perf: max_token count;feat: support resoner output;fix: member scroll (#3681)

* perf: supplement assistant empty response

* check array

* perf: max_token count

* feat: support resoner output

* member scroll

* update provider order

* i18n

docSite/content/zh-cn/docs/development/upgrading/4820.md

@@ -31,9 +31,14 @@ curl --location --request POST 'https://{{host}}/api/admin/initv4820' \
 ## 完整更新内容
 
 1. 新增 - 可视化模型参数配置。预设超过 100 个模型配置。同时支持所有类型模型的一键测试。（预计下个版本会完全支持在页面上配置渠道）。
-2. 新增 - 使用记录导出和仪表盘。
+2. 新增 - DeepSeek resoner 模型支持输出思考过程。
-3. 新增 - markdown 语法扩展，支持音视频（代码块 audio 和 video）。
+3. 新增 - 使用记录导出和仪表盘。
-4. 优化 - 页面组件抽离，减少页面组件路由。
+4. 新增 - markdown 语法扩展，支持音视频（代码块 audio 和 video）。
-5. 优化 - 全文检索，忽略大小写。
+5. 新增 - 调整 max_tokens 计算逻辑。优先保证 max_tokens 为配置值，如超出最大上下文，则减少历史记录。例如：如果申请 8000 的 max_tokens，则上下文长度会减少 8000。
-6. 优化 - 问答生成和增强索引改成流输出，避免部分模型超时。
+6. 优化 - 问题优化增加上下文过滤，避免超出上下文。
-7. 优化 - 自动给 assistant 空 content，补充 null，同时合并连续的 text assistant，避免部分模型抛错。
+7. 优化 - 页面组件抽离，减少页面组件路由。
+8. 优化 - 全文检索，忽略大小写。
+9. 优化 - 问答生成和增强索引改成流输出，避免部分模型超时。
+10. 优化 - 自动给 assistant 空 content，补充 null，同时合并连续的 text assistant，避免部分模型抛错。
+11. 优化 - 调整图片 Host， 取消上传时补充 FE_DOMAIN，改成发送对话前
+12. 修复 - 部分场景成员列表无法触底加载。

packages/global/core/ai/model.d.ts

@@ -29,10 +29,11 @@ export type LLMModelItemType = PriceType &
     maxContext: number;
     maxResponse: number;
     quoteMaxToken: number;
-    maxTemperature: number;
+    maxTemperature?: number;
 
     censor?: boolean;
     vision?: boolean;
+    reasoning?: boolean;
 
     // diff function model
     datasetProcess?: boolean; // dataset

packages/global/core/ai/provider.ts

@@ -11,8 +11,8 @@ export type ModelProviderIdType =
   | 'AliCloud'
   | 'Qwen'
   | 'Doubao'
-  | 'ChatGLM'
   | 'DeepSeek'
+  | 'ChatGLM'
   | 'Ernie'
   | 'Moonshot'
   | 'MiniMax'

packages/global/core/app/type.d.ts

@@ -117,6 +117,7 @@ export type SettingAIDataType = {
   isResponseAnswerText?: boolean;
   maxHistories?: number;
   [NodeInputKeyEnum.aiChatVision]?: boolean; // Is open vision mode
+  [NodeInputKeyEnum.aiChatReasoning]?: boolean; // Is open reasoning mode
 };
 
 // variable

packages/global/core/chat/constants.ts

@@ -25,7 +25,8 @@ export enum ChatItemValueTypeEnum {
   text = 'text',
   file = 'file',
   tool = 'tool',
-  interactive = 'interactive'
+  interactive = 'interactive',
+  reasoning = 'reasoning'
 }
 
 export enum ChatSourceEnum {

packages/global/core/chat/type.d.ts

@@ -70,14 +70,23 @@ export type SystemChatItemType = {
   obj: ChatRoleEnum.System;
   value: SystemChatItemValueItemType[];
 };
+
 export type AIChatItemValueItemType = {
-  type: ChatItemValueTypeEnum.text | ChatItemValueTypeEnum.tool | ChatItemValueTypeEnum.interactive;
+  type:
+    | ChatItemValueTypeEnum.text
+    | ChatItemValueTypeEnum.reasoning
+    | ChatItemValueTypeEnum.tool
+    | ChatItemValueTypeEnum.interactive;
   text?: {
     content: string;
   };
+  reasoning?: {
+    content: string;
+  };
   tools?: ToolModuleResponseItemType[];
   interactive?: WorkflowInteractiveResponseType;
 };
+
 export type AIChatItemType = {
   obj: ChatRoleEnum.AI;
   value: AIChatItemValueItemType[];

packages/global/core/workflow/constants.ts

@@ -141,6 +141,7 @@ export enum NodeInputKeyEnum {
   aiChatDatasetQuote = 'quoteQA',
   aiChatVision = 'aiChatVision',
   stringQuoteText = 'stringQuoteText',
+  aiChatReasoning = 'aiChatReasoning',
 
   // dataset
   datasetSelectList = 'datasets',
@@ -220,7 +221,8 @@ export enum NodeOutputKeyEnum {
   // common
   userChatInput = 'userChatInput',
   history = 'history',
-  answerText = 'answerText', // module answer. the value will be show and save to history
+  answerText = 'answerText', // node answer. the value will be show and save to history
+  reasoningText = 'reasoningText', // node reasoning. the value will be show but not save to history
   success = 'success',
   failed = 'failed',
   error = 'error',

packages/global/core/workflow/runtime/type.d.ts

@@ -220,6 +220,7 @@ export type AIChatNodeProps = {
   [NodeInputKeyEnum.aiChatMaxToken]?: number;
   [NodeInputKeyEnum.aiChatIsResponseText]: boolean;
   [NodeInputKeyEnum.aiChatVision]?: boolean;
+  [NodeInputKeyEnum.aiChatReasoning]?: boolean;
 
   [NodeInputKeyEnum.aiChatQuoteRole]?: AiChatQuoteRoleType;
   [NodeInputKeyEnum.aiChatQuoteTemplate]?: string;

packages/global/core/workflow/runtime/utils.ts

@@ -364,12 +364,14 @@ export function replaceEditorVariable({
 
 export const textAdaptGptResponse = ({
   text,
+  reasoning_content,
   model = '',
   finish_reason = null,
   extraData = {}
 }: {
   model?: string;
-  text: string | null;
+  text?: string | null;
+  reasoning_content?: string | null;
   finish_reason?: null | 'stop';
   extraData?: Object;
 }) => {
@@ -381,10 +383,11 @@ export const textAdaptGptResponse = ({
     model,
     choices: [
       {
-        delta:
+        delta: {
-          text === null
+          role: ChatCompletionRequestMessageRoleEnum.Assistant,
-            ? {}
+          content: text,
-            : { role: ChatCompletionRequestMessageRoleEnum.Assistant, content: text },
+          ...(reasoning_content && { reasoning_content })
+        },
         index: 0,
         finish_reason
       }

packages/global/core/workflow/template/system/aiChat/index.ts

@@ -63,14 +63,14 @@ export const AiChatModule: FlowNodeTemplateType = {
       key: NodeInputKeyEnum.aiChatTemperature,
       renderTypeList: [FlowNodeInputTypeEnum.hidden], // Set in the pop-up window
       label: '',
-      value: 0,
+      value: undefined,
       valueType: WorkflowIOValueTypeEnum.number
     },
     {
       key: NodeInputKeyEnum.aiChatMaxToken,
       renderTypeList: [FlowNodeInputTypeEnum.hidden], // Set in the pop-up window
       label: '',
-      value: 2000,
+      value: undefined,
       valueType: WorkflowIOValueTypeEnum.number
     },
 
@@ -91,6 +91,13 @@ export const AiChatModule: FlowNodeTemplateType = {
       valueType: WorkflowIOValueTypeEnum.boolean,
       value: true
     },
+    {
+      key: NodeInputKeyEnum.aiChatReasoning,
+      renderTypeList: [FlowNodeInputTypeEnum.hidden],
+      label: '',
+      valueType: WorkflowIOValueTypeEnum.boolean,
+      value: true
+    },
     // settings modal ---
     {
       ...Input_Template_System_Prompt,

packages/global/core/workflow/template/system/tools.ts

@@ -43,14 +43,14 @@ export const ToolModule: FlowNodeTemplateType = {
       key: NodeInputKeyEnum.aiChatTemperature,
       renderTypeList: [FlowNodeInputTypeEnum.hidden], // Set in the pop-up window
       label: '',
-      value: 0,
+      value: undefined,
       valueType: WorkflowIOValueTypeEnum.number
     },
     {
       key: NodeInputKeyEnum.aiChatMaxToken,
       renderTypeList: [FlowNodeInputTypeEnum.hidden], // Set in the pop-up window
       label: '',
-      value: 2000,
+      value: undefined,
       valueType: WorkflowIOValueTypeEnum.number
     },
     {

packages/service/common/file/image/controller.ts

@@ -40,7 +40,7 @@ export async function uploadMongoImg({
     expiredTime: forever ? undefined : addHours(new Date(), 1)
   });
 
