Test parse cite and add tool call parallel (#4737)

* add quote response filter (#4727) * chatting * add quote response filter * add test * remove comment * perf: cite hidden * perf: format llm response * feat: comment * update default chunk size * update default chunk size --------- Co-authored-by: heheer <heheer@sealos.io>
2025-10-15 23:55:36 +00:00 · 2025-04-30 17:43:50 +08:00
parent 683ab6c17d
commit fdd4e9edbd
53 changed files with 1131 additions and 716 deletions
--- a/docSite/content/zh-cn/docs/development/upgrading/498.md
+++ b/docSite/content/zh-cn/docs/development/upgrading/498.md
@@ -10,7 +10,9 @@ weight: 792

 ## 🚀 新增内容

-1. qwen3 模型预设
+1. 支持 Toolcalls 并行执行。
+2. 将所有内置任务，从非 stream 模式调整成 stream 模式，避免部分模型不支持非 stream 模式。如需覆盖，则可以在模型`额外 Body`参数中，强制指定`stream=false`。
+3. qwen3 模型预设

 ## ⚙️ 优化

--- a/docSite/content/zh-cn/docs/use-cases/app-cases/lab_appointment.md
+++ b/docSite/content/zh-cn/docs/use-cases/app-cases/lab_appointment.md
@@ -563,7 +563,7 @@ HTTP模块中，需要设置 3 个工具参数：
            "hidden"
          ],
          "label": "",
-          "value": 1500,
+          "value": 5000,
          "valueType": "number"
        },
        {
--- a/packages/global/common/string/password.ts
+++ b/packages/global/common/string/password.ts
@@ -0,0 +1,18 @@
+export const checkPasswordRule = (password: string) => {
+  const patterns = [
+    /\d/, // Contains digits
+    /[a-z]/, // Contains lowercase letters
+    /[A-Z]/, // Contains uppercase letters
+    /[!@#$%^&*()_+=-]/ // Contains special characters
+  ];
+  const validChars = /^[\dA-Za-z!@#$%^&*()_+=-]{6,100}$/;
+
+  // Check length and valid characters
+  if (!validChars.test(password)) return false;
+
+  // Count how many patterns are satisfied
+  const matchCount = patterns.filter((pattern) => pattern.test(password)).length;
+
+  // Must satisfy at least 2 patterns
+  return matchCount >= 2;
+};
--- a/packages/global/core/ai/prompt/AIChat.ts
+++ b/packages/global/core/ai/prompt/AIChat.ts
@@ -88,8 +88,8 @@ export const Prompt_userQuotePromptList: PromptTemplateItem[] = [
 - 保持答案与 <Reference></Reference> 中描述的一致。
 - 使用 Markdown 语法优化回答格式。
 - 使用与问题相同的语言回答。
- 使用 [id](QUOTE) 格式来引用<Reference></Reference>中的知识，其中 QUOTE 是固定常量, id 为引文中的 id。
- 在每段结尾自然地整合引用。例如: "FastGPT 是一个基于大语言模型(LLM)的知识库问答系统[67e517e74767063e882d6861](QUOTE)。"
+- 使用 [id](CITE) 格式来引用<Reference></Reference>中的知识，其中 CITE 是固定常量, id 为引文中的 id。
+- 在每段结尾自然地整合引用。例如: "FastGPT 是一个基于大语言模型(LLM)的知识库问答系统[67e517e74767063e882d6861](CITE)。"
 - 每段至少包含一个引用，也可根据内容需要加入多个引用，按顺序排列。`,
      ['4.9.2']: `使用 <Reference></Reference> 标记中的内容作为本次对话的参考:

@@ -146,8 +146,8 @@ export const Prompt_userQuotePromptList: PromptTemplateItem[] = [
 - 保持答案与 <Reference></Reference> 中描述的一致。
 - 使用 Markdown 语法优化回答格式。
 - 使用与问题相同的语言回答。
- 使用 [id](QUOTE) 格式来引用<Reference></Reference>中的知识，其中 QUOTE 是固定常量, id 为引文中的 id。
- 在每段结尾自然地整合引用。例如: "FastGPT 是一个基于大语言模型(LLM)的知识库问答系统[67e517e74767063e882d6861](QUOTE)。"
+- 使用 [id](CITE) 格式来引用<Reference></Reference>中的知识，其中 CITE 是固定常量, id 为引文中的 id。
+- 在每段结尾自然地整合引用。例如: "FastGPT 是一个基于大语言模型(LLM)的知识库问答系统[67e517e74767063e882d6861](CITE)。"
 - 每段至少包含一个引用，也可根据内容需要加入多个引用，按顺序排列。

 问题:"""{{question}}"""`,
@@ -217,8 +217,8 @@ export const Prompt_systemQuotePromptList: PromptTemplateItem[] = [
 - 保持答案与 <Reference></Reference> 中描述的一致。
 - 使用 Markdown 语法优化回答格式。
 - 使用与问题相同的语言回答。
- 使用 [id](QUOTE) 格式来引用<Reference></Reference>中的知识，其中 QUOTE 是固定常量, id 为引文中的 id。
- 在每段结尾自然地整合引用。例如: "FastGPT 是一个基于大语言模型(LLM)的知识库问答系统[67e517e74767063e882d6861](QUOTE)。"
+- 使用 [id](CITE) 格式来引用<Reference></Reference>中的知识，其中 CITE 是固定常量, id 为引文中的 id。
+- 在每段结尾自然地整合引用。例如: "FastGPT 是一个基于大语言模型(LLM)的知识库问答系统[67e517e74767063e882d6861](CITE)。"
 - 每段至少包含一个引用，也可根据内容需要加入多个引用，按顺序排列。`,
      ['4.9.2']: `使用 <Reference></Reference> 标记中的内容作为本次对话的参考:

@@ -271,8 +271,8 @@ export const Prompt_systemQuotePromptList: PromptTemplateItem[] = [
 - 保持答案与 <Reference></Reference> 中描述的一致。
 - 使用 Markdown 语法优化回答格式。
 - 使用与问题相同的语言回答。
- 使用 [id](QUOTE) 格式来引用<Reference></Reference>中的知识，其中 QUOTE 是固定常量, id 为引文中的 id。
- 在每段结尾自然地整合引用。例如: "FastGPT 是一个基于大语言模型(LLM)的知识库问答系统[67e517e74767063e882d6861](QUOTE)。"
+- 使用 [id](CITE) 格式来引用<Reference></Reference>中的知识，其中 CITE 是固定常量, id 为引文中的 id。
+- 在每段结尾自然地整合引用。例如: "FastGPT 是一个基于大语言模型(LLM)的知识库问答系统[67e517e74767063e882d6861](CITE)。"
 - 每段至少包含一个引用，也可根据内容需要加入多个引用，按顺序排列。

 问题:"""{{question}}"""`,
@@ -321,24 +321,13 @@ export const Prompt_systemQuotePromptList: PromptTemplateItem[] = [
  }
 ];

