feat: lafClaude

2025-07-23 21:13:50 +00:00 · 2023-05-04 10:53:55 +08:00
parent 3c8f38799c
commit 0d6897e180
22 changed files with 327 additions and 231 deletions
--- a/package.json
+++ b/package.json
@@ -26,7 +26,6 @@
    "axios": "^1.3.3",
    "crypto": "^1.0.1",
    "dayjs": "^1.11.7",
-    "delay": "^5.0.0",
    "eventsource-parser": "^0.1.0",
    "formidable": "^2.1.1",
    "framer-motion": "^9.0.6",
--- a/pnpm-lock.yaml
+++ b/pnpm-lock.yaml
@@ -28,7 +28,6 @@ specifiers:
  axios: ^1.3.3
  crypto: ^1.0.1
  dayjs: ^1.11.7
-  delay: ^5.0.0
  eslint: 8.34.0
  eslint-config-next: 13.1.6
  eventsource-parser: ^0.1.0
@@ -83,7 +82,6 @@ dependencies:
  axios: registry.npmmirror.com/axios/1.3.3
  crypto: registry.npmmirror.com/crypto/1.0.1
  dayjs: registry.npmmirror.com/dayjs/1.11.7
-  delay: 5.0.0
  eventsource-parser: registry.npmmirror.com/eventsource-parser/0.1.0
  formidable: registry.npmmirror.com/formidable/2.1.1
  framer-motion: registry.npmmirror.com/framer-motion/9.0.6_biqbaboplfbrettd7655fr4n2y
@@ -288,11 +286,6 @@ packages:
    dev: false
    optional: true

-  /delay/5.0.0:
-    resolution: {integrity: sha512-ReEBKkIfe4ya47wlPYf/gu5ib6yUG0/Aez0JQZQz94kiWtRQvZIQbTiehsnwHvLSWJnQdhVeqYue7Id1dKr0qw==}
-    engines: {node: '>=10'}
-    dev: false
-
  /fsevents/2.3.2:
    resolution: {integrity: sha512-xiqMQR4xAeHTuB9uWm+fFRcIOgKBMiOBP+eXiyT7jsgVCq1bkVygt00oASowB7EdtpOHaaPgKt812P9ab+DDKA==}
    engines: {node: ^8.16.0 || ^10.6.0 || >=11.0.0}
--- a/public/docs/intro.md
+++ b/public/docs/intro.md
@@ -3,12 +3,13 @@
 [Git 仓库](https://github.com/c121914yu/FastGPT)

 ### 交流群/问题反馈
+
 扫码满了，加个小号，定时拉
-wx号: fastgpt123
+wx 号: fastgpt123
 ![](/imgs/wx300.jpg)

-
 ### 快速开始
+
 1. 使用手机号注册账号。
 2. 进入账号页面，添加关联账号，目前只有 openai 的账号可以添加，直接去 openai 官网，把 API Key 粘贴过来。
 3. 如果填写了自己的 openai 账号，使用时会直接用你的账号。如果没有填写，需要付费使用平台的账号。
@@ -16,15 +17,16 @@ wx号: fastgpt123
 5. 在模型列表点击【对话】，即可使用 API 进行聊天。

 ### 价格表
+
 如果使用了自己的 Api Key，不会计费。可以在账号页，看到详细账单。单纯使用 chatGPT 模型进行对话，只有一个计费项目。使用知识库时，包含**对话**和**索引**生成两个计费项。
 | 计费项 | 价格: 元/ 1K tokens（包含上下文）|
 | --- | --- |
+| claude - 对话 | 免费 |
 | chatgpt - 对话 | 0.03 |
 | 知识库 - 对话 | 0.03 |
 | 知识库 - 索引 | 0.004 |
 | 文件拆分 | 0.03 |

-
 ### 定制 prompt

 1. 进入模型编辑页
--- a/src/api/chat.ts
+++ b/src/api/chat.ts
@@ -24,6 +24,7 @@ export const delChatHistoryById = (id: string) => GET(`/chat/removeHistory?id=${
 */
 export const postSaveChat = (data: {
  modelId: string;
+  newChatId: '' | string;
  chatId: '' | string;
  prompts: ChatItemType[];
 }) => POST<string>('/chat/saveChat', data);
--- a/src/api/fetch.ts
+++ b/src/api/fetch.ts
@@ -1,5 +1,5 @@
 import { getToken } from '../utils/user';
-import { SYSTEM_PROMPT_PREFIX } from '@/constants/chat';
+import { SYSTEM_PROMPT_HEADER, NEW_CHATID_HEADER } from '@/constants/chat';