-  return `${process.env.FE_DOMAIN || ''}${process.env.NEXT_PUBLIC_BASE_URL || ''}${imageBaseUrl}${String(_id)}.${extension}`;
+  return `${process.env.NEXT_PUBLIC_BASE_URL || ''}${imageBaseUrl}${String(_id)}.${extension}`;
 }
 
 const getIdFromPath = (path?: string) => {

packages/service/core/ai/config/provider/DeepSeek.json

@@ -27,8 +27,9 @@
       "maxContext": 64000,
       "maxResponse": 4096,
       "quoteMaxToken": 60000,
-      "maxTemperature": 1.5,
+      "maxTemperature": null,
       "vision": false,
+      "reasoning": true,
       "toolChoice": false,
       "functionCall": false,
       "defaultSystemChatPrompt": "",
@@ -39,9 +40,7 @@
       "usedInQueryExtension": true,
       "customExtractPrompt": "",
       "usedInToolCall": true,
-      "defaultConfig": {
+      "defaultConfig": {},
-        "temperature": null
-      },
       "fieldMap": {},
       "type": "llm"
     }

packages/service/core/ai/config/provider/OpenAI.json

@@ -50,10 +50,10 @@
       "maxContext": 128000,
       "maxResponse": 4000,
       "quoteMaxToken": 120000,
-      "maxTemperature": 1.2,
+      "maxTemperature": null,
       "vision": false,
       "toolChoice": false,
-      "functionCall": true,
+      "functionCall": false,
       "defaultSystemChatPrompt": "",
       "datasetProcess": true,
       "usedInClassify": true,
@@ -63,8 +63,10 @@
       "customExtractPrompt": "",
       "usedInToolCall": true,
       "defaultConfig": {
-        "temperature": 1,
+        "stream": false
-        "max_tokens": null
+      },
+      "fieldMap": {
+        "max_tokens": "max_completion_tokens"
       },
       "type": "llm"
     },
@@ -74,10 +76,10 @@
       "maxContext": 128000,
       "maxResponse": 4000,
       "quoteMaxToken": 120000,
-      "maxTemperature": 1.2,
+      "maxTemperature": null,
       "vision": false,
       "toolChoice": false,
-      "functionCall": true,
+      "functionCall": false,
       "defaultSystemChatPrompt": "",
       "datasetProcess": true,
       "usedInClassify": true,
@@ -87,10 +89,11 @@
       "customExtractPrompt": "",
       "usedInToolCall": true,
       "defaultConfig": {
-        "temperature": 1,
-        "max_tokens": null,
         "stream": false
       },
+      "fieldMap": {
+        "max_tokens": "max_completion_tokens"
+      },
       "type": "llm"
     },
     {
@@ -99,10 +102,10 @@
       "maxContext": 195000,
       "maxResponse": 8000,
       "quoteMaxToken": 120000,
-      "maxTemperature": 1.2,
+      "maxTemperature": null,
       "vision": false,
       "toolChoice": false,
-      "functionCall": true,
+      "functionCall": false,
       "defaultSystemChatPrompt": "",
       "datasetProcess": true,
       "usedInClassify": true,
@@ -112,10 +115,11 @@
       "customExtractPrompt": "",
       "usedInToolCall": true,
       "defaultConfig": {
-        "temperature": 1,
-        "max_tokens": null,
         "stream": false
       },
+      "fieldMap": {
+        "max_tokens": "max_completion_tokens"
+      },
       "type": "llm"
     },
     {

packages/service/core/ai/functions/queryExtension.ts

@@ -2,10 +2,12 @@ import { replaceVariable } from '@fastgpt/global/common/string/tools';
 import { createChatCompletion } from '../config';
 import { ChatItemType } from '@fastgpt/global/core/chat/type';
 import { countGptMessagesTokens, countPromptTokens } from '../../../common/string/tiktoken/index';
-import { chatValue2RuntimePrompt } from '@fastgpt/global/core/chat/adapt';
+import { chats2GPTMessages } from '@fastgpt/global/core/chat/adapt';
 import { getLLMModel } from '../model';
 import { llmCompletionsBodyFormat } from '../utils';
 import { addLog } from '../../../common/system/log';
+import { filterGPTMessageByMaxContext } from '../../chat/utils';
+import json5 from 'json5';
 
 /* 
     query extension - 问题扩展
@@ -13,72 +15,73 @@ import { addLog } from '../../../common/system/log';
 */
 
 const title = global.feConfigs?.systemTitle || 'FastAI';
-const defaultPrompt = `作为一个向量检索助手，你的任务是结合历史记录，从不同角度，为“原问题”生成个不同版本的“检索词”，从而提高向量检索的语义丰富度，提高向量检索的精度。
+const defaultPrompt = `## 你的任务
+你作为一个向量检索助手，你的任务是结合历史记录，从不同角度，为“原问题”生成个不同版本的“检索词”，从而提高向量检索的语义丰富度，提高向量检索的精度。
 生成的问题要求指向对象清晰明确，并与“原问题语言相同”。
 
-参考 <Example></Example> 标中的示例来完成任务。
+## 参考示例
 
-<Example>
 历史记录: 
 """
+null
 """
 原问题: 介绍下剧情。
 检索词: ["介绍下故事的背景。","故事的主题是什么？","介绍下故事的主要人物。"]
 ----------------
 历史记录: 
 """
-Q: 对话背景。
+user: 对话背景。
-A: 当前对话是关于 Nginx 的介绍和使用等。
+assistant: 当前对话是关于 Nginx 的介绍和使用等。
 """
 原问题: 怎么下载
 检索词: ["Nginx 如何下载？","下载 Nginx 需要什么条件？","有哪些渠道可以下载 Nginx？"]
 ----------------
 历史记录: 
 """
-Q: 对话背景。
+user: 对话背景。
-A: 当前对话是关于 Nginx 的介绍和使用等。
+assistant: 当前对话是关于 Nginx 的介绍和使用等。
-Q: 报错 "no connection"
+user: 报错 "no connection"
-A: 报错"no connection"可能是因为……
+assistant: 报错"no connection"可能是因为……
 """
 原问题: 怎么解决
 检索词: ["Nginx报错"no connection"如何解决？","造成'no connection'报错的原因。","Nginx提示'no connection'，要怎么办？"]
 ----------------
 历史记录: 
 """
-Q: 护产假多少天?
+user: How long is the maternity leave?
-A: 护产假的天数根据员工所在的城市而定。请提供您所在的城市，以便我回答您的问题。
+assistant: The number of days of maternity leave depends on the city in which the employee is located. Please provide your city so that I can answer your questions.
 """
-原问题: 沈阳
+原问题: ShenYang
-检索词: ["沈阳的护产假多少天？","沈阳的护产假政策。","沈阳的护产假标准。"]
+检索词: ["How many days is maternity leave in Shenyang?","Shenyang's maternity leave policy.","The standard of maternity leave in Shenyang."]
 ----------------
 历史记录: 
 """
-Q: 作者是谁？
+user: 作者是谁？
-A: ${title} 的作者是 labring。
+assistant: ${title} 的作者是 labring。
 """
 原问题: Tell me about him
 检索词: ["Introduce labring, the author of ${title}." ," Background information on author labring." "," Why does labring do ${title}?"]
 ----------------
 历史记录:
 """
-Q: 对话背景。
+user: 对话背景。
-A: 关于 ${title} 的介绍和使用等问题。
+assistant: 关于 ${title} 的介绍和使用等问题。
 """
 原问题: 你好。
 检索词: ["你好"]
 ----------------
 历史记录:
 """
-Q: ${title} 如何收费？
+user: ${title} 如何收费？
-A: ${title} 收费可以参考……
+assistant: ${title} 收费可以参考……
 """
 原问题: 你知道 laf 么？
 检索词: ["laf 的官网地址是多少？","laf 的使用教程。","laf 有什么特点和优势。"]
 ----------------
 历史记录:
 """
-Q: ${title} 的优势
+user: ${title} 的优势
-A: 1. 开源
+assistant: 1. 开源
    2. 简便
    3. 扩展性强
 """
@@ -87,18 +90,20 @@ A: 1. 开源
 ----------------
 历史记录:
 """
-Q: 什么是 ${title}？
+user: 什么是 ${title}？
-A: ${title} 是一个 RAG 平台。
+assistant: ${title} 是一个 RAG 平台。
-Q: 什么是 Laf？
+user: 什么是 Laf？
-A: Laf 是一个云函数开发平台。
+assistant: Laf 是一个云函数开发平台。
 """
 原问题: 它们有什么关系？
 检索词: ["${title}和Laf有什么关系？","介绍下${title}","介绍下Laf"]
-</Example>
 
------
+## 输出要求
 
-下面是正式的任务：
+1. 输出格式为 JSON 数组，数组中每个元素为字符串。无需对输出进行任何解释。
+2. 输出语言与原问题相同。原问题为中文则输出中文；原问题为英文则输出英文。
+
+## 开始任务
 
