     userId,
     text: input.join(''),
     tokenLen: result.tokenLen

client/src/pages/api/openapi/v1/chat/completions.ts

@@ -14,9 +14,9 @@ import { gptMessage2ChatType, textAdaptGptResponse } from '@/utils/adapt';
 import { getChatHistory } from './getHistory';
 import { saveChat } from '@/pages/api/chat/saveChat';
 import { sseResponse } from '@/service/utils/tools';
-import { getErrText } from '@/utils/tools';
 import { type ChatCompletionRequestMessage } from 'openai';
 import { Types } from 'mongoose';
+import { sensitiveCheck } from '../../text/sensitiveCheck';
 
 export type MessageItemType = ChatCompletionRequestMessage & { _id?: string };
 type FastGptWebChatProps = {
@@ -108,11 +108,12 @@ export default withNextCors(async function handler(req: NextApiRequest, res: Nex
     const {
       rawSearch = [],
       userSystemPrompt = [],
+      userLimitPrompt = [],
       quotePrompt = []
     } = await (async () => {
       // 使用了知识库搜索
       if (model.chat.relatedKbs?.length > 0) {
-        const { rawSearch, userSystemPrompt, quotePrompt } = await appKbSearch({
+        const { rawSearch, quotePrompt, userSystemPrompt, userLimitPrompt } = await appKbSearch({
           model,
           userId,
           fixedQuote: history[history.length - 1]?.quote,
@@ -123,21 +124,29 @@ export default withNextCors(async function handler(req: NextApiRequest, res: Nex
 
         return {
           rawSearch,
-          userSystemPrompt: userSystemPrompt ? [userSystemPrompt] : [],
+          userSystemPrompt,
+          userLimitPrompt,
           quotePrompt: [quotePrompt]
         };
       }
-      if (model.chat.systemPrompt) {
-        return {
-          userSystemPrompt: [
-            {
-              obj: ChatRoleEnum.System,
-              value: model.chat.systemPrompt
-            }
-          ]
-        };
-      }
-      return {};
+      return {
+        userSystemPrompt: model.chat.systemPrompt
+          ? [
+              {
+                obj: ChatRoleEnum.System,
+                value: model.chat.systemPrompt
+              }
+            ]
+          : [],
+        userLimitPrompt: model.chat.limitPrompt
+          ? [
+              {
+                obj: ChatRoleEnum.Human,
+                value: model.chat.limitPrompt
+              }
+            ]
+          : []
+      };
     })();
 
     // search result is empty
@@ -167,7 +176,13 @@ export default withNextCors(async function handler(req: NextApiRequest, res: Nex
     }
 
     // api messages. [quote,context,systemPrompt,question]
-    const completePrompts = [...quotePrompt, ...prompts.slice(0, -1), ...userSystemPrompt, prompt];
+    const completePrompts = [
+      ...quotePrompt,
+      ...userSystemPrompt,
+      ...prompts.slice(0, -1),
+      ...userLimitPrompt,
+      prompt
+    ];
     // chat temperature
     const modelConstantsData = ChatModelMap[model.chat.chatModel];
     // FastGpt temperature range: 1~10
@@ -175,6 +190,10 @@ export default withNextCors(async function handler(req: NextApiRequest, res: Nex
       2
     );
 
+    await sensitiveCheck({
+      input: `${userSystemPrompt[0]?.value}\n${userLimitPrompt[0]?.value}\n${prompt.value}`
+    });
+
     // start model api. responseText and totalTokens: valid only if stream = false
     const { streamResponse, responseMessages, responseText, totalTokens } =
       await modelServiceToolMap[model.chat.chatModel].chatCompletion({
@@ -231,8 +250,7 @@ export default withNextCors(async function handler(req: NextApiRequest, res: Nex
             tokens: totalTokens
           };
         } catch (error) {
-          console.log('stream response error', error);
-          return {};
+          return Promise.reject(error);
         }
       } else {
         return {
@@ -256,7 +274,7 @@ export default withNextCors(async function handler(req: NextApiRequest, res: Nex
             ...(showAppDetail
               ? {
                   quote: rawSearch,
-                  systemPrompt: userSystemPrompt?.[0]?.value
+                  systemPrompt: `${userSystemPrompt[0]?.value}\n\n${userLimitPrompt[0]?.value}`
                 }
               : {})
           }
@@ -301,7 +319,12 @@ export default withNextCors(async function handler(req: NextApiRequest, res: Nex
   } catch (err: any) {
     res.status(500);
     if (step === 1) {
-      res.end(getErrText(err, 'Stream response error'));
+      sseResponse({
+        res,
+        event: sseResponseEventEnum.error,
+        data: JSON.stringify(err)
+      });
+      res.end();
     } else {
       jsonRes(res, {
         code: 500,