 interface StreamFetchProps {
  url: string;
@@ -8,7 +8,8 @@ interface StreamFetchProps {
  abortSignal: AbortController;
 }
 export const streamFetch = ({ url, data, onMessage, abortSignal }: StreamFetchProps) =>
-  new Promise<{ responseText: string; systemPrompt: string }>(async (resolve, reject) => {
+  new Promise<{ responseText: string; systemPrompt: string; newChatId: string }>(
+    async (resolve, reject) => {
      try {
        const res = await fetch(url, {
          method: 'POST',
@@ -23,15 +24,18 @@ export const streamFetch = ({ url, data, onMessage, abortSignal }: StreamFetchPr
        if (!reader) return;

        const decoder = new TextDecoder();
+
+        const systemPrompt = decodeURIComponent(res.headers.get(SYSTEM_PROMPT_HEADER) || '');
+        const newChatId = decodeURIComponent(res.headers.get(NEW_CHATID_HEADER) || '');
+
        let responseText = '';
-      let systemPrompt = '';

        const read = async () => {
          try {
            const { done, value } = await reader?.read();
            if (done) {
              if (res.status === 200) {
-              resolve({ responseText, systemPrompt });
+                resolve({ responseText, systemPrompt, newChatId });
              } else {
                const parseError = JSON.parse(responseText);
                reject(parseError?.message || '请求异常');
@@ -39,29 +43,21 @@ export const streamFetch = ({ url, data, onMessage, abortSignal }: StreamFetchPr

              return;
            }
-          let text = decoder.decode(value).replace(/<br\/>/g, '\n');
-          // check system prompt
-          if (text.includes(SYSTEM_PROMPT_PREFIX)) {
-            const arr = text.split(SYSTEM_PROMPT_PREFIX);
-            systemPrompt = arr.pop() || '';
-
-            text = arr.join('');
-          }
+            const text = decoder.decode(value).replace(/<br\/>/g, '\n');
            responseText += text;
            onMessage(text);
-
            read();
          } catch (err: any) {
            if (err?.message === 'The user aborted a request.') {
-            return resolve({ responseText, systemPrompt });
+              return resolve({ responseText, systemPrompt, 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');
      }
-  });
+    }
+  );
--- a/src/constants/chat.ts
+++ b/src/constants/chat.ts
@@ -1,4 +1,5 @@
-export const SYSTEM_PROMPT_PREFIX = 'SYSTEM_PROMPT:';
+export const SYSTEM_PROMPT_HEADER = 'System-Prompt-Header';
+export const NEW_CHATID_HEADER = 'Chat-Id-Header';

 export enum ChatRoleEnum {
  System = 'System',
--- a/src/constants/model.ts
+++ b/src/constants/model.ts
@@ -8,13 +8,17 @@ export enum OpenAiChatEnum {
  'GPT4' = 'gpt-4',
  'GPT432k' = 'gpt-4-32k'
 }
+export enum ClaudeEnum {
+  'Claude' = 'Claude'
+}

-export type ChatModelType = `${OpenAiChatEnum}`;
+export type ChatModelType = `${OpenAiChatEnum}` | `${ClaudeEnum}`;

 export type ChatModelItemType = {
  chatModel: ChatModelType;
  name: string;
  contextMaxToken: number;
+  systemMaxToken: number;
  maxTemperature: number;
  price: number;
 };
@@ -24,6 +28,7 @@ export const ChatModelMap = {
    chatModel: OpenAiChatEnum.GPT35,
    name: 'ChatGpt',
    contextMaxToken: 4096,
+    systemMaxToken: 3000,
    maxTemperature: 1.5,
    price: 3
  },
@@ -31,6 +36,7 @@ export const ChatModelMap = {
    chatModel: OpenAiChatEnum.GPT4,
    name: 'Gpt4',
    contextMaxToken: 8000,
+    systemMaxToken: 4000,
    maxTemperature: 1.5,
    price: 30
  },
@@ -38,12 +44,24 @@ export const ChatModelMap = {
    chatModel: OpenAiChatEnum.GPT432k,
    name: 'Gpt4-32k',
    contextMaxToken: 32000,
+    systemMaxToken: 4000,
    maxTemperature: 1.5,
    price: 30
+  },
+  [ClaudeEnum.Claude]: {
+    chatModel: ClaudeEnum.Claude,
+    name: 'Claude(免费体验)',
+    contextMaxToken: 9000,
+    systemMaxToken: 2500,
+    maxTemperature: 1,
+    price: 0
  }
 };