 历史记录:
 """
@@ -125,26 +130,39 @@ export const queryExtension = async ({
   outputTokens: number;
 }> => {
   const systemFewShot = chatBg
-    ? `Q: 对话背景。
+    ? `user: 对话背景。
-A: ${chatBg}
+assistant: ${chatBg}
 `
     : '';
-  const historyFewShot = histories
-    .map((item) => {
-      const role = item.obj === 'Human' ? 'Q' : 'A';
-      return `${role}: ${chatValue2RuntimePrompt(item.value).text}`;
-    })
-    .join('\n');
-  const concatFewShot = `${systemFewShot}${historyFewShot}`.trim();
 
   const modelData = getLLMModel(model);
+  const filterHistories = await filterGPTMessageByMaxContext({
+    messages: chats2GPTMessages({ messages: histories, reserveId: false }),
+    maxContext: modelData.maxContext - 1000
+  });
+
+  const historyFewShot = filterHistories
+    .map((item) => {
+      const role = item.role;
+      const content = item.content;
+      if ((role === 'user' || role === 'assistant') && content) {
+        if (typeof content === 'string') {
+          return `${role}: ${content}`;
+        } else {
+          return `${role}: ${content.map((item) => (item.type === 'text' ? item.text : '')).join('\n')}`;
+        }
+      }
+    })
+    .filter(Boolean)
+    .join('\n');
+  const concatFewShot = `${systemFewShot}${historyFewShot}`.trim();
 
   const messages = [
     {
       role: 'user',
       content: replaceVariable(defaultPrompt, {
         query: `${query}`,
-        histories: concatFewShot
+        histories: concatFewShot || 'null'
       })
     }
   ] as any;
@@ -154,7 +172,7 @@ A: ${chatBg}
       {
         stream: false,
         model: modelData.model,
-        temperature: 0.01,
+        temperature: 0.1,
         messages
       },
       modelData
@@ -172,22 +190,41 @@ A: ${chatBg}
     };
   }
 
+  const start = answer.indexOf('[');
+  const end = answer.lastIndexOf(']');
+  if (start === -1 || end === -1) {
+    addLog.warn('Query extension failed, not a valid JSON', {
+      answer
+    });
+    return {
+      rawQuery: query,
+      extensionQueries: [],
+      model,
+      inputTokens: 0,
+      outputTokens: 0
+    };
+  }
+
   // Intercept the content of [] and retain []
-  answer = answer.match(/\[.*?\]/)?.[0] || '';
+  const jsonStr = answer
-  answer = answer.replace(/\\"/g, '"');
+    .substring(start, end + 1)
+    .replace(/(\\n|\\)/g, '')
+    .replace(/  /g, '');
 
   try {
-    const queries = JSON.parse(answer) as string[];
+    const queries = json5.parse(jsonStr) as string[];
 
     return {
       rawQuery: query,
-      extensionQueries: Array.isArray(queries) ? queries : [],
+      extensionQueries: (Array.isArray(queries) ? queries : []).slice(0, 5),
       model,
       inputTokens: await countGptMessagesTokens(messages),
       outputTokens: await countPromptTokens(answer)
     };
   } catch (error) {
-    addLog.error(`Query extension error`, error);
+    addLog.warn('Query extension failed, not a valid JSON', {
+      answer
+    });
     return {
       rawQuery: query,
       extensionQueries: [],

packages/service/core/ai/utils.ts

@@ -2,33 +2,23 @@ import { LLMModelItemType } from '@fastgpt/global/core/ai/model.d';
 import {
   ChatCompletionCreateParamsNonStreaming,
   ChatCompletionCreateParamsStreaming,
-  ChatCompletionMessageParam,
   StreamChatType
 } from '@fastgpt/global/core/ai/type';
-import { countGptMessagesTokens } from '../../common/string/tiktoken';
 import { getLLMModel } from './model';
 
-export const computedMaxToken = async ({
+/* 
+  Count response max token
+*/
+export const computedMaxToken = ({
   maxToken,
-  model,
+  model
-  filterMessages = []
 }: {
   maxToken?: number;
   model: LLMModelItemType;
-  filterMessages: ChatCompletionMessageParam[];
 }) => {
   if (maxToken === undefined) return;
 
   maxToken = Math.min(maxToken, model.maxResponse);
-  const tokensLimit = model.maxContext;
-
-  /* count response max token */
-  const promptsToken = await countGptMessagesTokens(filterMessages);
-  maxToken = promptsToken + maxToken > tokensLimit ? tokensLimit - promptsToken : maxToken;
-
-  if (maxToken <= 0) {
-    maxToken = 200;
-  }
   return maxToken;
 };
 
@@ -40,6 +30,7 @@ export const computedTemperature = ({
   model: LLMModelItemType;
   temperature: number;
 }) => {
+  if (typeof model.maxTemperature !== 'number') return undefined;
   temperature = +(model.maxTemperature * (temperature / 10)).toFixed(2);
   temperature = Math.max(temperature, 0.01);
 

packages/service/core/chat/utils.ts

@@ -14,36 +14,19 @@ import { serverRequestBaseUrl } from '../../common/api/serverRequest';
 import { i18nT } from '../../../web/i18n/utils';
 import { addLog } from '../../common/system/log';
 
-export const filterGPTMessageByMaxTokens = async ({
+export const filterGPTMessageByMaxContext = async ({
   messages = [],
-  maxTokens
+  maxContext
 }: {
   messages: ChatCompletionMessageParam[];
-  maxTokens: number;
+  maxContext: number;
 }) => {
   if (!Array.isArray(messages)) {
     return [];
   }
-  const rawTextLen = messages.reduce((sum, item) => {
-    if (typeof item.content === 'string') {
-      return sum + item.content.length;
-    }
-    if (Array.isArray(item.content)) {
-      return (
-        sum +
-        item.content.reduce((sum, item) => {
-          if (item.type === 'text') {
-            return sum + item.text.length;
-          }
-          return sum;
-        }, 0)
-      );
-    }
-    return sum;
-  }, 0);
 
   // If the text length is less than half of the maximum token, no calculation is required
-  if (rawTextLen < maxTokens * 0.5) {
+  if (messages.length < 4) {
     return messages;
   }
 
@@ -55,7 +38,7 @@ export const filterGPTMessageByMaxTokens = async ({
   const chatPrompts: ChatCompletionMessageParam[] = messages.slice(chatStartIndex);
 
   // reduce token of systemPrompt
-  maxTokens -= await countGptMessagesTokens(systemPrompts);
+  maxContext -= await countGptMessagesTokens(systemPrompts);
 
   // Save the last chat prompt(question)
   const question = chatPrompts.pop();
@@ -73,9 +56,9 @@ export const filterGPTMessageByMaxTokens = async ({
     }
 
     const tokens = await countGptMessagesTokens([assistant, user]);
-    maxTokens -= tokens;
+    maxContext -= tokens;
     /* 整体 tokens 超出范围，截断  */
-    if (maxTokens < 0) {
+    if (maxContext < 0) {
       break;
     }
 

packages/service/core/workflow/dispatch/agent/extract.ts

@@ -1,5 +1,5 @@
 import { chats2GPTMessages } from '@fastgpt/global/core/chat/adapt';
-import { filterGPTMessageByMaxTokens, loadRequestMessages } from '../../../chat/utils';
+import { filterGPTMessageByMaxContext, loadRequestMessages } from '../../../chat/utils';
 import type { ChatItemType } from '@fastgpt/global/core/chat/type.d';
 import {
   countMessagesTokens,
@@ -175,9 +175,9 @@ ${description ? `- ${description}` : ''}
     }
   ];
   const adaptMessages = chats2GPTMessages({ messages, reserveId: false });
-  const filterMessages = await filterGPTMessageByMaxTokens({
+  const filterMessages = await filterGPTMessageByMaxContext({
     messages: adaptMessages,
-    maxTokens: extractModel.maxContext
+    maxContext: extractModel.maxContext
   });
   const requestMessages = await loadRequestMessages({
     messages: filterMessages,

packages/service/core/workflow/dispatch/agent/runTool/functionCall.ts

@@ -1,5 +1,5 @@
 import { createChatCompletion } from '../../../../ai/config';
-import { filterGPTMessageByMaxTokens, loadRequestMessages } from '../../../../chat/utils';
+import { filterGPTMessageByMaxContext, loadRequestMessages } from '../../../../chat/utils';
 import {
   ChatCompletion,
   StreamChatType,
@@ -172,10 +172,14 @@ export const runToolWithFunctionCall = async (
     };
   });
 