-export const getQuotePrompt = (
-  version?: string,
-  role: 'user' | 'system' = 'user',
-  parseQuote = true
-) => {
+export const getQuotePrompt = (version?: string, role: 'user' | 'system' = 'user') => {
  const quotePromptTemplates =
    role === 'user' ? Prompt_userQuotePromptList : Prompt_systemQuotePromptList;

  const defaultTemplate = quotePromptTemplates[0].value;

-  return parseQuote
-    ? getPromptByVersion(version, defaultTemplate)
-    : getPromptByVersion(version, defaultTemplate).replace(
-        `- 使用 [id](QUOTE) 格式来引用<Reference></Reference>中的知识，其中 QUOTE 是固定常量, id 为引文中的 id。
- 在每段结尾自然地整合引用。例如: "FastGPT 是一个基于大语言模型(LLM)的知识库问答系统[67e517e74767063e882d6861](QUOTE)。"
- 每段至少包含一个引用，也可根据内容需要加入多个引用，按顺序排列。`,
-        ''
-      );
+  return getPromptByVersion(version, defaultTemplate);
 };

 // Document quote prompt
--- a/packages/global/core/ai/prompt/agent.ts
+++ b/packages/global/core/ai/prompt/agent.ts
@@ -60,7 +60,7 @@ export const getExtractJsonToolPrompt = (version?: string) => {
 """
 - {{description}}
 - 不是每个参数都是必须生成的，如果没有合适的参数值，不要生成该参数，或返回空字符串。
- 需要结合前面的对话内容，一起生成合适的参数。
+- 需要结合历史记录，一起生成合适的参数。
 """

 本次输入内容: """{{content}}"""
--- a/packages/global/core/ai/prompt/dataset.ts
+++ b/packages/global/core/ai/prompt/dataset.ts
@@ -1,6 +1,5 @@
-export const getDatasetSearchToolResponsePrompt = (parseQuote: boolean) => {
-  return parseQuote
-    ? `## Role
+export const getDatasetSearchToolResponsePrompt = () => {
+  return `## Role
 你是一个知识库回答助手，可以 "quotes" 中的内容作为本次对话的参考。为了使回答结果更加可信并且可追溯，你需要在每段话结尾添加引用标记。

 ## Rules
@@ -9,16 +8,7 @@ export const getDatasetSearchToolResponsePrompt = (parseQuote: boolean) => {
 - 保持答案与 "quotes" 中描述的一致。
 - 使用 Markdown 语法优化回答格式。尤其是图片、表格、序列号等内容，需严格完整输出。
 - 使用与问题相同的语言回答。
- 使用 [id](QUOTE) 格式来引用 "quotes" 中的知识，其中 QUOTE 是固定常量, id 为引文中的 id。
- 在每段话结尾自然地整合引用。例如: "FastGPT 是一个基于大语言模型(LLM)的知识库问答系统[67e517e74767063e882d6861](QUOTE)。"
- 每段话至少包含一个引用，也可根据内容需要加入多个引用，按顺序排列。`
-    : `## Role
-你是一个知识库回答助手，可以 "quotes" 中的内容作为本次对话的参考。
-
-## Rules
- 如果你不清楚答案，你需要澄清。
- 避免提及你是从 "quotes" 获取的知识。
- 保持答案与 "quotes" 中描述的一致。
- 使用 Markdown 语法优化回答格式。尤其是图片、表格、序列号等内容，需严格完整输出。
- 使用与问题相同的语言回答。`;
+- 使用 [id](CITE) 格式来引用 "quotes" 中的知识，其中 CITE 是固定常量, id 为引文中的 id。
+- 在每段话结尾自然地整合引用。例如: "FastGPT 是一个基于大语言模型(LLM)的知识库问答系统[67e517e74767063e882d6861](CITE)。"
+- 每段话至少包含一个引用，也可根据内容需要加入多个引用，按顺序排列。`;
 };
--- a/packages/global/core/ai/type.d.ts
+++ b/packages/global/core/ai/type.d.ts
@@ -60,6 +60,7 @@ export type ChatCompletionAssistantToolParam = {
  tool_calls: ChatCompletionMessageToolCall[];
 };
 export type ChatCompletionMessageToolCall = ChatCompletionMessageToolCall & {
+  index?: number;
  toolName?: string;
  toolAvatar?: string;
 };
--- a/packages/global/core/chat/utils.ts
+++ b/packages/global/core/chat/utils.ts
@@ -1,9 +1,15 @@
 import { DispatchNodeResponseType } from '../workflow/runtime/type';
 import { FlowNodeTypeEnum } from '../workflow/node/constant';
 import { ChatItemValueTypeEnum, ChatRoleEnum, ChatSourceEnum } from './constants';
-import { ChatHistoryItemResType, ChatItemType, UserChatItemValueItemType } from './type.d';
+import {
+  AIChatItemValueItemType,
+  ChatHistoryItemResType,
+  ChatItemType,
+  UserChatItemValueItemType
+} from './type.d';
 import { sliceStrStartEnd } from '../../common/string/tools';
 import { PublishChannelEnum } from '../../support/outLink/constant';
+import { removeDatasetCiteText } from '../../../service/core/ai/utils';

 // Concat 2 -> 1, and sort by role
 export const concatHistories = (histories1: ChatItemType[], histories2: ChatItemType[]) => {
@@ -77,6 +83,7 @@ export const getHistoryPreview = (
  });
 };

+// Filter workflow public response
 export const filterPublicNodeResponseData = ({
  flowResponses = [],
  responseDetail = false
@@ -112,6 +119,40 @@ export const filterPublicNodeResponseData = ({
    });
 };

+// Remove dataset cite in ai response
+export const removeAIResponseCite = <T extends AIChatItemValueItemType[] | string>(
+  value: T,
+  retainCite: boolean
+): T => {
+  if (retainCite) return value;
+
+  if (typeof value === 'string') {
+    return removeDatasetCiteText(value, false) as T;
+  }
+
+  return value.map<AIChatItemValueItemType>((item) => {
+    if (item.text?.content) {
+      return {
+        ...item,
+        text: {
+          ...item.text,
+          content: removeDatasetCiteText(item.text.content, false)
+        }
+      };
+    }
+    if (item.reasoning?.content) {
+      return {
+        ...item,
+        reasoning: {
+          ...item.reasoning,
+          content: removeDatasetCiteText(item.reasoning.content, false)
+        }
+      };
+    }
+    return item;
+  }) as T;
+};
+
 export const removeEmptyUserInput = (input?: UserChatItemValueItemType[]) => {
  return (
    input?.filter((item) => {
--- a/packages/global/core/dataset/training/utils.ts
+++ b/packages/global/core/dataset/training/utils.ts
@@ -8,7 +8,7 @@ import {
 export const minChunkSize = 64; // min index and chunk size