-export const chatModelList: ChatModelItemType[] = [ChatModelMap[OpenAiChatEnum.GPT35]];
+export const chatModelList: ChatModelItemType[] = [
+  ChatModelMap[OpenAiChatEnum.GPT35],
+  ChatModelMap[ClaudeEnum.Claude]
+];

 export enum ModelStatusEnum {
  running = 'running',
--- a/src/pages/api/chat/chat.ts
+++ b/src/pages/api/chat/chat.ts
@@ -42,7 +42,8 @@ export default async function handler(req: NextApiRequest, res: NextApiResponse)
    await connectToDatabase();
    let startTime = Date.now();

-    const { model, showModelDetail, content, userApiKey, systemApiKey, userId } = await authChat({
+    const { model, showModelDetail, content, userOpenAiKey, systemAuthKey, userId } =
+      await authChat({
        modelId,
        chatId,
        authorization
@@ -56,8 +57,7 @@ export default async function handler(req: NextApiRequest, res: NextApiResponse)
    // 使用了知识库搜索
    if (model.chat.useKb) {
      const { code, searchPrompt } = await searchKb({
-        userApiKey,
-        systemApiKey,
+        userOpenAiKey,
        prompts,
        similarity: ModelVectorSearchModeMap[model.chat.searchMode]?.similarity,
        model,
@@ -86,10 +86,12 @@ export default async function handler(req: NextApiRequest, res: NextApiResponse)

    // 发出请求
    const { streamResponse } = await modelServiceToolMap[model.chat.chatModel].chatCompletion({
-      apiKey: userApiKey || systemApiKey,
+      apiKey: userOpenAiKey || systemAuthKey,
      temperature: +temperature,
      messages: prompts,
-      stream: true
+      stream: true,
+      res,
+      chatId
    });

    console.log('api response time:', `${(Date.now() - startTime) / 1000}s`);
@@ -108,7 +110,7 @@ export default async function handler(req: NextApiRequest, res: NextApiResponse)

    // 只有使用平台的 key 才计费
    pushChatBill({
-      isPay: !userApiKey,
+      isPay: !userOpenAiKey,
      chatModel: model.chat.chatModel,
      userId,
      chatId,
--- a/src/pages/api/chat/saveChat.ts
+++ b/src/pages/api/chat/saveChat.ts
@@ -9,7 +9,8 @@ import mongoose from 'mongoose';
 /* 聊天内容存存储 */
 export default async function handler(req: NextApiRequest, res: NextApiResponse) {
  try {
-    const { chatId, modelId, prompts } = req.body as {
+    const { chatId, modelId, prompts, newChatId } = req.body as {
+      newChatId: '' | string;
      chatId: '' | string;
      modelId: string;
      prompts: ChatItemType[];
@@ -35,6 +36,7 @@ export default async function handler(req: NextApiRequest, res: NextApiResponse)
    // 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,
--- a/src/pages/api/openapi/chat/chat.ts
+++ b/src/pages/api/openapi/chat/chat.ts
@@ -65,7 +65,6 @@ export default async function handler(req: NextApiRequest, res: NextApiResponse)
      const similarity = ModelVectorSearchModeMap[model.chat.searchMode]?.similarity || 0.22;

      const { code, searchPrompt } = await searchKb({
-        systemApiKey: apiKey,
        prompts,
        similarity,
        model,
--- a/src/pages/api/openapi/chat/lafGpt.ts
+++ b/src/pages/api/openapi/chat/lafGpt.ts
@@ -116,7 +116,6 @@ export default async function handler(req: NextApiRequest, res: NextApiResponse)

    // 获取向量匹配到的提示词
    const { searchPrompt } = await searchKb({
-      systemApiKey: apiKey,
      similarity: ModelVectorSearchModeMap[model.chat.searchMode]?.similarity,
      prompts,
      model,
--- a/src/pages/chat/index.tsx
+++ b/src/pages/chat/index.tsx
@@ -206,7 +206,7 @@ const Chat = ({ modelId, chatId }: { modelId: string; chatId: string }) => {
      };

      // 流请求，获取数据
-      const { responseText, systemPrompt } = await streamFetch({
+      let { responseText, systemPrompt, newChatId } = await streamFetch({
        url: '/api/chat/chat',
        data: {
          prompt,
@@ -234,10 +234,10 @@ const Chat = ({ modelId, chatId }: { modelId: string; chatId: string }) => {
        return;
      }

-      let newChatId = '';
      // save chat record
      try {
        newChatId = await postSaveChat({
+          newChatId, // 如果有newChatId，会自动以这个Id创建对话框
          modelId,
          chatId,
          prompts: [
--- a/src/service/events/generateQA.ts
+++ b/src/service/events/generateQA.ts
@@ -45,12 +45,12 @@ export async function generateQA(next = false): Promise<any> {
    const textList: string[] = dataItem.textList.slice(-5);