+  const max_tokens = computedMaxToken({
+    model: toolModel,
+    maxToken
+  });
   const filterMessages = (
-    await filterGPTMessageByMaxTokens({
+    await filterGPTMessageByMaxContext({
       messages,
-      maxTokens: toolModel.maxContext - 300 // filter token. not response maxToken
+      maxContext: toolModel.maxContext - (max_tokens || 0) // filter token. not response maxToken
     })
   ).map((item) => {
     if (item.role === ChatCompletionRequestMessageRoleEnum.Assistant && item.function_call) {
@@ -190,16 +194,11 @@ export const runToolWithFunctionCall = async (
     }
     return item;
   });
-  const [requestMessages, max_tokens] = await Promise.all([
+  const [requestMessages] = await Promise.all([
     loadRequestMessages({
       messages: filterMessages,
       useVision: toolModel.vision && aiChatVision,
       origin: requestOrigin
-    }),
-    computedMaxToken({
-      model: toolModel,
-      maxToken,
-      filterMessages
     })
   ]);
   const requestBody = llmCompletionsBodyFormat(

packages/service/core/workflow/dispatch/agent/runTool/promptCall.ts

@@ -1,5 +1,5 @@
 import { createChatCompletion } from '../../../../ai/config';
-import { filterGPTMessageByMaxTokens, loadRequestMessages } from '../../../../chat/utils';
+import { filterGPTMessageByMaxContext, loadRequestMessages } from '../../../../chat/utils';
 import {
   ChatCompletion,
   StreamChatType,
@@ -196,21 +196,20 @@ export const runToolWithPromptCall = async (
     return Promise.reject('Prompt call invalid input');
   }
 
-  const filterMessages = await filterGPTMessageByMaxTokens({
+  const max_tokens = computedMaxToken({
+    model: toolModel,
+    maxToken
+  });
+  const filterMessages = await filterGPTMessageByMaxContext({
     messages,
-    maxTokens: toolModel.maxContext - 500 // filter token. not response maxToken
+    maxContext: toolModel.maxContext - (max_tokens || 0) // filter token. not response maxToken
   });
 
-  const [requestMessages, max_tokens] = await Promise.all([
+  const [requestMessages] = await Promise.all([
     loadRequestMessages({
       messages: filterMessages,
       useVision: toolModel.vision && aiChatVision,
       origin: requestOrigin
-    }),
-    computedMaxToken({
-      model: toolModel,
-      maxToken,
-      filterMessages
     })
   ]);
   const requestBody = llmCompletionsBodyFormat(

packages/service/core/workflow/dispatch/agent/runTool/toolChoice.ts

@@ -1,5 +1,5 @@
 import { createChatCompletion } from '../../../../ai/config';
-import { filterGPTMessageByMaxTokens, loadRequestMessages } from '../../../../chat/utils';
+import { filterGPTMessageByMaxContext, loadRequestMessages } from '../../../../chat/utils';
 import {
   ChatCompletion,
   ChatCompletionMessageToolCall,
@@ -228,11 +228,16 @@ export const runToolWithToolChoice = async (
     };
   });
 
+  const max_tokens = computedMaxToken({
+    model: toolModel,
+    maxToken
+  });
+
   // Filter histories by maxToken
   const filterMessages = (
-    await filterGPTMessageByMaxTokens({
+    await filterGPTMessageByMaxContext({
       messages,
-      maxTokens: toolModel.maxContext - 300 // filter token. not response maxToken
+      maxContext: toolModel.maxContext - (max_tokens || 0) // filter token. not response maxToken
     })
   ).map((item) => {
     if (item.role === 'assistant' && item.tool_calls) {
@@ -248,16 +253,11 @@ export const runToolWithToolChoice = async (
     return item;
   });
 
-  const [requestMessages, max_tokens] = await Promise.all([
+  const [requestMessages] = await Promise.all([
     loadRequestMessages({
       messages: filterMessages,
       useVision: toolModel.vision && aiChatVision,
       origin: requestOrigin
-    }),
-    computedMaxToken({
-      model: toolModel,
-      maxToken,
-      filterMessages
     })
   ]);
   const requestBody = llmCompletionsBodyFormat(

packages/service/core/workflow/dispatch/chat/oneapi.ts

@@ -1,5 +1,5 @@
 import type { NextApiResponse } from 'next';
-import { filterGPTMessageByMaxTokens, loadRequestMessages } from '../../../chat/utils';
+import { filterGPTMessageByMaxContext, loadRequestMessages } from '../../../chat/utils';
 import type { ChatItemType, UserChatItemValueItemType } from '@fastgpt/global/core/chat/type.d';
 import { ChatRoleEnum } from '@fastgpt/global/core/chat/constants';
 import { SseResponseEventEnum } from '@fastgpt/global/core/workflow/runtime/constants';
@@ -58,6 +58,7 @@ export type ChatProps = ModuleDispatchProps<
 >;
 export type ChatResponse = DispatchNodeResultType<{
   [NodeOutputKeyEnum.answerText]: string;
+  [NodeOutputKeyEnum.reasoningText]?: string;
   [NodeOutputKeyEnum.history]: ChatItemType[];
 }>;
 
@@ -87,22 +88,24 @@ export const dispatchChatCompletion = async (props: ChatProps): Promise<ChatResp
       quoteTemplate,
       quotePrompt,
       aiChatVision,
+      aiChatReasoning,
       fileUrlList: fileLinks, // node quote file links
       stringQuoteText //abandon
     }
   } = props;
   const { files: inputFiles } = chatValue2RuntimePrompt(query); // Chat box input files
 
-  stream = stream && isResponseAnswerText;
-
-  const chatHistories = getHistories(history, histories);
-  quoteQA = checkQuoteQAValue(quoteQA);
-
   const modelConstantsData = getLLMModel(model);
   if (!modelConstantsData) {
     return Promise.reject('The chat model is undefined, you need to select a chat model.');
   }
 
+  stream = stream && isResponseAnswerText;
+  aiChatReasoning = !!aiChatReasoning && !!modelConstantsData.reasoning;
+
+  const chatHistories = getHistories(history, histories);
+  quoteQA = checkQuoteQAValue(quoteQA);
+
   const [{ datasetQuoteText }, { documentQuoteText, userFiles }] = await Promise.all([
     filterDatasetQuote({
       quoteQA,
@@ -124,9 +127,15 @@ export const dispatchChatCompletion = async (props: ChatProps): Promise<ChatResp
     return Promise.reject(i18nT('chat:AI_input_is_empty'));
   }
 
+  const max_tokens = computedMaxToken({
+    model: modelConstantsData,
+    maxToken
+  });
+
   const [{ filterMessages }] = await Promise.all([
     getChatMessages({
       model: modelConstantsData,
+      maxTokens: max_tokens,
       histories: chatHistories,
       useDatasetQuote: quoteQA !== undefined,
       datasetQuoteText,
@@ -137,8 +146,8 @@ export const dispatchChatCompletion = async (props: ChatProps): Promise<ChatResp
       userFiles,
       documentQuoteText
     }),
+    // Censor = true and system key, will check content
     (() => {
-      // censor model and system key
       if (modelConstantsData.censor && !externalProvider.openaiAccount?.key) {
         return postTextCensor({
           text: `${systemPrompt}
@@ -149,18 +158,11 @@ export const dispatchChatCompletion = async (props: ChatProps): Promise<ChatResp
     })()
   ]);
 
-  const [requestMessages, max_tokens] = await Promise.all([
+  const requestMessages = await loadRequestMessages({
-    loadRequestMessages({
+    messages: filterMessages,
-      messages: filterMessages,
+    useVision: modelConstantsData.vision && aiChatVision,
-      useVision: modelConstantsData.vision && aiChatVision,
+    origin: requestOrigin
-      origin: requestOrigin
+  });
-    }),
-    computedMaxToken({
-      model: modelConstantsData,
-      maxToken,
-      filterMessages
-    })
-  ]);
 
   const requestBody = llmCompletionsBodyFormat(
     {
@@ -183,34 +185,42 @@ export const dispatchChatCompletion = async (props: ChatProps): Promise<ChatResp
     }
   });
 
-  const { answerText } = await (async () => {
+  const { answerText, reasoningText } = await (async () => {
     if (res && isStreamResponse) {
       // sse response
-      const { answer } = await streamResponse({
+      const { answer, reasoning } = await streamResponse({
         res,
         stream: response,
+        aiChatReasoning,
         workflowStreamResponse
       });
 
       return {
-        answerText: answer
+        answerText: answer,
+        reasoningText: reasoning
       };
     } else {
       const unStreamResponse = response as ChatCompletion;
       const answer = unStreamResponse.choices?.[0]?.message?.content || '';
-
+      const reasoning = aiChatReasoning
+        ? // @ts-ignore
+          unStreamResponse.choices?.[0]?.message?.reasoning_content || ''
+        : '';
       if (stream) {
         // Some models do not support streaming
-        workflowStreamResponse?.({
+        reasoning &&
-          event: SseResponseEventEnum.fastAnswer,
+          workflowStreamResponse?.({
-          data: textAdaptGptResponse({
+            event: SseResponseEventEnum.fastAnswer,
-            text: answer
+            data: textAdaptGptResponse({
-          })
+              text: answer,
-        });
+              reasoning_content: reasoning
+            })
+          });
       }
 
       return {
-        answerText: answer
+        answerText: answer,
+        reasoningText: reasoning
       };
     }
   })();
@@ -241,6 +251,7 @@ export const dispatchChatCompletion = async (props: ChatProps): Promise<ChatResp
 
   return {
     answerText,
+    reasoningText,
     [DispatchNodeResponseKeyEnum.nodeResponse]: {
       totalPoints: externalProvider.openaiAccount?.key ? 0 : totalPoints,
       model: modelName,
@@ -367,6 +378,7 @@ async function getMultiInput({
 
 async function getChatMessages({
   model,
+  maxTokens = 0,
   aiChatQuoteRole,
   datasetQuotePrompt = '',
   datasetQuoteText,
@@ -378,6 +390,7 @@ async function getChatMessages({
   documentQuoteText
 }: {
   model: LLMModelItemType;
+  maxTokens?: number;
   // dataset quote
   aiChatQuoteRole: AiChatQuoteRoleType; // user: replace user prompt; system: replace system prompt
   datasetQuotePrompt?: string;
@@ -444,9 +457,9 @@ async function getChatMessages({
 
   const adaptMessages = chats2GPTMessages({ messages, reserveId: false });
 