client/src/pages/api/system/getModels.ts

@@ -7,11 +7,9 @@ import { ChatModelMap, OpenAiChatEnum } from '@/constants/model';
 export default async function handler(req: NextApiRequest, res: NextApiResponse) {
   const chatModelList: ChatModelItemType[] = [];
 
-  if (process.env.OPENAIKEY) {
-    chatModelList.push(ChatModelMap[OpenAiChatEnum.GPT3516k]);
-    chatModelList.push(ChatModelMap[OpenAiChatEnum.GPT35]);
-    chatModelList.push(ChatModelMap[OpenAiChatEnum.GPT4]);
-  }
+  chatModelList.push(ChatModelMap[OpenAiChatEnum.GPT3516k]);
+  chatModelList.push(ChatModelMap[OpenAiChatEnum.GPT35]);
+  chatModelList.push(ChatModelMap[OpenAiChatEnum.GPT4]);
 
   jsonRes(res, {
     data: chatModelList

client/src/pages/chat/index.tsx

@@ -174,7 +174,7 @@ const Chat = ({ modelId, chatId }: { modelId: string; chatId: string }) => {
       const messages = adaptChatItem_openAI({ messages: prompts, reserveId: true });
 
       // 流请求，获取数据
-      const { newChatId, quoteLen } = await streamFetch({
+      const { newChatId, quoteLen, errMsg } = await streamFetch({
         data: {
           messages,
           chatId,
@@ -219,7 +219,9 @@ const Chat = ({ modelId, chatId }: { modelId: string; chatId: string }) => {
             ...item,
             status: 'finish',
             quoteLen,
-            systemPrompt: chatData.systemPrompt
+            systemPrompt: `${chatData.systemPrompt}${`${
+              chatData.limitPrompt ? `\n\n${chatData.limitPrompt}` : ''
+            }`}`
           };
         })
       }));
@@ -230,17 +232,26 @@ const Chat = ({ modelId, chatId }: { modelId: string; chatId: string }) => {
         loadHistory({ pageNum: 1, init: true });
         loadMyModels(true);
       }, 100);
+
+      if (errMsg) {
+        toast({
+          status: 'warning',
+          title: errMsg
+        });
+      }
     },
     [
       chatId,
       modelId,
-      chatData.systemPrompt,
       setChatData,
-      loadHistory,
-      loadMyModels,
       generatingMessage,
       setForbidLoadChatData,
-      router
+      router,
+      chatData.systemPrompt,
+      chatData.limitPrompt,
+      loadHistory,
+      loadMyModels,
+      toast
     ]
   );
 
@@ -749,7 +760,7 @@ const Chat = ({ modelId, chatId }: { modelId: string; chatId: string }) => {
                                 px={[2, 4]}
                                 onClick={() => setShowSystemPrompt(item.systemPrompt || '')}
                               >
-                                提示词
+                                提示词 & 限定词
                               </Button>
                             )}
                             {!!item.quoteLen && (

client/src/pages/kb/components/Test.tsx

@@ -42,6 +42,12 @@ const Test = () => {
       pushKbTestItem(testItem);
       setInputText('');
       setKbTestItem(testItem);
+    },
+    onError(err) {
+      toast({
+        title: getErrText(err),
+        status: 'error'
+      });
     }
   });
 

client/src/pages/model/components/detail/components/Settings.tsx

@@ -1,5 +1,15 @@
 import React, { useCallback, useState, useMemo } from 'react';
-import { Box, Flex, Button, FormControl, Input, Textarea, Divider } from '@chakra-ui/react';
+import {
+  Box,
+  Flex,
+  Button,
+  FormControl,
+  Input,
+  Textarea,
+  Divider,
+  Tooltip
+} from '@chakra-ui/react';
+import { QuestionOutlineIcon } from '@chakra-ui/icons';
 import { useQuery } from '@tanstack/react-query';
 import { useForm } from 'react-hook-form';
 import { useRouter } from 'next/router';
@@ -20,6 +30,11 @@ import Avatar from '@/components/Avatar';
 import MySelect from '@/components/Select';
 import MySlider from '@/components/Slider';
 