    // 获取 openapi Key
-    let userApiKey = '',
-      systemApiKey = '';
+    let userOpenAiKey = '',
+      systemAuthKey = '';
    try {
      const key = await getApiKey({ model: OpenAiChatEnum.GPT35, userId: dataItem.userId });
-      userApiKey = key.userApiKey;
-      systemApiKey = key.systemApiKey;
+      userOpenAiKey = key.userOpenAiKey;
+      systemAuthKey = key.systemAuthKey;
    } catch (error: any) {
      if (error?.code === 501) {
        // 余额不够了, 清空该记录
@@ -73,18 +73,18 @@ export async function generateQA(next = false): Promise<any> {
      textList.map((text) =>
        modelServiceToolMap[OpenAiChatEnum.GPT35]
          .chatCompletion({
-            apiKey: userApiKey || systemApiKey,
+            apiKey: userOpenAiKey || systemAuthKey,
            temperature: 0.8,
            messages: [
              {
                obj: ChatRoleEnum.System,
                value: `你是出题人
-          ${dataItem.prompt || '下面是"一段长文本"'}
-          从中选出5至20个题目和答案.答案详细.按格式返回: Q1:
-          A1:
-          Q2:
-          A2:
-          ...`
+${dataItem.prompt || '下面是"一段长文本"'}
+从中选出5至20个题目和答案.答案详细.按格式返回: Q1:
+A1:
+Q2:
+A2:
+...`
              },
              {
                obj: 'Human',
@@ -98,7 +98,7 @@ export async function generateQA(next = false): Promise<any> {
            console.log(`split result length: `, result.length);
            // 计费
            pushSplitDataBill({
-              isPay: !userApiKey && result.length > 0,
+              isPay: !userOpenAiKey && result.length > 0,
              userId: dataItem.userId,
              type: 'QA',
              textLen: responseMessages.map((item) => item.value).join('').length,
--- a/src/service/events/generateVector.ts
+++ b/src/service/events/generateVector.ts
@@ -2,7 +2,6 @@ import { openaiCreateEmbedding } from '../utils/chat/openai';
 import { getApiKey } from '../utils/auth';
 import { openaiError2 } from '../errorCode';
 import { PgClient } from '@/service/pg';
-import { embeddingModel } from '@/constants/model';

 export async function generateVector(next = false): Promise<any> {
  if (process.env.queueTask !== '1') {
@@ -42,11 +41,10 @@ export async function generateVector(next = false): Promise<any> {
    dataId = dataItem.id;

    // 获取 openapi Key
-    let userApiKey, systemApiKey;
+    let userOpenAiKey;
    try {
-      const res = await getApiKey({ model: embeddingModel, userId: dataItem.userId });
-      userApiKey = res.userApiKey;
-      systemApiKey = res.systemApiKey;
+      const res = await getApiKey({ model: 'gpt-3.5-turbo', userId: dataItem.userId });
+      userOpenAiKey = res.userOpenAiKey;
    } catch (error: any) {
      if (error?.code === 501) {
        await PgClient.delete('modelData', {
@@ -63,8 +61,7 @@ export async function generateVector(next = false): Promise<any> {
    const { vectors } = await openaiCreateEmbedding({
      textArr: [dataItem.q],
      userId: dataItem.userId,
-      userApiKey,
-      systemApiKey
+      userOpenAiKey
    });

    // 更新 pg 向量和状态数据
--- a/src/service/plugins/searchKb.ts
+++ b/src/service/plugins/searchKb.ts
@@ -3,22 +3,20 @@ import { ModelDataStatusEnum, ModelVectorSearchModeEnum, ChatModelMap } from '@/
 import { ModelSchema } from '@/types/mongoSchema';
 import { openaiCreateEmbedding } from '../utils/chat/openai';
 import { ChatRoleEnum } from '@/constants/chat';
-import { sliceTextByToken } from '@/utils/chat';
+import { modelToolMap } from '@/utils/chat';
 import { ChatItemSimpleType } from '@/types/chat';

 /**
 *  use openai embedding search kb
 */
 export const searchKb = async ({
-  userApiKey,
-  systemApiKey,
+  userOpenAiKey,
  prompts,
  similarity = 0.2,
  model,
  userId
 }: {
-  userApiKey?: string;
-  systemApiKey: string;
+  userOpenAiKey?: string;
  prompts: ChatItemSimpleType[];
  model: ModelSchema;
  userId: string;
@@ -33,8 +31,7 @@ export const searchKb = async ({
  async function search(textArr: string[] = []) {
    // 获取提示词的向量
    const { vectors: promptVectors } = await openaiCreateEmbedding({
-      userApiKey,
-      systemApiKey,
+      userOpenAiKey,
      userId,
      textArr
    });
@@ -81,11 +78,24 @@ export const searchKb = async ({
  ].filter((item) => item);
  const systemPrompts = await search(searchArr);