-  const filterMessages = await filterGPTMessageByMaxTokens({
+  const filterMessages = await filterGPTMessageByMaxContext({
     messages: adaptMessages,
-    maxTokens: model.maxContext - 300 // filter token. not response maxToken
+    maxContext: model.maxContext - maxTokens // filter token. not response maxToken
   });
 
   return {
@@ -457,33 +470,43 @@ async function getChatMessages({
 async function streamResponse({
   res,
   stream,
-  workflowStreamResponse
+  workflowStreamResponse,
+  aiChatReasoning
 }: {
   res: NextApiResponse;
   stream: StreamChatType;
   workflowStreamResponse?: WorkflowResponseType;
+  aiChatReasoning?: boolean;
 }) {
   const write = responseWriteController({
     res,
     readStream: stream
   });
   let answer = '';
+  let reasoning = '';
   for await (const part of stream) {
     if (res.closed) {
       stream.controller?.abort();
       break;
     }
+
     const content = part.choices?.[0]?.delta?.content || '';
     answer += content;
 
+    const reasoningContent = aiChatReasoning
+      ? part.choices?.[0]?.delta?.reasoning_content || ''
+      : '';
+    reasoning += reasoningContent;
+
     workflowStreamResponse?.({
       write,
       event: SseResponseEventEnum.answer,
       data: textAdaptGptResponse({
-        text: content
+        text: content,
+        reasoning_content: reasoningContent
       })
     });
   }
 
-  return { answer };
+  return { answer, reasoning };
 }

packages/service/core/workflow/dispatch/index.ts

@@ -204,6 +204,7 @@ export async function dispatchWorkFlow(data: Props): Promise<DispatchFlowRespons
     { inputs = [] }: RuntimeNodeItemType,
     {
       answerText = '',
+      reasoningText,
       responseData,
       nodeDispatchUsages,
       toolResponses,
@@ -213,6 +214,7 @@ export async function dispatchWorkFlow(data: Props): Promise<DispatchFlowRespons
     }: Omit<
       DispatchNodeResultType<{
         [NodeOutputKeyEnum.answerText]?: string;
+        [NodeOutputKeyEnum.reasoningText]?: string;
         [DispatchNodeResponseKeyEnum.nodeResponse]?: ChatHistoryItemResType;
       }>,
       'nodeResponse'
@@ -251,6 +253,13 @@ export async function dispatchWorkFlow(data: Props): Promise<DispatchFlowRespons
           }
         });
       }
+    } else if (reasoningText) {
+      chatAssistantResponse.push({
+        type: ChatItemValueTypeEnum.reasoning,
+        reasoning: {
+          content: reasoningText
+        }
+      });
     }
 
     if (rewriteHistories) {

packages/web/components/common/Icon/constants.ts

@@ -178,6 +178,7 @@ export const iconPaths = {
   'core/chat/sideLine': () => import('./icons/core/chat/sideLine.svg'),
   'core/chat/speaking': () => import('./icons/core/chat/speaking.svg'),
   'core/chat/stopSpeech': () => import('./icons/core/chat/stopSpeech.svg'),
+  'core/chat/think': () => import('./icons/core/chat/think.svg'),
   'core/dataset/commonDataset': () => import('./icons/core/dataset/commonDataset.svg'),
   'core/dataset/commonDatasetColor': () => import('./icons/core/dataset/commonDatasetColor.svg'),
   'core/dataset/commonDatasetOutline': () =>

packages/web/components/common/Icon/icons/core/chat/think.svg

@@ -0,0 +1 @@
+<svg t="1737983662269" class="icon" viewBox="0 0 1024 1024" version="1.1" xmlns="http://www.w3.org/2000/svg" p-id="6134" width="64" height="64"><path d="M512 512m-91.264 0a91.264 91.264 0 1 0 182.528 0 91.264 91.264 0 1 0-182.528 0Z" fill="" p-id="6135"></path><path d="M256.341333 693.546667l-20.138666-5.12C86.101333 650.496 0 586.112 0 511.829333s86.101333-138.666667 236.202667-176.597333l20.138666-5.077333 5.674667 19.968a1003.946667 1003.946667 0 0 0 58.154667 152.661333l4.309333 9.088-4.309333 9.088a994.432 994.432 0 0 0-58.154667 152.661333l-5.674667 19.925334zM226.858667 381.866667c-114.090667 32.042667-184.106667 81.066667-184.106667 129.962666 0 48.853333 70.016 97.877333 184.106667 129.962667a1064.533333 1064.533333 0 0 1 50.432-129.962667A1056.085333 1056.085333 0 0 1 226.858667 381.866667z m540.8 311.68l-5.674667-20.010667a996.565333 996.565333 0 0 0-58.197333-152.618667l-4.309334-9.088 4.309334-9.088a999.253333 999.253333 0 0 0 58.197333-152.661333l5.674667-19.968 20.181333 5.077333c150.058667 37.930667 236.16 102.314667 236.16 176.64s-86.101333 138.666667-236.16 176.597334l-20.181333 5.12z m-20.949334-181.717334c20.48 44.330667 37.418667 87.893333 50.432 129.962667 114.133333-32.085333 184.106667-81.109333 184.106667-129.962667 0-48.896-70.016-97.877333-184.106667-129.962666a1057.621333 1057.621333 0 0 1-50.432 129.962666z" fill="" p-id="6136"></path><path d="M226.56 381.653333l-5.674667-19.925333C178.688 212.992 191.488 106.410667 256 69.205333c63.274667-36.522667 164.864 6.613333 271.317333 115.882667l14.506667 14.890667-14.506667 14.890666a1004.885333 1004.885333 0 0 0-103.338666 126.592l-5.76 8.234667-10.026667 0.853333a1009.365333 1009.365333 0 0 0-161.493333 26.026667l-20.138667 5.077333z m80.896-282.88c-11.434667 0-21.546667 2.474667-30.08 7.381334-42.410667 24.448-49.92 109.44-20.693333 224.128a1071.872 1071.872 0 0 1 137.941333-21.376 1060.138667 1060.138667 0 0 1 87.552-108.544c-66.56-64.810667-129.578667-101.589333-174.72-101.589334z m409.130667 868.778667c-0.042667 0-0.042667 0 0 0-60.8 0-138.88-45.781333-219.904-128.981333l-14.506667-14.890667 14.506667-14.890667a1003.946667 1003.946667 0 0 0 103.296-126.634666l5.76-8.234667 9.984-0.853333a1008.213333 1008.213333 0 0 0 161.578666-25.984l20.138667-5.077334 5.717333 19.968c42.112 148.650667 29.354667 255.274667-35.157333 292.437334a101.546667 101.546667 0 0 1-51.413333 13.141333z m-174.762667-144.256c66.56 64.810667 129.578667 101.589333 174.72 101.589333h0.042667c11.392 0 21.546667-2.474667 30.037333-7.381333 42.410667-24.448 49.962667-109.482667 20.693333-224.170667a1067.52 1067.52 0 0 1-137.984 21.376 1052.757333 1052.757333 0 0 1-87.509333 108.586667z" fill="" p-id="6137"></path><path d="M797.44 381.653333l-20.138667-5.077333a1001.770667 1001.770667 0 0 0-161.578666-26.026667l-9.984-0.853333-5.76-8.234667a998.997333 998.997333 0 0 0-103.296-126.592l-14.506667-14.890666 14.506667-14.890667C603.093333 75.861333 704.64 32.725333 768 69.205333c64.512 37.205333 77.312 143.786667 35.157333 292.48l-5.717333 19.968zM629.333333 308.906667c48.725333 4.437333 95.018667 11.648 137.984 21.376 29.269333-114.688 21.717333-199.68-20.693333-224.128-42.154667-24.362667-121.386667 12.970667-204.8 94.208A1060.224 1060.224 0 0 1 629.333333 308.906667zM307.456 967.552A101.546667 101.546667 0 0 1 256 954.410667c-64.512-37.162667-77.312-143.744-35.114667-292.437334l5.632-19.968 20.138667 5.077334c49.28 12.416 103.637333 21.162667 161.493333 25.984l10.026667 0.853333 5.717333 8.234667a1006.762667 1006.762667 0 0 0 103.338667 126.634666l14.506667 14.890667-14.506667 14.890667c-80.981333 83.2-159.061333 128.981333-219.776 128.981333z m-50.773333-274.218667c-29.269333 114.688-21.717333 199.722667 20.693333 224.170667 42.112 24.021333 121.301333-13.013333 204.8-94.208a1066.581333 1066.581333 0 0 1-87.552-108.586667 1065.642667 1065.642667 0 0 1-137.941333-21.376z" fill="" p-id="6138"></path><path d="M512 720.128c-35.114667 0-71.210667-1.536-107.349333-4.522667l-10.026667-0.853333-5.76-8.234667a1296.554667 1296.554667 0 0 1-57.6-90.538666 1295.104 1295.104 0 0 1-49.749333-95.061334l-4.266667-9.088 4.266667-9.088a1292.8 1292.8 0 0 1 49.749333-95.061333c17.664-30.549333 37.077333-61.013333 57.6-90.538667l5.76-8.234666 10.026667-0.853334a1270.826667 1270.826667 0 0 1 214.741333 0l9.984 0.853334 5.717333 8.234666a1280.256 1280.256 0 0 1 107.392 185.6l4.309334 9.088-4.309334 9.088a1262.933333 1262.933333 0 0 1-107.392 185.6l-5.717333 8.234667-9.984 0.853333c-36.138667 2.986667-72.277333 4.522667-107.392 4.522667z m-93.738667-46.250667c63.146667 4.736 124.330667 4.736 187.52 0a1237.589333 1237.589333 0 0 0 93.696-162.048 1219.626667 1219.626667 0 0 0-93.738666-162.048 1238.656 1238.656 0 0 0-187.477334 0 1215.018667 1215.018667 0 0 0-93.738666 162.048 1242.197333 1242.197333 0 0 0 93.738666 162.048z"  p-id="6139"></path></svg>