 // Chunk size
-export const chunkAutoChunkSize = 1500;
+export const chunkAutoChunkSize = 1000;
 export const getMaxChunkSize = (model: LLMModelItemType) => {
  return Math.max(model.maxContext - model.maxResponse, 2000);
 };
--- a/packages/global/core/workflow/runtime/type.d.ts
+++ b/packages/global/core/workflow/runtime/type.d.ts
@@ -58,7 +58,7 @@ export type ChatDispatchProps = {
  chatConfig: AppSchema['chatConfig'];
  lastInteractive?: WorkflowInteractiveResponseType; // last interactive response
  stream: boolean;
-  parseQuote?: boolean;
+  retainDatasetCite?: boolean;
  maxRunTimes: number;
  isToolCall?: boolean;
  workflowStreamResponse?: WorkflowResponseType;
--- a/packages/global/core/workflow/template/system/datasetSearch.ts
+++ b/packages/global/core/workflow/template/system/datasetSearch.ts
@@ -54,7 +54,7 @@ export const DatasetSearchModule: FlowNodeTemplateType = {
      key: NodeInputKeyEnum.datasetMaxTokens,
      renderTypeList: [FlowNodeInputTypeEnum.hidden],
      label: '',
-      value: 1500,
+      value: 5000,
      valueType: WorkflowIOValueTypeEnum.number
    },
    {
--- a/packages/service/core/ai/config/provider/Qwen.json
+++ b/packages/service/core/ai/config/provider/Qwen.json
@@ -1,79 +1,6 @@
 {
  "provider": "Qwen",
  "list": [
-    {
-      "model": "qwen-vl-plus",
-      "name": "qwen-vl-plus",
-      "maxContext": 32000,
-      "maxResponse": 2000,
-      "quoteMaxToken": 20000,
-      "maxTemperature": 1.2,
-      "vision": true,
-      "toolChoice": false,
-      "functionCall": false,
-      "defaultSystemChatPrompt": "",
-      "datasetProcess": true,
-      "usedInClassify": true,
-      "customCQPrompt": "",
-      "usedInExtractFields": true,
-      "usedInQueryExtension": true,
-      "customExtractPrompt": "",
-      "usedInToolCall": true,
-      "type": "llm",
-      "showTopP": true,
-      "showStopSign": true
-    },
-    {
-      "model": "qwen-plus",
-      "name": "Qwen-plus",
-      "maxContext": 64000,
-      "maxResponse": 8000,
-      "quoteMaxToken": 60000,
-      "maxTemperature": 1,
-      "vision": false,
-      "toolChoice": true,
-      "functionCall": false,
-      "defaultSystemChatPrompt": "",
-      "datasetProcess": true,
-      "usedInClassify": true,
-      "customCQPrompt": "",
-      "usedInExtractFields": true,
-      "usedInQueryExtension": true,
-      "customExtractPrompt": "",
-      "usedInToolCall": true,
-      "defaultConfig": {},
-      "fieldMap": {},
-      "type": "llm",
-      "showTopP": true,
-      "showStopSign": true,
-      "responseFormatList": ["text", "json_object"]
-    },
-    {
-      "model": "qwen-turbo",
-      "name": "Qwen-turbo",
-      "maxContext": 128000,
-      "maxResponse": 8000,
-      "quoteMaxToken": 100000,
-      "maxTemperature": 1,
-      "vision": false,
-      "toolChoice": true,
-      "functionCall": false,
-      "defaultSystemChatPrompt": "",
-      "datasetProcess": true,
-      "usedInClassify": true,
-      "customCQPrompt": "",
-      "usedInExtractFields": true,
-      "usedInQueryExtension": true,
-      "customExtractPrompt": "",
-      "usedInToolCall": true,
-      "defaultConfig": {},
-      "fieldMap": {},
-      "type": "llm",
-      "showTopP": true,
-      "showStopSign": true,
-      "responseFormatList": ["text", "json_object"]
-    },
-
    {
      "model": "qwen-max",
      "name": "Qwen-max",
@@ -123,6 +50,78 @@
      "showTopP": true,
      "showStopSign": true
    },
+    {
+      "model": "qwen-plus",
+      "name": "Qwen-plus",
+      "maxContext": 64000,
+      "maxResponse": 8000,
+      "quoteMaxToken": 60000,
+      "maxTemperature": 1,
+      "vision": false,
+      "toolChoice": true,
+      "functionCall": false,
+      "defaultSystemChatPrompt": "",
+      "datasetProcess": true,
+      "usedInClassify": true,
+      "customCQPrompt": "",
+      "usedInExtractFields": true,
+      "usedInQueryExtension": true,
+      "customExtractPrompt": "",
+      "usedInToolCall": true,
+      "defaultConfig": {},
+      "fieldMap": {},
+      "type": "llm",
+      "showTopP": true,
+      "showStopSign": true,
+      "responseFormatList": ["text", "json_object"]
+    },
+    {
+      "model": "qwen-vl-plus",
+      "name": "qwen-vl-plus",
+      "maxContext": 32000,
+      "maxResponse": 2000,
+      "quoteMaxToken": 20000,
+      "maxTemperature": 1.2,
+      "vision": true,
+      "toolChoice": false,
+      "functionCall": false,
+      "defaultSystemChatPrompt": "",
+      "datasetProcess": true,
+      "usedInClassify": true,
+      "customCQPrompt": "",
+      "usedInExtractFields": true,
+      "usedInQueryExtension": true,
+      "customExtractPrompt": "",
+      "usedInToolCall": true,
+      "type": "llm",
+      "showTopP": true,
+      "showStopSign": true
+    },
+    {
+      "model": "qwen-turbo",
+      "name": "Qwen-turbo",
+      "maxContext": 128000,
+      "maxResponse": 8000,
+      "quoteMaxToken": 100000,
+      "maxTemperature": 1,
+      "vision": false,
+      "toolChoice": true,
+      "functionCall": false,
+      "defaultSystemChatPrompt": "",
+      "datasetProcess": true,
+      "usedInClassify": true,
+      "customCQPrompt": "",
+      "usedInExtractFields": true,
+      "usedInQueryExtension": true,
+      "customExtractPrompt": "",
+      "usedInToolCall": true,
+      "defaultConfig": {},
+      "fieldMap": {},
+      "type": "llm",
+      "showTopP": true,
+      "showStopSign": true,
+      "responseFormatList": ["text", "json_object"]
+    },
    {
      "model": "qwen3-235b-a22b",
      "name": "qwen3-235b-a22b",
@@ -142,7 +141,9 @@
      "usedInQueryExtension": true,
      "customExtractPrompt": "",
      "usedInToolCall": true,
-      "defaultConfig": {},
+      "defaultConfig": {
+        "stream": true
+      },
      "fieldMap": {},
      "type": "llm",
      "showTopP": true,
@@ -168,7 +169,9 @@
      "usedInQueryExtension": true,
      "customExtractPrompt": "",
      "usedInToolCall": true,
-      "defaultConfig": {},
+      "defaultConfig": {
+        "stream": true
+      },
      "fieldMap": {},
      "type": "llm",
      "showTopP": true,
@@ -194,7 +197,9 @@
      "usedInQueryExtension": true,
      "customExtractPrompt": "",
      "usedInToolCall": true,
-      "defaultConfig": {},
+      "defaultConfig": {
+        "stream": true
+      },
      "fieldMap": {},
      "type": "llm",
      "showTopP": true,
@@ -220,7 +225,9 @@
      "usedInQueryExtension": true,
      "customExtractPrompt": "",
      "usedInToolCall": true,
-      "defaultConfig": {},
+      "defaultConfig": {
+        "stream": true
+      },
      "fieldMap": {},
      "type": "llm",
      "showTopP": true,
@@ -246,7 +253,9 @@
      "usedInQueryExtension": true,
      "customExtractPrompt": "",
      "usedInToolCall": true,
-      "defaultConfig": {},
+      "defaultConfig": {
+        "stream": true
+      },
      "fieldMap": {},
      "type": "llm",
      "showTopP": true,
@@ -272,7 +281,9 @@
      "usedInQueryExtension": true,
      "customExtractPrompt": "",
      "usedInToolCall": true,
-      "defaultConfig": {},
+      "defaultConfig": {
+        "stream": true
+      },
      "fieldMap": {},
      "type": "llm",
      "showTopP": true,
@@ -298,7 +309,9 @@
      "usedInQueryExtension": true,
      "customExtractPrompt": "",
      "usedInToolCall": true,
-      "defaultConfig": {},
+      "defaultConfig": {
+        "stream": true
+      },
      "fieldMap": {},
      "type": "llm",
      "showTopP": true,
@@ -324,7 +337,9 @@
      "usedInQueryExtension": true,
      "customExtractPrompt": "",
      "usedInToolCall": true,
-      "defaultConfig": {},
+      "defaultConfig": {
+        "stream": true
+      },
      "fieldMap": {},
      "type": "llm",
      "showTopP": true,
@@ -350,7 +365,9 @@
      "usedInQueryExtension": false,
      "customExtractPrompt": "",
      "usedInToolCall": true,
-      "defaultConfig": {},
+      "defaultConfig": {
+        "stream": true
+      },
      "fieldMap": {},
      "type": "llm",
      "showTopP": false,
@@ -375,7 +392,9 @@
      "usedInQueryExtension": false,
      "customExtractPrompt": "",
      "usedInToolCall": true,
-      "defaultConfig": {},
+      "defaultConfig": {
+        "stream": true
+      },
      "fieldMap": {},
      "type": "llm",
      "showTopP": false,
--- a/packages/service/core/ai/functions/createQuestionGuide.ts
+++ b/packages/service/core/ai/functions/createQuestionGuide.ts
@@ -2,7 +2,7 @@ import type { ChatCompletionMessageParam } from '@fastgpt/global/core/ai/type.d'
 import { createChatCompletion } from '../config';
 import { countGptMessagesTokens, countPromptTokens } from '../../../common/string/tiktoken/index';
 import { loadRequestMessages } from '../../chat/utils';
-import { llmCompletionsBodyFormat, llmResponseToAnswerText } from '../utils';
+import { llmCompletionsBodyFormat, formatLLMResponse } from '../utils';
 import {
  QuestionGuidePrompt,
  QuestionGuideFooterPrompt
@@ -42,12 +42,12 @@ export async function createQuestionGuide({
        temperature: 0.1,
        max_tokens: 200,
        messages: requestMessages,
-        stream: false
+        stream: true
      },
      model
    )
  });
-  const { text: answer, usage } = await llmResponseToAnswerText(response);
+  const { text: answer, usage } = await formatLLMResponse(response);