+const systemPromptTip =
+  '模型固定的引导词，通过调整该内容，可以引导模型聊天方向。该内容会被固定在上下文的开头。';
+const limitPromptTip =
+  '限定模型对话范围，会被放置在本次提问前，拥有强引导和限定性。例如:\n1. 知识库是关于 Laf 的介绍，参考知识库回答问题，与 "Laf" 无关内容，直接回复: "我不知道"。\n2. 你仅回答关于 "xxx" 的问题，其他问题回复: "xxxx"';
+
 const Settings = ({ modelId }: { modelId: string }) => {
   const { toast } = useToast();
   const router = useRouter();
@@ -60,7 +75,7 @@ const Settings = ({ modelId }: { modelId: string }) => {
     }
 
     return max;
-  }, [getValues, setValue, refresh]);
+  }, [getValues, setValue]);
 
   // 提交保存模型修改
   const saveSubmitSuccess = useCallback(
@@ -211,7 +226,7 @@ const Settings = ({ modelId }: { modelId: string }) => {
           介绍
         </Box>
         <Textarea
-          rows={5}
+          rows={4}
           maxLength={500}
           placeholder={'给你的 AI 应用一个介绍'}
           {...register('intro')}
@@ -225,11 +240,14 @@ const Settings = ({ modelId }: { modelId: string }) => {
           对话模型
         </Box>
         <MySelect
-          width={['200px', '240px']}
+          width={['100%', '280px']}
           value={getValues('chat.chatModel')}
           list={chatModelList.map((item) => ({
             id: item.chatModel,
-            label: item.name
+            label: `${item.name} (${formatPrice(
+              ChatModelMap[getValues('chat.chatModel')]?.price,
+              1000
+            )} 元/1k tokens)`
           }))}
           onchange={(val: any) => {
             setValue('chat.chatModel', val);
@@ -237,15 +255,6 @@ const Settings = ({ modelId }: { modelId: string }) => {
           }}
         />
       </Flex>
-      <Flex alignItems={'center'} mt={5}>
-        <Box w={['60px', '100px', '140px']} flexShrink={0}>
-          价格
-        </Box>
-        <Box fontSize={['sm', 'md']}>
-          {formatPrice(ChatModelMap[getValues('chat.chatModel')]?.price, 1000)}
-          元/1K tokens(包括上下文和回答)
-        </Box>
-      </Flex>
       <Flex alignItems={'center'} my={10}>
         <Box w={['60px', '100px', '140px']} flexShrink={0}>
           温度
@@ -269,7 +278,7 @@ const Settings = ({ modelId }: { modelId: string }) => {
       </Flex>
       <Flex alignItems={'center'} mt={12} mb={10}>
         <Box w={['60px', '100px', '140px']} flexShrink={0}>
-          最大长度
+          回复上限
         </Box>
         <Box flex={1} ml={'10px'}>
           <MySlider
@@ -292,15 +301,29 @@ const Settings = ({ modelId }: { modelId: string }) => {
       <Flex mt={10} alignItems={'flex-start'}>
         <Box w={['60px', '100px', '140px']} flexShrink={0}>
           提示词
+          <Tooltip label={systemPromptTip}>
+            <QuestionOutlineIcon display={['none', 'inline']} ml={1} />
+          </Tooltip>
         </Box>
         <Textarea
           rows={8}
-          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。'
-          }
+          placeholder={systemPromptTip}
           {...register('chat.systemPrompt')}
         ></Textarea>
       </Flex>
+      <Flex mt={5} alignItems={'flex-start'}>
+        <Box w={['60px', '100px', '140px']} flexShrink={0}>
+          限定词
+          <Tooltip label={limitPromptTip}>
+            <QuestionOutlineIcon display={['none', 'inline']} ml={1} />
+          </Tooltip>
+        </Box>
+        <Textarea
+          rows={5}
+          placeholder={limitPromptTip}
+          {...register('chat.limitPrompt')}
+        ></Textarea>
+      </Flex>
 