packages/web/hooks/useScrollPagination.tsx

@@ -304,7 +304,7 @@ export function useScrollPagination<
       );
 
       return (
-        <MyBox {...props} ref={ref} overflow={'overlay'} isLoading={isLoading}>
+        <MyBox ref={ref} h={'100%'} overflow={'auto'} isLoading={isLoading} {...props}>
           {scrollLoadType === 'top' && total > 0 && isLoading && (
             <Box mt={2} fontSize={'xs'} color={'blackAlpha.500'} textAlign={'center'}>
               {t('common:common.is_requesting')}

packages/web/i18n/en/account.json

@@ -49,10 +49,10 @@
   "model.output_price": "Output price",
   "model.output_price_tip": "The language model output price. If this item is configured, the model comprehensive price will be invalid.",
   "model.param_name": "Parameter name",
-  "model.request_auth": "Custom token",
+  "model.request_auth": "Custom key",
   "model.request_auth_tip": "When making a request to a custom request address, carry the request header: Authorization: Bearer xxx to make the request.",
   "model.request_url": "Custom url",
-  "model.request_url_tip": "If this value is filled in, a request will be made directly to this address without going through OneAPI",
+  "model.request_url_tip": "If you fill in this value, you will initiate a request directly without passing. \nYou need to follow the API format of Openai and fill in the full request address, such as\n\nLLM: {Host}}/v1/Chat/Completions\n\nEmbedding: {host}}/v1/embeddings\n\nSTT: {Host}/v1/Audio/Transcriptions\n\nTTS: {Host}}/v1/Audio/Speech\n\nRERARARARARARARANK: {Host}}/v1/RERARARARARARARARARARANK",
   "model.test_model": "Model testing",
   "model.tool_choice": "Tool choice",
   "model.tool_choice_tag": "ToolCall",

packages/web/i18n/en/app.json

@@ -109,6 +109,7 @@
   "publish_channel": "Publish",
   "publish_success": "Publish Successful",
   "question_guide_tip": "After the conversation, 3 guiding questions will be generated for you.",
+  "reasoning_response": "Output thinking",
   "saved_success": "Saved successfully! \nTo use this version externally, click Save and Publish",
   "search_app": "Search apps",
   "setting_app": "Workflow",

packages/web/i18n/en/chat.json

@@ -2,6 +2,7 @@
   "AI_input_is_empty": "The content passed to the AI ​​node is empty",
   "Delete_all": "Clear All Lexicon",
   "LLM_model_response_empty": "The model flow response is empty, please check whether the model flow output is normal.",
+  "ai_reasoning": "Thinking process",
   "chat_history": "Conversation History",
   "chat_input_guide_lexicon_is_empty": "Lexicon not configured yet",
   "chat_test_app": "Debug-{{name}}",

packages/web/i18n/zh-CN/account.json

@@ -49,10 +49,10 @@
   "model.output_price": "模型输出价格",
   "model.output_price_tip": "语言模型输出价格，如果配置了该项，则模型综合价格会失效",
   "model.param_name": "参数名",
-  "model.request_auth": "自定义请求 Tokens",
+  "model.request_auth": "自定义请求 Key",
   "model.request_auth_tip": "向自定义请求地址发起请求时候，携带请求头：Authorization: Bearer xxx 进行请求",
   "model.request_url": "自定义请求地址",
-  "model.request_url_tip": "如果填写该值，则会直接向该地址发起请求，不经过 OneAPI",
+  "model.request_url_tip": "如果填写该值，则会直接向该地址发起请求，不经过 OneAPI。需要遵循 OpenAI 的 API格式，并填写完整请求地址，例如：\nLLM: {{host}}/v1/chat/completions\nEmbedding: {{host}}/v1/embeddings\nSTT: {{host}}/v1/audio/transcriptions\nTTS: {{host}}/v1/audio/speech\nRerank: {{host}}/v1/rerank",
   "model.test_model": "模型测试",
   "model.tool_choice": "支持工具调用",
   "model.tool_choice_tag": "工具调用",

packages/web/i18n/zh-CN/app.json

@@ -109,6 +109,7 @@
   "publish_channel": "发布渠道",
   "publish_success": "发布成功",
   "question_guide_tip": "对话结束后，会为你生成 3 个引导性问题。",
+  "reasoning_response": "输出思考",
   "saved_success": "保存成功！如需在外部使用该版本，请点击“保存并发布”",
   "search_app": "搜索应用",
   "setting_app": "应用配置",

packages/web/i18n/zh-CN/chat.json

@@ -2,6 +2,7 @@
   "AI_input_is_empty": "传入AI 节点的内容为空",
   "Delete_all": "清空词库",
   "LLM_model_response_empty": "模型流响应为空，请检查模型流输出是否正常",
+  "ai_reasoning": "思考过程",
   "chat_history": "聊天记录",
   "chat_input_guide_lexicon_is_empty": "还没有配置词库",
   "chat_test_app": "调试-{{name}}",

packages/web/i18n/zh-Hant/account.json

@@ -48,10 +48,10 @@
   "model.output_price": "模型輸出價格",
   "model.output_price_tip": "語言模型輸出價格，如果配置了該項，則模型綜合價格會失效",
   "model.param_name": "參數名",
-  "model.request_auth": "自訂請求 Tokens",
+  "model.request_auth": "自訂請求 Key",
   "model.request_auth_tip": "向自訂請求地址發起請求時候，攜帶請求頭：Authorization: Bearer xxx 進行請求",
   "model.request_url": "自訂請求地址",
-  "model.request_url_tip": "如果填入該值，則會直接向該位址發起請求，不經過 OneAPI",
+  "model.request_url_tip": "如果填寫該值，則會直接向該地址發起請求，不經過 OneAPI。\n需要遵循 OpenAI 的 API格式，並填寫完整請求地址，例如：\n\nLLM: {{host}}/v1/chat/completions\n\nEmbedding: {{host}}/v1/embeddings\n\nSTT: {{host}}/v1/audio/transcriptions\n\nTTS: {{host}}/v1/audio/speech\n\nRerank: {{host}}/v1/rerank",
   "model.test_model": "模型測試",
   "model.tool_choice": "支援工具調用",
   "model.tool_choice_tag": "工具調用",

packages/web/i18n/zh-Hant/app.json

@@ -109,6 +109,7 @@
   "publish_channel": "發布通道",
   "publish_success": "發布成功",
   "question_guide_tip": "對話結束後，會為你產生 3 個引導性問題。",
+  "reasoning_response": "輸出思考",
   "saved_success": "保存成功！\n如需在外部使用該版本，請點擊“儲存並發布”",
   "search_app": "搜尋應用程式",
   "setting_app": "應用程式設定",