-  // filter system prompt
-  if (
-    systemPrompts.length === 0 &&
-    model.chat.searchMode === ModelVectorSearchModeEnum.hightSimilarity
-  ) {
+  // filter system prompts.
+  const filterRateMap: Record<number, number[]> = {
+    1: [1],
+    2: [0.7, 0.3]
+  };
+  const filterRate = filterRateMap[systemPrompts.length] || filterRateMap[0];
+
+  const filterSystemPrompt = filterRate
+    .map((rate, i) =>
+      modelToolMap[model.chat.chatModel].sliceText({
+        text: systemPrompts[i],
+        length: Math.floor(modelConstantsData.systemMaxToken * rate)
+      })
+    )
+    .join('\n');
+
+  /* 高相似度+不回复 */
+  if (!filterSystemPrompt && model.chat.searchMode === ModelVectorSearchModeEnum.hightSimilarity) {
    return {
      code: 201,
      searchPrompt: {
@@ -95,7 +105,7 @@ export const searchKb = async ({
    };
  }
  /* 高相似度+无上下文，不添加额外知识,仅用系统提示词 */
-  if (systemPrompts.length === 0 && model.chat.searchMode === ModelVectorSearchModeEnum.noContext) {
+  if (!filterSystemPrompt && model.chat.searchMode === ModelVectorSearchModeEnum.noContext) {
    return {
      code: 200,
      searchPrompt: model.chat.systemPrompt
@@ -107,25 +117,7 @@ export const searchKb = async ({
    };
  }

-  /* 有匹配情况下，system 添加知识库内容。 */
-
-  // filter system prompts. max 70% tokens
-  const filterRateMap: Record<number, number[]> = {
-    1: [0.7],
-    2: [0.5, 0.2]
-  };
-  const filterRate = filterRateMap[systemPrompts.length] || filterRateMap[0];
-
-  const filterSystemPrompt = filterRate
-    .map((rate, i) =>
-      sliceTextByToken({
-        model: model.chat.chatModel,
-        text: systemPrompts[i],
-        length: Math.floor(modelConstantsData.contextMaxToken * rate)
-      })
-    )
-    .join('\n');
-
+  /* 有匹配 */
  return {
    code: 200,
    searchPrompt: {
@@ -133,9 +125,9 @@ export const searchKb = async ({
      value: `
 ${model.chat.systemPrompt}
 ${
-  model.chat.searchMode === ModelVectorSearchModeEnum.hightSimilarity ? `不回答知识库外的内容.` : ''
+  model.chat.searchMode === ModelVectorSearchModeEnum.hightSimilarity ? '不回答知识库外的内容.' : ''
 }
-知识库内容为: ${filterSystemPrompt}'
+知识库内容为: '${filterSystemPrompt}'
 `
    }
  };
--- a/src/service/utils/auth.ts
+++ b/src/service/utils/auth.ts
@@ -4,15 +4,10 @@ import { Chat, Model, OpenApi, User } from '../mongo';
 import type { ModelSchema } from '@/types/mongoSchema';
 import type { ChatItemSimpleType } from '@/types/chat';
 import mongoose from 'mongoose';
-import { defaultModel } from '@/constants/model';
+import { ClaudeEnum, defaultModel } from '@/constants/model';
 import { formatPrice } from '@/utils/user';
 import { ERROR_ENUM } from '../errorCode';
-import {
-  ChatModelType,
-  OpenAiChatEnum,
-  embeddingModel,
-  EmbeddingModelType
-} from '@/constants/model';
+import { ChatModelType, OpenAiChatEnum } from '@/constants/model';

 /* 校验 token */
 export const authToken = (token?: string): Promise<string> => {
@@ -34,13 +29,7 @@ export const authToken = (token?: string): Promise<string> => {
 };