       <Flex mt={5} alignItems={'center'}>
         <Box w={['60px', '100px', '140px']} flexShrink={0}></Box>

client/src/service/models/model.ts

@@ -43,7 +43,10 @@ const ModelSchema = new Schema({
       default: ''
     },
     systemPrompt: {
-      // 系统提示词
+      type: String,
+      default: ''
+    },
+    limitPrompt: {
       type: String,
       default: ''
     },

client/src/service/utils/auth.ts

@@ -163,7 +163,7 @@ export const authUser = async ({
 
 /* random get openai api key */
 export const getSystemOpenAiKey = () => {
-  return process.env.OPENAIKEY || '';
+  return process.env.ONEAPI_KEY || process.env.OPENAIKEY || '';
 };
 
 /* 获取 api 请求的 key */

client/src/service/utils/chat/index.ts

@@ -6,8 +6,8 @@ import { sseResponse } from '../tools';
 import { OpenAiChatEnum } from '@/constants/model';
 import { chatResponse, openAiStreamResponse } from './openai';
 import type { NextApiResponse } from 'next';
-import { createParser, ParsedEvent, ReconnectInterval } from 'eventsource-parser';
 import { textAdaptGptResponse } from '@/utils/adapt';
+import { parseStreamChunk } from '@/utils/adapt';
 
 export type ChatCompletionType = {
   apiKey: string;
@@ -185,65 +185,63 @@ export const V2_StreamResponse = async ({
   model: ChatModelType;
 }) => {
   let responseContent = '';
+  let error: any = null;
+
+  const clientRes = async (data: string) => {
+    const { content = '' } = (() => {
+      try {
+        const json = JSON.parse(data);
+        const content: string = json?.choices?.[0].delta.content || '';
+        error = json.error;
+        responseContent += content;
+        return { content };
+      } catch (error) {
+        return {};
+      }
+    })();
+
+    if (res.closed || error) return;
+
+    if (data === '[DONE]') {
+      sseResponse({
+        res,
+        event: sseResponseEventEnum.answer,
+        data: textAdaptGptResponse({
+          text: null,
+          finish_reason: 'stop'
+        })
+      });
+      sseResponse({
+        res,
+        event: sseResponseEventEnum.answer,
+        data: '[DONE]'
+      });
+    } else {
+      sseResponse({
+        res,
+        event: sseResponseEventEnum.answer,
+        data: textAdaptGptResponse({
+          text: content
+        })
+      });
+    }
+  };
 
   try {
-    const onParse = async (e: ParsedEvent | ReconnectInterval) => {
-      if (e.type !== 'event') return;
-
-      const data = e.data;
-
-      const { content = '' } = (() => {
-        try {
-          const json = JSON.parse(data);
-          const content: string = json?.choices?.[0].delta.content || '';
-          responseContent += content;
-          return { content };
-        } catch (error) {}
-        return {};
-      })();
-
-      if (res.closed) return;
-
-      if (data === '[DONE]') {
-        sseResponse({
-          res,
-          event: sseResponseEventEnum.answer,
-          data: textAdaptGptResponse({
-            text: null,
-            finish_reason: 'stop'
-          })
-        });
-        sseResponse({
-          res,
-          event: sseResponseEventEnum.answer,
-          data: '[DONE]'
-        });
-      } else {
-        sseResponse({
-          res,
-          event: sseResponseEventEnum.answer,
-          data: textAdaptGptResponse({
-            text: content
-          })
-        });
-      }
-    };
-
-    try {
-      const parser = createParser(onParse);
-      const decoder = new TextDecoder();
-      for await (const chunk of chatResponse.data as any) {
-        if (res.closed) {
-          break;
-        }
-        parser.feed(decoder.decode(chunk, { stream: true }));
-      }
-    } catch (error) {
-      console.log('pipe error', error);
+    for await (const chunk of chatResponse.data as any) {
+      if (res.closed) break;
+      const parse = parseStreamChunk(chunk);
+      parse.forEach((item) => clientRes(item.data));
     }
   } catch (error) {
-    console.log('stream error', error);
+    console.log('pipe error', error);
   }
+
+  if (error) {
+    console.log(error);
+    return Promise.reject(error);
+  }
+
   // count tokens
   const finishMessages = prompts.concat({
     obj: ChatRoleEnum.AI,