packages/web/i18n/zh-Hant/chat.json

@@ -2,6 +2,7 @@
   "AI_input_is_empty": "傳送至 AI 節點的內容為空",
   "Delete_all": "清除所有詞彙",
   "LLM_model_response_empty": "模型流程回應為空，請檢查模型流程輸出是否正常",
+  "ai_reasoning": "思考過程",
   "chat_history": "對話紀錄",
   "chat_input_guide_lexicon_is_empty": "尚未設定詞彙庫",
   "chat_test_app": "調試-{{name}}",

projects/app/src/components/core/ai/AISettingModal/index.tsx

@@ -72,6 +72,7 @@ const AIChatSettingsModal = ({
     defaultValues: defaultData
   });
   const model = watch('model');
+  const reasoning = watch(NodeInputKeyEnum.aiChatReasoning);
   const showResponseAnswerText = watch(NodeInputKeyEnum.aiChatIsResponseText) !== undefined;
   const showVisionSwitch = watch(NodeInputKeyEnum.aiChatVision) !== undefined;
   const showMaxHistoriesSlider = watch('maxHistories') !== undefined;
@@ -84,6 +85,8 @@ const AIChatSettingsModal = ({
     return getWebLLMModel(model);
   }, [model]);
   const llmSupportVision = !!selectedModel?.vision;
+  const llmSupportTemperature = typeof selectedModel?.maxTemperature === 'number';
+  const llmSupportReasoning = !!selectedModel?.reasoning;
 
   const tokenLimit = useMemo(() => {
     return selectedModel?.maxResponse || 4096;
@@ -258,36 +261,51 @@ const AIChatSettingsModal = ({
             />
           </Box>
         </Flex>
-        <Flex {...FlexItemStyles}>
+        {llmSupportTemperature && (
-          <Box {...LabelStyles}>
+          <Flex {...FlexItemStyles}>
-            <Flex alignItems={'center'}>
+            <Box {...LabelStyles}>
-              {t('app:temperature')}
+              <Flex alignItems={'center'}>
-              <QuestionTip label={t('app:temperature_tip')} />
+                {t('app:temperature')}
-            </Flex>
+                <QuestionTip label={t('app:temperature_tip')} />
-            <Switch
+              </Flex>
-              isChecked={temperature !== undefined}
+              <Switch
-              size={'sm'}
+                isChecked={temperature !== undefined}
-              onChange={(e) => {
+                size={'sm'}
-                setValue('temperature', e.target.checked ? 0 : undefined);
+                onChange={(e) => {
-              }}
+                  setValue('temperature', e.target.checked ? 0 : undefined);
-            />
+                }}
-          </Box>
+              />
-
+            </Box>
-          <Box flex={'1 0 0'}>
+            <Box flex={'1 0 0'}>
-            <InputSlider
+              <InputSlider
-              min={0}
+                min={0}
-              max={10}
+                max={10}
-              step={1}
+                step={1}
-              value={temperature}
+                value={temperature}
-              isDisabled={temperature === undefined}
+                isDisabled={temperature === undefined}
-              onChange={(e) => {
+                onChange={(e) => {
-                setValue(NodeInputKeyEnum.aiChatTemperature, e);
+                  setValue(NodeInputKeyEnum.aiChatTemperature, e);
-                setRefresh(!refresh);
+                  setRefresh(!refresh);
-              }}
+                }}
-            />
+              />
-          </Box>
+            </Box>
-        </Flex>
+          </Flex>
-
+        )}
+        {llmSupportReasoning && (
+          <Flex {...FlexItemStyles} h={'25px'}>
+            <Box {...LabelStyles}>
+              <Flex alignItems={'center'}>{t('app:reasoning_response')}</Flex>
+              <Switch
+                isChecked={reasoning || false}
+                size={'sm'}
+                onChange={(e) => {
+                  const value = e.target.checked;
+                  setValue(NodeInputKeyEnum.aiChatReasoning, value);
+                }}
+              />
+            </Box>
+          </Flex>
+        )}
         {showResponseAnswerText && (
           <Flex {...FlexItemStyles} h={'25px'}>
             <Box {...LabelStyles}>

projects/app/src/components/core/chat/ChatContainer/ChatBox/index.tsx

@@ -201,6 +201,7 @@ const ChatBox = ({
     ({
       event,
       text = '',
+      reasoningText,
       status,
       name,
       tool,
@@ -247,6 +248,25 @@ const ChatBox = ({
                 value: item.value.slice(0, -1).concat(lastValue)
               };
             }
+          } else if (event === SseResponseEventEnum.answer && reasoningText) {
+            if (lastValue.type === ChatItemValueTypeEnum.reasoning && lastValue.reasoning) {
+              lastValue.reasoning.content += reasoningText;
+              return {
+                ...item,
+                value: item.value.slice(0, -1).concat(lastValue)
+              };
+            } else {
+              const val: AIChatItemValueItemType = {
+                type: ChatItemValueTypeEnum.reasoning,
+                reasoning: {
+                  content: reasoningText
+                }
+              };
+              return {
+                ...item,
+                value: item.value.concat(val)
+              };
+            }
           } else if (event === SseResponseEventEnum.toolCall && tool) {
             const val: AIChatItemValueItemType = {
               type: ChatItemValueTypeEnum.tool,

projects/app/src/components/core/chat/ChatContainer/type.d.ts

@@ -6,6 +6,7 @@ import { WorkflowInteractiveResponseType } from '@fastgpt/global/core/workflow/t
 export type generatingMessageProps = {
   event: SseResponseEventEnum;
   text?: string;
+  reasoningText?: string;
   name?: string;
   status?: 'running' | 'finish';
   tool?: ToolModuleResponseItemType;

projects/app/src/components/core/chat/components/AIResponseBox.tsx

@@ -8,6 +8,7 @@ import {
   Box,
   Button,
   Flex,
+  HStack,
   Textarea
 } from '@chakra-ui/react';
 import { ChatItemValueTypeEnum } from '@fastgpt/global/core/chat/constants';
@@ -139,6 +140,55 @@ ${toolResponse}`}
   },
   (prevProps, nextProps) => isEqual(prevProps, nextProps)
 );
+const RenderResoningContent = React.memo(function RenderResoningContent({
+  content,
+  showAnimation
+}: {
+  content: string;
+  showAnimation: boolean;
+}) {
+  const { t } = useTranslation();
+
+  return (
+    <Accordion allowToggle defaultIndex={0}>
+      <AccordionItem borderTop={'none'} borderBottom={'none'}>
+        <AccordionButton
+          w={'auto'}
+          bg={'white'}
+          borderRadius={'md'}
+          borderWidth={'1px'}
+          borderColor={'myGray.200'}
+          boxShadow={'1'}
+          pl={3}
+          pr={2.5}
+          py={1}
+          _hover={{
+            bg: 'auto'
+          }}
+        >
+          <HStack mr={2} spacing={1}>
+            <MyIcon name={'core/chat/think'} w={'0.85rem'} />
+            <Box fontSize={'sm'}>{t('chat:ai_reasoning')}</Box>
+          </HStack>
+
+          {showAnimation && <MyIcon name={'common/loading'} w={'0.85rem'} />}
+          <AccordionIcon color={'myGray.600'} ml={5} />
+        </AccordionButton>
+        <AccordionPanel
+          py={0}
+          pr={0}
+          pl={3}
+          mt={2}
+          borderLeft={'2px solid'}
+          borderColor={'myGray.300'}
+          color={'myGray.500'}
+        >
+          <Markdown source={content} />
+        </AccordionPanel>
+      </AccordionItem>
+    </Accordion>
+  );
+});
 const RenderUserSelectInteractive = React.memo(function RenderInteractive({
   interactive
 }: {
@@ -290,6 +340,8 @@ const AIResponseBox = ({ value, isLastResponseValue, isChatting }: props) => {
     return (
       <RenderText showAnimation={isChatting && isLastResponseValue} text={value.text.content} />
     );
+  if (value.type === ChatItemValueTypeEnum.reasoning && value.reasoning)
+    return <RenderResoningContent showAnimation={isChatting} content={value.reasoning.content} />;
   if (value.type === ChatItemValueTypeEnum.tool && value.tools)
     return <RenderTool showAnimation={isChatting} tools={value.tools} />;
   if (value.type === ChatItemValueTypeEnum.interactive && value.interactive) {

projects/app/src/pageComponents/account/model/ModelConfigTable.tsx

@@ -803,6 +803,10 @@ const ModelEditModal = ({
                           <JsonEditor
                             value={JSON.stringify(getValues('defaultConfig'), null, 2)}
                             onChange={(e) => {
+                              if (!e) {
+                                setValue('defaultConfig', undefined);
+                                return;
+                              }
                               try {
                                 setValue('defaultConfig', JSON.parse(e));
                               } catch (error) {
@@ -1009,6 +1013,10 @@ const ModelEditModal = ({
                         value={JSON.stringify(getValues('defaultConfig'), null, 2)}
                         resize
                         onChange={(e) => {
+                          if (!e) {
+                            setValue('defaultConfig', undefined);
+                            return;
+                          }
                           try {
                             setValue('defaultConfig', JSON.parse(e));
                           } catch (error) {