  const start = answer.indexOf('[');
  const end = answer.lastIndexOf(']');
--- a/packages/service/core/ai/functions/queryExtension.ts
+++ b/packages/service/core/ai/functions/queryExtension.ts
@@ -4,7 +4,7 @@ import { ChatItemType } from '@fastgpt/global/core/chat/type';
 import { countGptMessagesTokens, countPromptTokens } from '../../../common/string/tiktoken/index';
 import { chats2GPTMessages } from '@fastgpt/global/core/chat/adapt';
 import { getLLMModel } from '../model';
-import { llmCompletionsBodyFormat, llmResponseToAnswerText } from '../utils';
+import { llmCompletionsBodyFormat, formatLLMResponse } from '../utils';
 import { addLog } from '../../../common/system/log';
 import { filterGPTMessageByMaxContext } from '../../chat/utils';
 import json5 from 'json5';
@@ -170,7 +170,7 @@ assistant: ${chatBg}
  const { response } = await createChatCompletion({
    body: llmCompletionsBodyFormat(
      {
-        stream: false,
+        stream: true,
        model: modelData.model,
        temperature: 0.1,
        messages
@@ -178,7 +178,7 @@ assistant: ${chatBg}
      modelData
    )
  });
-  const { text: answer, usage } = await llmResponseToAnswerText(response);
+  const { text: answer, usage } = await formatLLMResponse(response);
  const inputTokens = usage?.prompt_tokens || (await countGptMessagesTokens(messages));
  const outputTokens = usage?.completion_tokens || (await countPromptTokens(answer));

--- a/packages/service/core/ai/utils.test.ts
+++ b/packages/service/core/ai/utils.test.ts
@@ -1,145 +0,0 @@
-import { parseReasoningStreamContent } from './utils';
-import { expect, test } from 'vitest';
-
-test('Parse reasoning stream content test', async () => {
-  const partList = [
-    {
-      data: [{ content: '你好1' }, { content: '你好2' }, { content: '你好3' }],
-      correct: { answer: '你好1你好2你好3', reasoning: '' }
-    },
-    {
-      data: [
-        { reasoning_content: '这是' },
-        { reasoning_content: '思考' },
-        { reasoning_content: '过程' },
-        { content: '你好1' },
-        { content: '你好2' },
-        { content: '你好3' }
-      ],
-      correct: { answer: '你好1你好2你好3', reasoning: '这是思考过程' }
-    },
-    {
-      data: [
-        { content: '<t' },
-        { content: 'hink>' },
-        { content: '这是' },
-        { content: '思考' },
-        { content: '过程' },
-        { content: '</think>' },
-        { content: '你好1' },
-        { content: '你好2' },
-        { content: '你好3' }
-      ],
-      correct: { answer: '你好1你好2你好3', reasoning: '这是思考过程' }
-    },
-    {
-      data: [
-        { content: '<think>' },
-        { content: '这是' },
-        { content: '思考' },
-        { content: '过程' },
-        { content: '</think>' },
-        { content: '你好1' },
-        { content: '你好2' },
-        { content: '你好3' }
-      ],
-      correct: { answer: '你好1你好2你好3', reasoning: '这是思考过程' }
-    },
-    {
-      data: [
-        { content: '<think>这是' },
-        { content: '思考' },
-        { content: '过程' },
-        { content: '</think>' },
-        { content: '你好1' },
-        { content: '你好2' },
-        { content: '你好3' }
-      ],
-      correct: { answer: '你好1你好2你好3', reasoning: '这是思考过程' }
-    },
-    {
-      data: [
-        { content: '<think>这是' },
-        { content: '思考' },
-        { content: '过程</' },
-        { content: 'think>' },
-        { content: '你好1' },
-        { content: '你好2' },
-        { content: '你好3' }
-      ],
-      correct: { answer: '你好1你好2你好3', reasoning: '这是思考过程' }
-    },
-    {
-      data: [
-        { content: '<think>这是' },
-        { content: '思考' },
-        { content: '过程</think>' },
-        { content: '你好1' },
-        { content: '你好2' },
-        { content: '你好3' }
-      ],
-      correct: { answer: '你好1你好2你好3', reasoning: '这是思考过程' }
-    },
-    {
-      data: [
-        { content: '<think>这是' },
-        { content: '思考' },
-        { content: '过程</think>你好1' },
-        { content: '你好2' },
-        { content: '你好3' }
-      ],
-      correct: { answer: '你好1你好2你好3', reasoning: '这是思考过程' }
-    },
-    {
-      data: [
-        { content: '<think>这是' },
-        { content: '思考' },
-        { content: '过程</th' },
-        { content: '假的' },
-        { content: '你好2' },
-        { content: '你好3' },
-        { content: '过程</think>你好1' },
-        { content: '你好2' },
-        { content: '你好3' }
-      ],
-      correct: { answer: '你好1你好2你好3', reasoning: '这是思考过程</th假的你好2你好3过程' }
-    },
-    {
-      data: [
-        { content: '<think>这是' },
-        { content: '思考' },
-        { content: '过程</th' },
-        { content: '假的' },
-        { content: '你好2' },
-        { content: '你好3' }
-      ],
-      correct: { answer: '', reasoning: '这是思考过程</th假的你好2你好3' }
-    }
-  ];
-
-  partList.forEach((part) => {
-    const { parsePart } = parseReasoningStreamContent();
-
-    let answer = '';
-    let reasoning = '';
-    part.data.forEach((item) => {
-      const formatPart = {
-        choices: [
-          {
-            delta: {
-              role: 'assistant',
-              content: item.content,
-              reasoning_content: item.reasoning_content
-            }
-          }
-        ]
-      };
-      const [reasoningContent, content] = parsePart(formatPart, true);
-      answer += content;
-      reasoning += reasoningContent;
-    });
-
-    expect(answer).toBe(part.correct.answer);
-    expect(reasoning).toBe(part.correct.reasoning);
-  });
-});
--- a/packages/service/core/ai/utils.ts
+++ b/packages/service/core/ai/utils.ts
@@ -5,10 +5,12 @@ import {
  CompletionFinishReason,
  StreamChatType,
  UnStreamChatType,
-  CompletionUsage
+  CompletionUsage,
+  ChatCompletionMessageToolCall
 } from '@fastgpt/global/core/ai/type';
 import { getLLMModel } from './model';
 import { getLLMDefaultUsage } from '@fastgpt/global/core/ai/constants';
+import { getNanoid } from '@fastgpt/global/common/string/tools';