 /* 获取 api 请求的 key */
-export const getApiKey = async ({
-  model,
-  userId
-}: {
-  model: ChatModelType | EmbeddingModelType;
-  userId: string;
-}) => {
+export const getApiKey = async ({ model, userId }: { model: ChatModelType; userId: string }) => {
  const user = await User.findById(userId);
  if (!user) {
    return Promise.reject({
@@ -51,29 +40,29 @@ export const getApiKey = async ({

  const keyMap = {
    [OpenAiChatEnum.GPT35]: {
-      userApiKey: user.openaiKey || '',
-      systemApiKey: process.env.OPENAIKEY as string
+      userOpenAiKey: user.openaiKey || '',
+      systemAuthKey: process.env.OPENAIKEY as string
    },
    [OpenAiChatEnum.GPT4]: {
-      userApiKey: user.openaiKey || '',
-      systemApiKey: process.env.OPENAIKEY as string
+      userOpenAiKey: user.openaiKey || '',
+      systemAuthKey: process.env.OPENAIKEY as string
    },
    [OpenAiChatEnum.GPT432k]: {
-      userApiKey: user.openaiKey || '',
-      systemApiKey: process.env.OPENAIKEY as string
+      userOpenAiKey: user.openaiKey || '',
+      systemAuthKey: process.env.OPENAIKEY as string
    },
-    [embeddingModel]: {
-      userApiKey: user.openaiKey || '',
-      systemApiKey: process.env.OPENAIKEY as string
+    [ClaudeEnum.Claude]: {
+      userOpenAiKey: '',
+      systemAuthKey: process.env.LAFKEY as string
    }
  };

  // 有自己的key
-  if (keyMap[model].userApiKey) {
+  if (keyMap[model].userOpenAiKey) {
    return {
      user,
-      userApiKey: keyMap[model].userApiKey,
-      systemApiKey: ''
+      userOpenAiKey: keyMap[model].userOpenAiKey,
+      systemAuthKey: ''
    };
  }

@@ -87,8 +76,8 @@ export const getApiKey = async ({

  return {
    user,
-    userApiKey: '',
-    systemApiKey: keyMap[model].systemApiKey
+    userOpenAiKey: '',
+    systemAuthKey: keyMap[model].systemAuthKey
  };
 };

@@ -176,11 +165,11 @@ export const authChat = async ({
    ]);
  }
  // 获取 user 的 apiKey
-  const { userApiKey, systemApiKey } = await getApiKey({ model: model.chat.chatModel, userId });
+  const { userOpenAiKey, systemAuthKey } = await getApiKey({ model: model.chat.chatModel, userId });

  return {
-    userApiKey,
-    systemApiKey,
+    userOpenAiKey,
+    systemAuthKey,
    content,
    userId,
    model,
--- a/src/service/utils/chat/claude.ts
+++ b/src/service/utils/chat/claude.ts
@@ -0,0 +1,103 @@
+import { modelToolMap } from '@/utils/chat';
+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';
+import { ClaudeEnum } from '@/constants/model';
+
+/* 模型对话 */
+export const lafClaudChat = async ({
+  apiKey,
+  messages,
+  stream,
+  chatId,
+  res
+}: 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')
+    .map((item) => item.value)
+    .join('\n');
+  const systemPromptText = systemPrompt ? `这是我的知识:'${systemPrompt}'\n` : '';
+
+  const prompt = systemPromptText + messages[messages.length - 1].value;
+
+  const lafResponse = await axios.post(
+    'https://hnvacz.laf.run/claude-gpt',
+    {
+      prompt,
+      stream,
+      conversationId
+    },
+    {
+      headers: {
+        Authorization: apiKey
+      },
+      timeout: stream ? 40000 : 240000,
+      responseType: stream ? 'stream' : 'json'
+    }
+  );
+
+  let responseText = '';
+  let totalTokens = 0;
+
+  if (!stream) {
+    responseText = lafResponse.data?.text || '';
+  }
+
+  return {
+    streamResponse: lafResponse,
+    responseMessages: messages.concat({ obj: ChatRoleEnum.AI, value: responseText }),
+    responseText,
+    totalTokens
+  };
+};
+
+/* openai stream response */
+export const lafClaudStreamResponse = async ({
+  stream,
+  chatResponse,
+  prompts
+}: StreamResponseType) => {
+  try {
+    let responseContent = '';
+
+    try {
+      const decoder = new TextDecoder();
+      for await (const chunk of chatResponse.data as any) {
+        if (stream.destroyed) {
+          // 流被中断了，直接忽略后面的内容
+          break;
+        }
+        const content = decoder.decode(chunk);
+        responseContent += content;
+        content && stream.push(content.replace(/\n/g, '<br/>'));
+      }
+    } catch (error) {
+      console.log('pipe error', error);
+    }
+    // count tokens
+    const finishMessages = prompts.concat({
+      obj: ChatRoleEnum.AI,
+      value: responseContent
+    });
+    const totalTokens = modelToolMap[ClaudeEnum.Claude].countTokens({
+      messages: finishMessages
+    });
+
+    return {
+      responseContent,
+      totalTokens,
+      finishMessages
+    };
+  } catch (error) {
+    return Promise.reject(error);
+  }
+};
--- a/src/service/utils/chat/index.ts
+++ b/src/service/utils/chat/index.ts
@@ -1,19 +1,19 @@
 import { ChatItemSimpleType } from '@/types/chat';
 import { modelToolMap } from '@/utils/chat';
 import type { ChatModelType } from '@/constants/model';
-import { ChatRoleEnum, SYSTEM_PROMPT_PREFIX } from '@/constants/chat';
-import { OpenAiChatEnum } from '@/constants/model';
+import { ChatRoleEnum, SYSTEM_PROMPT_HEADER } from '@/constants/chat';
+import { OpenAiChatEnum, ClaudeEnum } from '@/constants/model';
 import { chatResponse, openAiStreamResponse } from './openai';
+import { lafClaudChat, lafClaudStreamResponse } from './claude';
 import type { NextApiResponse } from 'next';
 import type { PassThrough } from 'stream';
-import delay from 'delay';