projects/app/src/pageComponents/account/team/GroupManage/GroupManageMember.tsx

@@ -14,7 +14,7 @@ import Avatar from '@fastgpt/web/components/common/Avatar';
 import Tag from '@fastgpt/web/components/common/Tag';
 
 import { useTranslation } from 'next-i18next';
-import React, { useMemo, useState } from 'react';
+import React, { useMemo, useRef, useState } from 'react';
 import { useRequest2 } from '@fastgpt/web/hooks/useRequest';
 import { useContextSelector } from 'use-context-selector';
 import { TeamContext } from '../context';
@@ -50,6 +50,8 @@ function GroupEditModal({ onClose, editGroupId }: { onClose: () => void; editGro
   const refetchMembers = useContextSelector(TeamContext, (v) => v.refetchMembers);
   const MemberScrollData = useContextSelector(TeamContext, (v) => v.MemberScrollData);
   const [hoveredMemberId, setHoveredMemberId] = useState<string>();
+
+  const selectedMembersRef = useRef<HTMLDivElement>(null);
   const [members, setMembers] = useState(group?.members || []);
 
   const [searchKey, setSearchKey] = useState('');
@@ -155,7 +157,7 @@ function GroupEditModal({ onClose, editGroupId }: { onClose: () => void; editGro
                 setSearchKey(e.target.value);
               }}
             />
-            <MemberScrollData mt={3} flex={'1 0 0'} h={0}>
+            <MemberScrollData mt={3} flexGrow="1" overflow={'auto'} maxH={'400px'}>
               {filtered.map((member) => {
                 return (
                   <HStack
@@ -185,7 +187,7 @@ function GroupEditModal({ onClose, editGroupId }: { onClose: () => void; editGro
           </Flex>
           <Flex borderLeft="1px" borderColor="myGray.200" flexDirection="column" p="4" h={'100%'}>
             <Box mt={2}>{t('common:chosen') + ': ' + members.length}</Box>
-            <MemberScrollData mt={3} flex={'1 0 0'} h={0}>
+            <MemberScrollData ScrollContainerRef={selectedMembersRef} mt={3} flex={'1 0 0'} h={0}>
               {members.map((member) => {
                 return (
                   <HStack

projects/app/src/pageComponents/account/team/MemberTable.tsx

@@ -169,8 +169,8 @@ function MemberTable({ Tabs }: { Tabs: React.ReactNode }) {
       </Flex>
 
       <Box flex={'1 0 0'} overflow={'auto'}>
-        <TableContainer overflow={'unset'} fontSize={'sm'}>
+        <MemberScrollData>
-          <MemberScrollData>
+          <TableContainer overflow={'unset'} fontSize={'sm'}>
             <Table overflow={'unset'}>
               <Thead>
                 <Tr bgColor={'white !important'}>
@@ -246,9 +246,9 @@ function MemberTable({ Tabs }: { Tabs: React.ReactNode }) {
                 ))}
               </Tbody>
             </Table>
-          </MemberScrollData>
+            <ConfirmRemoveMemberModal />
-          <ConfirmRemoveMemberModal />
+          </TableContainer>
-        </TableContainer>
+        </MemberScrollData>
       </Box>
 
       <ConfirmLeaveTeamModal />

projects/app/src/pageComponents/account/team/OrgManage/OrgMemberManageModal.tsx

@@ -121,36 +121,34 @@ function OrgMemberManageModal({
                 setSearchKey(e.target.value);
               }}
             />
-            <Flex flexDirection="column" mt={3} flexGrow="1" overflow={'auto'} maxH={'400px'}>
+            <MemberScrollData mt={3} flexGrow="1" overflow={'auto'} maxH={'400px'}>
-              <MemberScrollData>
+              {filterMembers.map((member) => {
-                {filterMembers.map((member) => {
+                return (
-                  return (
+                  <HStack
-                    <HStack
+                    py="2"
-                      py="2"
+                    px={3}
-                      px={3}
+                    borderRadius={'md'}
-                      borderRadius={'md'}
+                    alignItems="center"
-                      alignItems="center"
+                    key={member.tmbId}
-                      key={member.tmbId}
+                    cursor={'pointer'}
-                      cursor={'pointer'}
+                    _hover={{
-                      _hover={{
+                      bg: 'myGray.50',
-                        bg: 'myGray.50',
+                      ...(!isSelected(member.tmbId) ? { svg: { color: 'myGray.50' } } : {})
-                        ...(!isSelected(member.tmbId) ? { svg: { color: 'myGray.50' } } : {})
+                    }}
-                      }}
+                    _notLast={{ mb: 2 }}
-                      _notLast={{ mb: 2 }}
+                    onClick={() => handleToggleSelect(member.tmbId)}
-                      onClick={() => handleToggleSelect(member.tmbId)}
+                  >
-                    >
+                    <Checkbox
-                      <Checkbox
+                      isChecked={!!isSelected(member.tmbId)}
-                        isChecked={!!isSelected(member.tmbId)}
+                      icon={<CheckboxIcon name={'common/check'} />}
-                        icon={<CheckboxIcon name={'common/check'} />}
+                      pointerEvents="none"
-                        pointerEvents="none"
+                    />
-                      />
+                    <Avatar src={member.avatar} w="1.5rem" borderRadius={'50%'} />
-                      <Avatar src={member.avatar} w="1.5rem" borderRadius={'50%'} />
+                    <Box>{member.memberName}</Box>
-                      <Box>{member.memberName}</Box>
+                  </HStack>
-                    </HStack>
+                );
-                  );
+              })}
-                })}
+            </MemberScrollData>
-              </MemberScrollData>
-            </Flex>
           </Flex>
           <Flex borderLeft="1px" borderColor="myGray.200" flexDirection="column" p="4" h={'100%'}>
             <Box mt={2}>{`${t('common:chosen')}:${selectedMembers.length}`}</Box>

projects/app/src/pageComponents/app/detail/WorkflowComponents/Flow/nodes/render/RenderInput/templates/SettingLLMModel.tsx

@@ -38,7 +38,9 @@ const SelectAiModelRender = ({ item, inputs = [], nodeId }: RenderInputProps) =>
         (input) => input.key === NodeInputKeyEnum.aiChatIsResponseText
       )?.value,
       aiChatVision:
-        inputs.find((input) => input.key === NodeInputKeyEnum.aiChatVision)?.value ?? true
+        inputs.find((input) => input.key === NodeInputKeyEnum.aiChatVision)?.value ?? true,
+      aiChatReasoning:
+        inputs.find((input) => input.key === NodeInputKeyEnum.aiChatReasoning)?.value ?? true
     }),
     [inputs]
   );

projects/app/src/web/common/api/fetch.ts

@@ -186,6 +186,12 @@ export const streamFetch = ({
                 text: item
               });
             }
+
+            const reasoningText = parseJson.choices?.[0]?.delta?.reasoning_content || '';
+            onMessage({
+              event,
+              reasoningText
+            });
           } else if (event === SseResponseEventEnum.fastAnswer) {
             const text = parseJson.choices?.[0]?.delta?.content || '';
             pushDataToQueue({

projects/app/src/web/core/app/templates.ts

@@ -1,7 +1,7 @@
 import { parseCurl } from '@fastgpt/global/common/string/http';
 import { AppTypeEnum } from '@fastgpt/global/core/app/constants';
 import { AppSchema } from '@fastgpt/global/core/app/type';
-import { WorkflowIOValueTypeEnum } from '@fastgpt/global/core/workflow/constants';
+import { NodeInputKeyEnum, WorkflowIOValueTypeEnum } from '@fastgpt/global/core/workflow/constants';
 import {
   FlowNodeInputTypeEnum,
   FlowNodeOutputTypeEnum,
@@ -150,7 +150,7 @@ export const emptyTemplates: Record<
             key: 'temperature',
             renderTypeList: [FlowNodeInputTypeEnum.hidden],
             label: '',
-            value: 0,
+            value: undefined,
             valueType: WorkflowIOValueTypeEnum.number,
             min: 0,
             max: 10,
@@ -160,7 +160,7 @@ export const emptyTemplates: Record<
             key: 'maxToken',
             renderTypeList: [FlowNodeInputTypeEnum.hidden],
             label: '',
-            value: 2000,
+            value: undefined,
             valueType: WorkflowIOValueTypeEnum.number,
             min: 100,
             max: 4000,
@@ -221,6 +221,13 @@ export const emptyTemplates: Record<
             debugLabel: i18nT('common:core.module.Dataset quote.label'),
             description: '',
             valueType: WorkflowIOValueTypeEnum.datasetQuote
+          },
+          {
+            key: NodeInputKeyEnum.aiChatReasoning,
+            renderTypeList: [FlowNodeInputTypeEnum.hidden],
+            label: '',
+            valueType: WorkflowIOValueTypeEnum.boolean,
+            value: true
           }
         ],
         outputs: [