 /* 
  Count response max token
@@ -105,33 +107,84 @@ export const llmStreamResponseToAnswerText = async (
 ): Promise<{
  text: string;
  usage?: CompletionUsage;
+  toolCalls?: ChatCompletionMessageToolCall[];
 }> => {
  let answer = '';
  let usage = getLLMDefaultUsage();
+  let toolCalls: ChatCompletionMessageToolCall[] = [];
+  let callingTool: { name: string; arguments: string } | null = null;
+
  for await (const part of response) {
    usage = part.usage || usage;
+    const responseChoice = part.choices?.[0]?.delta;

-    const content = part.choices?.[0]?.delta?.content || '';
+    const content = responseChoice?.content || '';
    answer += content;
+
+    // Tool calls
+    if (responseChoice?.tool_calls?.length) {
+      responseChoice.tool_calls.forEach((toolCall) => {
+        const index = toolCall.index;
+
+        if (toolCall.id || callingTool) {
+          // 有 id，代表新 call 工具
+          if (toolCall.id) {
+            callingTool = {
+              name: toolCall.function?.name || '',
+              arguments: toolCall.function?.arguments || ''
+            };
+          } else if (callingTool) {
+            // Continue call(Perhaps the name of the previous function was incomplete)
+            callingTool.name += toolCall.function?.name || '';
+            callingTool.arguments += toolCall.function?.arguments || '';
+          }
+
+          if (!callingTool) {
+            return;
+          }
+
+          // New tool, add to list.
+          const toolId = getNanoid();
+          toolCalls[index] = {
+            ...toolCall,
+            id: toolId,
+            type: 'function',
+            function: callingTool
+          };
+          callingTool = null;
+        } else {
+          /* arg 追加到当前工具的参数里 */
+          const arg: string = toolCall?.function?.arguments ?? '';
+          const currentTool = toolCalls[index];
+          if (currentTool && arg) {
+            currentTool.function.arguments += arg;
+          }
+        }
+      });
+    }
  }
  return {
    text: parseReasoningContent(answer)[1],
-    usage
+    usage,
+    toolCalls
  };
 };
 export const llmUnStreamResponseToAnswerText = async (
  response: UnStreamChatType
 ): Promise<{
  text: string;
+  toolCalls?: ChatCompletionMessageToolCall[];
  usage?: CompletionUsage;
 }> => {
  const answer = response.choices?.[0]?.message?.content || '';
+  const toolCalls = response.choices?.[0]?.message?.tool_calls;
  return {
    text: answer,
-    usage: response.usage
+    usage: response.usage,
+    toolCalls
  };
 };
-export const llmResponseToAnswerText = async (response: StreamChatType | UnStreamChatType) => {
+export const formatLLMResponse = async (response: StreamChatType | UnStreamChatType) => {
  if ('iterator' in response) {
    return llmStreamResponseToAnswerText(response);
  }
@@ -155,20 +208,31 @@ export const parseReasoningContent = (text: string): [string, string] => {
  return [thinkContent, answerContent];
 };

-// Parse <think></think> tags to think and answer - stream response
-export const parseReasoningStreamContent = () => {
-  let isInThinkTag: boolean | undefined;
+export const removeDatasetCiteText = (text: string, retainDatasetCite: boolean) => {
+  return retainDatasetCite ? text : text.replace(/\[([a-f0-9]{24})\]\(CITE\)/g, '');
+};

-  const startTag = '<think>';
+// Parse llm stream part
+export const parseLLMStreamResponse = () => {
+  let isInThinkTag: boolean | undefined = undefined;
  let startTagBuffer = '';
-
-  const endTag = '</think>';
  let endTagBuffer = '';

+  const thinkStartChars = '<think>';
+  const thinkEndChars = '</think>';
+
+  let citeBuffer = '';
+  const maxCiteBufferLength = 32; // [Object](CITE)总长度为32
+
  /* 
    parseThinkTag - 只控制是否主动解析 <think></think>，如果接口已经解析了，则不再解析。
+    retainDatasetCite - 
  */
-  const parsePart = (
+  const parsePart = ({
+    part,
+    parseThinkTag = true,
+    retainDatasetCite = true
+  }: {
    part: {
      choices: {
        delta: {
@@ -177,28 +241,32 @@ export const parseReasoningStreamContent = () => {
        };
        finish_reason?: CompletionFinishReason;
      }[];
-    },
-    parseThinkTag = false
-  ): {
+    };
+    parseThinkTag?: boolean;
+    retainDatasetCite?: boolean;
+  }): {
    reasoningContent: string;
    content: string;
+    responseContent: string;
    finishReason: CompletionFinishReason;
  } => {
-    const content = part.choices?.[0]?.delta?.content || '';
    const finishReason = part.choices?.[0]?.finish_reason || null;
-
+    const content = part.choices?.[0]?.delta?.content || '';
    // @ts-ignore
    const reasoningContent = part.choices?.[0]?.delta?.reasoning_content || '';
+    const isStreamEnd = !!finishReason;
+
+    // Parse think
+    const { reasoningContent: parsedThinkReasoningContent, content: parsedThinkContent } = (() => {
      if (reasoningContent || !parseThinkTag) {
        isInThinkTag = false;
-      return { reasoningContent, content, finishReason };
+        return { reasoningContent, content };
      }

      if (!content) {
        return {
          reasoningContent: '',
-        content: '',
-        finishReason
+          content: ''
        };
      }