 export type ChatCompletionType = {
  apiKey: string;
  temperature: number;
  messages: ChatItemSimpleType[];
  stream: boolean;
-  params?: any;
+  [key: string]: any;
 };
 export type ChatCompletionResponseType = {
  streamResponse: any;
@@ -25,6 +25,9 @@ export type StreamResponseType = {
  stream: PassThrough;
  chatResponse: any;
  prompts: ChatItemSimpleType[];
+  res: NextApiResponse;
+  systemPrompt?: string;
+  [key: string]: any;
 };
 export type StreamResponseReturnType = {
  responseContent: string;
@@ -65,6 +68,10 @@ export const modelServiceToolMap: Record<
        model: OpenAiChatEnum.GPT432k,
        ...data
      })
+  },
+  [ClaudeEnum.Claude]: {
+    chatCompletion: lafClaudChat,
+    streamResponse: lafClaudStreamResponse
  }
 };

@@ -143,14 +150,13 @@ export const resStreamResponse = async ({
  prompts
 }: StreamResponseType & {
  model: ChatModelType;
-  res: NextApiResponse;
-  systemPrompt?: string;
 }) => {
  // 创建响应流
  res.setHeader('Content-Type', 'text/event-stream;charset-utf-8');
  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));
  stream.pipe(res);

  const { responseContent, totalTokens, finishMessages } = await modelServiceToolMap[
@@ -158,16 +164,11 @@ export const resStreamResponse = async ({
  ].streamResponse({
    chatResponse,
    stream,
-    prompts
+    prompts,
+    res,
+    systemPrompt
  });

-  await delay(100);
-
-  // push system prompt
-  !stream.destroyed &&
-    systemPrompt &&
-    stream.push(`${SYSTEM_PROMPT_PREFIX}${systemPrompt.replace(/\n/g, '<br/>')}`);
-
  // close stream
  !stream.destroyed && stream.push(null);
  stream.destroy();
--- a/src/service/utils/chat/openai.ts
+++ b/src/service/utils/chat/openai.ts
@@ -19,18 +19,18 @@ export const getOpenAIApi = (apiKey: string) => {

 /* 获取向量 */
 export const openaiCreateEmbedding = async ({
-  userApiKey,
-  systemApiKey,
+  userOpenAiKey,
  userId,
  textArr
 }: {
-  userApiKey?: string;
-  systemApiKey: string;
+  userOpenAiKey?: string;
  userId: string;
  textArr: string[];
 }) => {
+  const systemAuthKey = process.env.OPENAIKEY as string;
+
  // 获取 chatAPI
-  const chatAPI = getOpenAIApi(userApiKey || systemApiKey);
+  const chatAPI = getOpenAIApi(userOpenAiKey || systemAuthKey);

  // 把输入的内容转成向量
  const res = await chatAPI
@@ -50,7 +50,7 @@ export const openaiCreateEmbedding = async ({
    }));

  pushGenerateVectorBill({
-    isPay: !userApiKey,
+    isPay: !userOpenAiKey,
    userId,
    text: textArr.join(''),
    tokenLen: res.tokenLen
--- a/src/utils/chat/claude.ts
+++ b/src/utils/chat/claude.ts
@@ -0,0 +1,3 @@
+export const ClaudeSliceTextByToken = ({ text, length }: { text: string; length: number }) => {
+  return text.slice(0, length);
+};
--- a/src/utils/chat/index.ts
+++ b/src/utils/chat/index.ts
@@ -1,46 +1,30 @@
-import { OpenAiChatEnum } from '@/constants/model';
+import { ClaudeEnum, OpenAiChatEnum } from '@/constants/model';
 import type { ChatModelType } from '@/constants/model';
 import type { ChatItemSimpleType } from '@/types/chat';
-import { countOpenAIToken, getOpenAiEncMap, adaptChatItem_openAI } from './openai';
-import { ChatCompletionRequestMessage } from 'openai';
-
-export type CountTokenType = { messages: ChatItemSimpleType[] };
+import { countOpenAIToken, openAiSliceTextByToken } from './openai';
+import { ClaudeSliceTextByToken } from './claude';