client/src/service/utils/chat/openai.ts

@@ -1,18 +1,20 @@
 import { Configuration, OpenAIApi } from 'openai';
-import { createParser, ParsedEvent, ReconnectInterval } from 'eventsource-parser';
 import { axiosConfig } from '../tools';
 import { ChatModelMap, OpenAiChatEnum } from '@/constants/model';
 import { adaptChatItem_openAI } from '@/utils/plugin/openai';
 import { modelToolMap } from '@/utils/plugin';
 import { ChatCompletionType, ChatContextFilter, StreamResponseType } from './index';
 import { ChatRoleEnum } from '@/constants/chat';
+import { parseStreamChunk } from '@/utils/adapt';
 
-export const getOpenAIApi = () =>
-  new OpenAIApi(
+export const getOpenAIApi = (apiKey: string) => {
+  const openaiBaseUrl = process.env.OPENAI_BASE_URL || 'https://api.openai.com/v1';
+  return new OpenAIApi(
     new Configuration({
-      basePath: process.env.OPENAI_BASE_URL || 'https://api.openai.com/v1'
+      basePath: apiKey === process.env.ONEAPI_KEY ? process.env.ONEAPI_URL : openaiBaseUrl
     })
   );
+};
 
 /* 模型对话 */
 export const chatResponse = async ({
@@ -31,7 +33,7 @@ export const chatResponse = async ({
   });
 
   const adaptMessages = adaptChatItem_openAI({ messages: filterMessages, reserveId: false });
-  const chatAPI = getOpenAIApi();
+  const chatAPI = getOpenAIApi(apiKey);
 
   const promptsToken = modelToolMap[model].countTokens({
     messages: filterMessages
@@ -47,8 +49,8 @@ export const chatResponse = async ({
       messages: adaptMessages,
       frequency_penalty: 0.5, // 越大，重复内容越少
       presence_penalty: -0.5, // 越大，越容易出现新内容
-      stream,
-      stop: ['.!?。']
+      stream
+      // stop: ['.!?。']
     },
     {
       timeout: stream ? 60000 : 480000,
@@ -80,29 +82,29 @@ export const openAiStreamResponse = async ({
   try {
     let responseContent = '';
 
-    const onParse = async (event: ParsedEvent | ReconnectInterval) => {
-      if (event.type !== 'event') return;
-      const data = event.data;
-      if (data === '[DONE]') return;
-      try {
-        const json = JSON.parse(data);
-        const content: string = json?.choices?.[0].delta.content || '';
-        responseContent += content;
+    const clientRes = async (data: string) => {
+      const { content = '' } = (() => {
+        try {
+          const json = JSON.parse(data);
+          const content: string = json?.choices?.[0].delta.content || '';
+          responseContent += content;
+          return { content };
+        } catch (error) {
+          return {};
+        }
+      })();
 
-        !res.closed && content && res.write(content);
-      } catch (error) {
-        error;
-      }
+      if (data === '[DONE]') return;
+
+      !res.closed && content && res.write(content);
     };
 
     try {
-      const decoder = new TextDecoder();
-      const parser = createParser(onParse);
       for await (const chunk of chatResponse.data as any) {
-        if (res.closed) {
-          break;
-        }
-        parser.feed(decoder.decode(chunk, { stream: true }));
+        if (res.closed) break;
+
+        const parse = parseStreamChunk(chunk);
+        parse.forEach((item) => clientRes(item.data));
       }
     } catch (error) {
       console.log('pipe error', error);

client/src/service/utils/tools.ts

@@ -34,14 +34,18 @@ export const clearCookie = (res: NextApiResponse) => {
 };
 