@@ -206,8 +274,7 @@ export const parseReasoningStreamContent = () => {
      if (isInThinkTag === false) {
        return {
          reasoningContent: '',
-        content,
-        finishReason
+          content
        };
      }

@@ -216,20 +283,26 @@ export const parseReasoningStreamContent = () => {
        // Parse content think and answer
        startTagBuffer += content;
        // 太少内容时候，暂时不解析
-      if (startTagBuffer.length < startTag.length) {
+        if (startTagBuffer.length < thinkStartChars.length) {
+          if (isStreamEnd) {
+            const tmpContent = startTagBuffer;
+            startTagBuffer = '';
            return {
              reasoningContent: '',
-          content: '',
-          finishReason
+              content: tmpContent
+            };
+          }
+          return {
+            reasoningContent: '',
+            content: ''
          };
        }

-      if (startTagBuffer.startsWith(startTag)) {
+        if (startTagBuffer.startsWith(thinkStartChars)) {
          isInThinkTag = true;
          return {
-          reasoningContent: startTagBuffer.slice(startTag.length),
-          content: '',
-          finishReason
+            reasoningContent: startTagBuffer.slice(thinkStartChars.length),
+            content: ''
          };
        }

@@ -237,8 +310,7 @@ export const parseReasoningStreamContent = () => {
        isInThinkTag = false;
        return {
          reasoningContent: '',
-        content: startTagBuffer,
-        finishReason
+          content: startTagBuffer
        };
      }

@@ -257,50 +329,45 @@ export const parseReasoningStreamContent = () => {
      // endTagBuffer 专门用来记录疑似尾标签的内容
      if (endTagBuffer) {
        endTagBuffer += content;
-      if (endTagBuffer.includes(endTag)) {
+        if (endTagBuffer.includes(thinkEndChars)) {
          isInThinkTag = false;
-        const answer = endTagBuffer.slice(endTag.length);
+          const answer = endTagBuffer.slice(thinkEndChars.length);
          return {
            reasoningContent: '',
-          content: answer,
-          finishReason
+            content: answer
          };
-      } else if (endTagBuffer.length >= endTag.length) {
+        } else if (endTagBuffer.length >= thinkEndChars.length) {
          // 缓存内容超出尾标签长度，且仍未命中 </think>，则认为本次猜测 </think> 失败，仍处于 think 阶段。
          const tmp = endTagBuffer;
          endTagBuffer = '';
          return {
            reasoningContent: tmp,
-          content: '',
-          finishReason
+            content: ''
          };
        }
        return {
          reasoningContent: '',
-        content: '',
-        finishReason
+          content: ''
        };
-    } else if (content.includes(endTag)) {
+      } else if (content.includes(thinkEndChars)) {
        // 返回内容，完整命中</think>，直接结束
        isInThinkTag = false;
-      const [think, answer] = content.split(endTag);
+        const [think, answer] = content.split(thinkEndChars);
        return {
          reasoningContent: think,
-        content: answer,
-        finishReason
+          content: answer
        };
      } else {
        // 无 buffer，且未命中 </think>，开始疑似 </think> 检测。
-      for (let i = 1; i < endTag.length; i++) {
-        const partialEndTag = endTag.slice(0, i);
+        for (let i = 1; i < thinkEndChars.length; i++) {
+          const partialEndTag = thinkEndChars.slice(0, i);
          // 命中一部分尾标签
          if (content.endsWith(partialEndTag)) {
            const think = content.slice(0, -partialEndTag.length);
            endTagBuffer += partialEndTag;
            return {
              reasoningContent: think,
-            content: '',
-            finishReason
+              content: ''
            };
          }
        }
@@ -309,15 +376,74 @@ export const parseReasoningStreamContent = () => {
      // 完全未命中尾标签，还是 think 阶段。
      return {
        reasoningContent: content,
-      content: '',
+        content: ''
+      };
+    })();
+
+    // Parse datset cite
+    if (retainDatasetCite) {
+      return {
+        reasoningContent: parsedThinkReasoningContent,
+        content: parsedThinkContent,
+        responseContent: parsedThinkContent,
+        finishReason
+      };
+    }
+
+    // 缓存包含 [ 的字符串，直到超出 maxCiteBufferLength 再一次性返回
+    const parseCite = (text: string) => {
+      // 结束时，返回所有剩余内容
+      if (isStreamEnd) {
+        const content = citeBuffer + text;
+        return {
+          content: removeDatasetCiteText(content, false)
+        };
+      }
+
+      // 新内容包含 [，初始化缓冲数据
+      if (text.includes('[')) {
+        const index = text.indexOf('[');
+        const beforeContent = citeBuffer + text.slice(0, index);
+        citeBuffer = text.slice(index);
+
+        // beforeContent 可能是：普通字符串，带 [ 的字符串
+        return {
+          content: removeDatasetCiteText(beforeContent, false)
+        };
+      }
+      // 处于 Cite 缓冲区，判断是否满足条件
+      else if (citeBuffer) {
+        citeBuffer += text;
+
+        // 检查缓冲区长度是否达到完整Quote长度或已经流结束
+        if (citeBuffer.length >= maxCiteBufferLength) {
+          const content = removeDatasetCiteText(citeBuffer, false);
+          citeBuffer = '';
+
+          return {
+            content
+          };
+        } else {
+          // 暂时不返回内容
+          return { content: '' };
+        }
+      }
+
+      return {
+        content: text
+      };
+    };
+    const { content: pasedCiteContent } = parseCite(parsedThinkContent);
+
+    return {
+      reasoningContent: parsedThinkReasoningContent,
+      content: parsedThinkContent,
+      responseContent: pasedCiteContent,
      finishReason
    };
  };

-  const getStartTagBuffer = () => startTagBuffer;
-
  return {
-    parsePart,
-    getStartTagBuffer
+    parsePart
  };
 };
--- a/packages/service/core/app/plugin/utils.ts
+++ b/packages/service/core/app/plugin/utils.ts
@@ -31,5 +31,6 @@ export const computedPluginUsage = async ({
    return plugin.hasTokenFee ? pluginCurrentCost + childrenUsages : pluginCurrentCost;
  }