 export const modelToolMap: Record<
  ChatModelType,
  {
-    countTokens: (data: CountTokenType) => number;
-    adaptChatMessages: (data: CountTokenType) => ChatCompletionRequestMessage[];
+    countTokens: (data: { messages: ChatItemSimpleType[] }) => number;
+    sliceText: (data: { text: string; length: number }) => string;
  }
 > = {
  [OpenAiChatEnum.GPT35]: {
-    countTokens: ({ messages }: CountTokenType) =>
-      countOpenAIToken({ model: OpenAiChatEnum.GPT35, messages }),
-    adaptChatMessages: adaptChatItem_openAI
+    countTokens: ({ messages }) => countOpenAIToken({ model: OpenAiChatEnum.GPT35, messages }),
+    sliceText: (data) => openAiSliceTextByToken({ model: OpenAiChatEnum.GPT35, ...data })
  },
  [OpenAiChatEnum.GPT4]: {
-    countTokens: ({ messages }: CountTokenType) =>
-      countOpenAIToken({ model: OpenAiChatEnum.GPT4, messages }),
-    adaptChatMessages: adaptChatItem_openAI
+    countTokens: ({ messages }) => countOpenAIToken({ model: OpenAiChatEnum.GPT4, messages }),
+    sliceText: (data) => openAiSliceTextByToken({ model: OpenAiChatEnum.GPT35, ...data })
  },
  [OpenAiChatEnum.GPT432k]: {
-    countTokens: ({ messages }: CountTokenType) =>
-      countOpenAIToken({ model: OpenAiChatEnum.GPT432k, messages }),
-    adaptChatMessages: adaptChatItem_openAI
+    countTokens: ({ messages }) => countOpenAIToken({ model: OpenAiChatEnum.GPT432k, messages }),
+    sliceText: (data) => openAiSliceTextByToken({ model: OpenAiChatEnum.GPT35, ...data })
+  },
+  [ClaudeEnum.Claude]: {
+    countTokens: () => 0,
+    sliceText: ClaudeSliceTextByToken
  }
 };
-
-export const sliceTextByToken = ({
-  model = 'gpt-3.5-turbo',
-  text,
-  length
-}: {
-  model: ChatModelType;
-  text: string;
-  length: number;
-}) => {
-  const enc = getOpenAiEncMap()[model];
-  const encodeText = enc.encode(text);
-  const decoder = new TextDecoder();
-  return decoder.decode(enc.decode(encodeText.slice(0, length)));
-};
--- a/src/utils/chat/openai.ts
+++ b/src/utils/chat/openai.ts
@@ -2,7 +2,7 @@ import { encoding_for_model, type Tiktoken } from '@dqbd/tiktoken';
 import type { ChatItemSimpleType } from '@/types/chat';
 import { ChatRoleEnum } from '@/constants/chat';
 import { ChatCompletionRequestMessage, ChatCompletionRequestMessageRoleEnum } from 'openai';
-
+import { OpenAiChatEnum } from '@/constants/model';
 import Graphemer from 'graphemer';

 const textDecoder = new TextDecoder();
@@ -52,7 +52,7 @@ export function countOpenAIToken({
  model
 }: {
  messages: ChatItemSimpleType[];
-  model: 'gpt-3.5-turbo' | 'gpt-4' | 'gpt-4-32k';
+  model: `${OpenAiChatEnum}`;
 }) {
  function getChatGPTEncodingText(
    messages: { role: 'system' | 'user' | 'assistant'; content: string; name?: string }[],
@@ -104,3 +104,18 @@ export function countOpenAIToken({

  return text2TokensLen(getOpenAiEncMap()[model], getChatGPTEncodingText(adaptMessages, model));
 }
+
+export const openAiSliceTextByToken = ({
+  model = 'gpt-3.5-turbo',
+  text,
+  length
+}: {
+  model: `${OpenAiChatEnum}`;
+  text: string;
+  length: number;
+}) => {
+  const enc = getOpenAiEncMap()[model];
+  const encodeText = enc.encode(text);
+  const decoder = new TextDecoder();
+  return decoder.decode(enc.decode(encodeText.slice(0, length)));
+};