 /* openai axios config */
-export const axiosConfig = (apikey: string) => ({
-  baseURL: process.env.OPENAI_BASE_URL || 'https://api.openai.com/v1',
-  httpsAgent: global.httpsAgent,
-  headers: {
-    Authorization: `Bearer ${apikey}`,
-    auth: process.env.OPENAI_BASE_URL_AUTH || ''
-  }
-});
+export const axiosConfig = (apikey: string) => {
+  const openaiBaseUrl = process.env.OPENAI_BASE_URL || 'https://api.openai.com/v1';
+
+  return {
+    baseURL: apikey === process.env.ONEAPI_KEY ? process.env.ONEAPI_URL : openaiBaseUrl, // 此处仅对非 npm 模块有效
+    httpsAgent: global.httpsAgent,
+    headers: {
+      Authorization: `Bearer ${apikey}`,
+      auth: process.env.OPENAI_BASE_URL_AUTH || ''
+    }
+  };
+};
 
 export function withNextCors(handler: NextApiHandler): NextApiHandler {
   return async function nextApiHandlerWrappedWithNextCors(

client/src/types/mongoSchema.d.ts

@@ -43,6 +43,7 @@ export interface ModelSchema {
     searchLimit: number;
     searchEmptyText: string;
     systemPrompt: string;
+    limitPrompt: string;
     temperature: number;
     maxToken: number;
     chatModel: ChatModelType; // 聊天时用的模型，训练后就是训练的模型

client/src/utils/adapt.ts

@@ -54,3 +54,24 @@ export const textAdaptGptResponse = ({
     choices: [{ delta: text === null ? {} : { content: text }, index: 0, finish_reason }]
   });
 };
+
+const decoder = new TextDecoder();
+export const parseStreamChunk = (value: BufferSource) => {
+  const chunk = decoder.decode(value);
+  const chunkLines = chunk.split('\n\n').filter((item) => item);
+  const chunkResponse = chunkLines.map((item) => {
+    const splitEvent = item.split('\n');
+    if (splitEvent.length === 2) {
+      return {
+        event: splitEvent[0].replace('event: ', ''),
+        data: splitEvent[1].replace('data: ', '')
+      };
+    }
+    return {
+      event: '',
+      data: splitEvent[0].replace('data: ', '')
+    };
+  });
+
+  return chunkResponse;
+};

client/src/utils/tools.ts

@@ -116,7 +116,7 @@ export const voiceBroadcast = ({ text }: { text: string }) => {
 };
 
 export const getErrText = (err: any, def = '') => {
-  const msg = typeof err === 'string' ? err : err?.message || def || '';
+  const msg: string = typeof err === 'string' ? err : err?.message || def || '';
   msg && console.log('error =>', msg);
   return msg;
 };

docs/deploy/fastgpt/docker-compose.yml

@@ -67,10 +67,13 @@ services:
       - PG_USER=fastgpt
       - PG_PASSWORD=1234
       - PG_DB_NAME=fastgpt
-      # openai, 推荐使用 one-api 管理key
+      # oneapi 配置 推荐使用 one-api 管理key
+      - ONEAPI_URL=https://kfcwurtbijvh.cloud.sealos.io/v1
+      - ONEAPI_KEY=sk-itJ9v8qthRiFDzfs62Ea21Aa9b004c8791937dCf4cC568Ff
+      # openai 相关配置：使用了 oneapi 后，下面只需要填下 OPENAI_BASE_URL （国外可全忽略）
       - OPENAIKEY=sk-xxxxx
       - OPENAI_BASE_URL=https://api.openai.com/v1
-      - OPENAI_BASE_URL_AUTH=可选的安全凭证
+      - OPENAI_BASE_URL_AUTH=可选的安全凭证，会放到 header.auth 里
   fastgpt-admin:
     image: registry.cn-hangzhou.aliyuncs.com/fastgpt/fastgpt-admin:latest
     container_name: fastgpt-admin

docs/deploy/one-api/sealos.md

@@ -12,6 +12,8 @@ https://cloud.sealos.io/
 
 ## 4. 填写对应参数
 
+镜像：ghcr.io/songquanpeng/one-api:latest
+
 ![step2](./sealosImg/step2.png)
 打开外网访问开关后，Sealos 会自动分配一个可访问的地址，不需要自己配置。