+  // Personal plugins are charged regardless of whether they are successful or not
  return childrenUsages;
 };
--- a/packages/service/core/workflow/dispatch/agent/classifyQuestion.ts
+++ b/packages/service/core/workflow/dispatch/agent/classifyQuestion.ts
@@ -19,7 +19,7 @@ import { DispatchNodeResultType } from '@fastgpt/global/core/workflow/runtime/ty
 import { chatValue2RuntimePrompt } from '@fastgpt/global/core/chat/adapt';
 import { getHandleId } from '@fastgpt/global/core/workflow/utils';
 import { loadRequestMessages } from '../../../chat/utils';
-import { llmCompletionsBodyFormat, llmResponseToAnswerText } from '../../../ai/utils';
+import { llmCompletionsBodyFormat, formatLLMResponse } from '../../../ai/utils';
 import { addLog } from '../../../../common/system/log';
 import { ModelTypeEnum } from '../../../../../global/core/ai/model';
 import { replaceVariable } from '@fastgpt/global/common/string/tools';
@@ -135,13 +135,13 @@ const completions = async ({
        model: cqModel.model,
        temperature: 0.01,
        messages: requestMessages,
-        stream: false
+        stream: true
      },
      cqModel
    ),
    userKey: externalProvider.openaiAccount
  });
-  const { text: answer, usage } = await llmResponseToAnswerText(response);
+  const { text: answer, usage } = await formatLLMResponse(response);

  // console.log(JSON.stringify(chats2GPTMessages({ messages, reserveId: false }), null, 2));
  // console.log(answer, '----');
--- a/packages/service/core/workflow/dispatch/agent/extract.ts
+++ b/packages/service/core/workflow/dispatch/agent/extract.ts
@@ -30,7 +30,7 @@ import {
 import { ChatCompletionRequestMessageRoleEnum } from '@fastgpt/global/core/ai/constants';
 import { DispatchNodeResultType } from '@fastgpt/global/core/workflow/runtime/type';
 import { chatValue2RuntimePrompt } from '@fastgpt/global/core/chat/adapt';
-import { llmCompletionsBodyFormat, llmResponseToAnswerText } from '../../../ai/utils';
+import { llmCompletionsBodyFormat, formatLLMResponse } from '../../../ai/utils';
 import { ModelTypeEnum } from '../../../../../global/core/ai/model';
 import {
  getExtractJsonPrompt,
@@ -226,10 +226,10 @@ const toolChoice = async (props: ActionProps) => {
    }
  ];

-  const { response } = (await createChatCompletion({
+  const { response } = await createChatCompletion({
    body: llmCompletionsBodyFormat(
      {
-        stream: false,
+        stream: true,
        model: extractModel.model,
        temperature: 0.01,
        messages: filterMessages,
@@ -239,16 +239,15 @@ const toolChoice = async (props: ActionProps) => {
      extractModel
    ),
    userKey: externalProvider.openaiAccount
-  })) as { response: UnStreamChatType };
+  });
+  const { toolCalls, usage } = await formatLLMResponse(response);

  const arg: Record<string, any> = (() => {
    try {
-      return json5.parse(
-        response?.choices?.[0]?.message?.tool_calls?.[0]?.function?.arguments || ''
-      );
+      return json5.parse(toolCalls?.[0]?.function?.arguments || '');
    } catch (error) {
      console.log(agentFunction.parameters);
-      console.log(response.choices?.[0]?.message?.tool_calls?.[0]?.function);
+      console.log(toolCalls?.[0]?.function);
      console.log('Your model may not support tool_call', error);
      return {};
    }
@@ -257,11 +256,10 @@ const toolChoice = async (props: ActionProps) => {
  const AIMessages: ChatCompletionMessageParam[] = [
    {
      role: ChatCompletionRequestMessageRoleEnum.Assistant,
-      tool_calls: response.choices?.[0]?.message?.tool_calls
+      tool_calls: toolCalls
    }
  ];

-  const usage = response.usage;
  const inputTokens = usage?.prompt_tokens || (await countGptMessagesTokens(filterMessages, tools));
  const outputTokens = usage?.completion_tokens || (await countGptMessagesTokens(AIMessages));
  return {
@@ -321,13 +319,13 @@ Human: ${content}`
        model: extractModel.model,
        temperature: 0.01,
        messages: requestMessages,
-        stream: false
+        stream: true
      },
      extractModel
    ),
    userKey: externalProvider.openaiAccount
  });
-  const { text: answer, usage } = await llmResponseToAnswerText(response);
+  const { text: answer, usage } = await formatLLMResponse(response);
  const inputTokens = usage?.prompt_tokens || (await countMessagesTokens(messages));
  const outputTokens = usage?.completion_tokens || (await countPromptTokens(answer));

--- a/packages/service/core/workflow/dispatch/agent/runTool/functionCall.ts
+++ b/packages/service/core/workflow/dispatch/agent/runTool/functionCall.ts
@@ -26,7 +26,12 @@ import { getNanoid, sliceStrStartEnd } from '@fastgpt/global/common/string/tools
 import { AIChatItemType } from '@fastgpt/global/core/chat/type';
 import { GPTMessages2Chats } from '@fastgpt/global/core/chat/adapt';
 import { formatToolResponse, initToolCallEdges, initToolNodes } from './utils';
-import { computedMaxToken, llmCompletionsBodyFormat } from '../../../../ai/utils';
+import {
+  computedMaxToken,
+  llmCompletionsBodyFormat,
+  removeDatasetCiteText,
+  parseLLMStreamResponse
+} from '../../../../ai/utils';
 import { toolValueTypeList } from '@fastgpt/global/core/workflow/constants';
 import { WorkflowInteractiveResponseType } from '@fastgpt/global/core/workflow/template/system/interactive/type';
 import { ChatItemValueTypeEnum } from '@fastgpt/global/core/chat/constants';
@@ -48,6 +53,7 @@ export const runToolWithFunctionCall = async (
    runtimeEdges,
    externalProvider,
    stream,
+    retainDatasetCite = true,
    workflowStreamResponse,
    params: {
      temperature,
@@ -261,7 +267,8 @@ export const runToolWithFunctionCall = async (
        res,
        toolNodes,
        stream: aiResponse,
-        workflowStreamResponse
+        workflowStreamResponse,
+        retainDatasetCite
      });

      return {
@@ -288,8 +295,18 @@ export const runToolWithFunctionCall = async (
          ]
        : [];

+      const answer = result.choices?.[0]?.message?.content || '';
+      if (answer) {
+        workflowStreamResponse?.({
+          event: SseResponseEventEnum.fastAnswer,
+          data: textAdaptGptResponse({
+            text: removeDatasetCiteText(answer, retainDatasetCite)
+          })
+        });
+      }
+
      return {
-        answer: result.choices?.[0]?.message?.content || '',
+        answer,
        functionCalls: toolCalls,
        inputTokens: usage?.prompt_tokens,
        outputTokens: usage?.completion_tokens
@@ -509,12 +526,14 @@ async function streamResponse({
  res,
  toolNodes,
  stream,
-  workflowStreamResponse
+  workflowStreamResponse,
+  retainDatasetCite
 }: {
  res: NextApiResponse;
  toolNodes: ToolNodeItemType[];
  stream: StreamChatType;
  workflowStreamResponse?: WorkflowResponseType;
+  retainDatasetCite?: boolean;
 }) {
  const write = responseWriteController({
    res,
@@ -526,6 +545,8 @@ async function streamResponse({
  let functionId = getNanoid();
  let usage = getLLMDefaultUsage();

+  const { parsePart } = parseLLMStreamResponse();
+
  for await (const part of stream) {
    usage = part.usage || usage;
    if (res.closed) {
@@ -533,17 +554,21 @@ async function streamResponse({
      break;
    }

+    const { content: toolChoiceContent, responseContent } = parsePart({
+      part,
+      parseThinkTag: false,
+      retainDatasetCite
+    });
+
    const responseChoice = part.choices?.[0]?.delta;
+    textAnswer += toolChoiceContent;

-    if (responseChoice.content) {
-      const content = responseChoice?.content || '';
-      textAnswer += content;
-
+    if (responseContent) {
      workflowStreamResponse?.({
        write,
        event: SseResponseEventEnum.answer,
        data: textAdaptGptResponse({
-          text: content
+          text: responseContent
        })
      });
    } else if (responseChoice.function_call) {
--- a/packages/service/core/workflow/dispatch/agent/runTool/promptCall.ts
+++ b/packages/service/core/workflow/dispatch/agent/runTool/promptCall.ts
@@ -29,8 +29,9 @@ import { formatToolResponse, initToolCallEdges, initToolNodes } from './utils';
 import {
  computedMaxToken,
  llmCompletionsBodyFormat,
+  removeDatasetCiteText,
  parseReasoningContent,
-  parseReasoningStreamContent
+  parseLLMStreamResponse
 } from '../../../../ai/utils';
 import { WorkflowResponseType } from '../../type';
 import { toolValueTypeList } from '@fastgpt/global/core/workflow/constants';
@@ -60,6 +61,7 @@ export const runToolWithPromptCall = async (
    runtimeEdges,
    externalProvider,
    stream,
+    retainDatasetCite = true,
    workflowStreamResponse,
    params: {
      temperature,
@@ -275,7 +277,8 @@ export const runToolWithPromptCall = async (
        toolNodes,
        stream: aiResponse,
        workflowStreamResponse,
-        aiChatReasoning
+        aiChatReasoning,
+        retainDatasetCite
      });

      return {
@@ -318,7 +321,7 @@ export const runToolWithPromptCall = async (
    workflowStreamResponse?.({
      event: SseResponseEventEnum.fastAnswer,
      data: textAdaptGptResponse({
-        reasoning_content: reasoning
+        reasoning_content: removeDatasetCiteText(reasoning, retainDatasetCite)
      })
    });
  }
@@ -344,7 +347,7 @@ export const runToolWithPromptCall = async (
      workflowStreamResponse?.({
        event: SseResponseEventEnum.fastAnswer,
        data: textAdaptGptResponse({
-          text: replaceAnswer
+          text: removeDatasetCiteText(replaceAnswer, retainDatasetCite)
        })
      });
    }
@@ -566,13 +569,15 @@ async function streamResponse({
  res,
  stream,
  workflowStreamResponse,
-  aiChatReasoning
+  aiChatReasoning,
+  retainDatasetCite
 }: {
  res: NextApiResponse;
  toolNodes: ToolNodeItemType[];
  stream: StreamChatType;
  workflowStreamResponse?: WorkflowResponseType;
  aiChatReasoning?: boolean;
+  retainDatasetCite?: boolean;
 }) {
  const write = responseWriteController({
    res,
@@ -585,7 +590,7 @@ async function streamResponse({
  let finish_reason: CompletionFinishReason = null;
  let usage = getLLMDefaultUsage();

-  const { parsePart, getStartTagBuffer } = parseReasoningStreamContent();
+  const { parsePart } = parseLLMStreamResponse();

  for await (const part of stream) {
    usage = part.usage || usage;
@@ -595,11 +600,16 @@ async function streamResponse({
      break;
    }

-    const { reasoningContent, content, finishReason } = parsePart(part, aiChatReasoning);
+    const { reasoningContent, content, responseContent, finishReason } = parsePart({
+      part,
+      parseThinkTag: aiChatReasoning,
+      retainDatasetCite
+    });
    finish_reason = finish_reason || finishReason;
    answer += content;
    reasoning += reasoningContent;

+    // Reasoning response
    if (aiChatReasoning && reasoningContent) {
      workflowStreamResponse?.({
        write,
@@ -612,13 +622,15 @@ async function streamResponse({

    if (content) {
      if (startResponseWrite) {
+        if (responseContent) {
          workflowStreamResponse?.({
            write,
            event: SseResponseEventEnum.answer,
            data: textAdaptGptResponse({
-            text: content
+              text: responseContent
            })
          });
+        }
      } else if (answer.length >= 3) {
        answer = answer.trimStart();
        if (/0(:|：)/.test(answer)) {
@@ -640,22 +652,6 @@ async function streamResponse({
    }
  }

-  if (answer === '') {
-    answer = getStartTagBuffer();
-    if (/0(:|：)/.test(answer)) {
-      // find first : index
-      const firstIndex = answer.indexOf('0:') !== -1 ? answer.indexOf('0:') : answer.indexOf('0：');
-      answer = answer.substring(firstIndex + 2).trim();
-      workflowStreamResponse?.({
-        write,
-        event: SseResponseEventEnum.answer,
-        data: textAdaptGptResponse({
-          text: answer
-        })
-      });
-    }
-  }
-
  return { answer, reasoning, finish_reason, usage };
 }

--- a/packages/service/core/workflow/dispatch/agent/runTool/toolChoice.ts
+++ b/packages/service/core/workflow/dispatch/agent/runTool/toolChoice.ts
@@ -26,7 +26,12 @@ import { countGptMessagesTokens } from '../../../../../common/string/tiktoken/in
 import { GPTMessages2Chats } from '@fastgpt/global/core/chat/adapt';
 import { AIChatItemType } from '@fastgpt/global/core/chat/type';
 import { formatToolResponse, initToolCallEdges, initToolNodes } from './utils';
-import { computedMaxToken, llmCompletionsBodyFormat } from '../../../../ai/utils';
+import {
+  computedMaxToken,
+  llmCompletionsBodyFormat,
+  removeDatasetCiteText,
+  parseLLMStreamResponse
+} from '../../../../ai/utils';
 import { getNanoid, sliceStrStartEnd } from '@fastgpt/global/common/string/tools';
 import { toolValueTypeList } from '@fastgpt/global/core/workflow/constants';
 import { WorkflowInteractiveResponseType } from '@fastgpt/global/core/workflow/template/system/interactive/type';
@@ -89,12 +94,13 @@ export const runToolWithToolChoice = async (
    interactiveEntryToolParams,
    ...workflowProps
  } = props;
-  const {
+  let {
    res,
    requestOrigin,
    runtimeNodes,
    runtimeEdges,
    stream,
+    retainDatasetCite = true,
    externalProvider,
    workflowStreamResponse,
    params: {
@@ -104,9 +110,11 @@ export const runToolWithToolChoice = async (
      aiChatTopP,
      aiChatStopSign,
      aiChatResponseFormat,
-      aiChatJsonSchema
+      aiChatJsonSchema,
+      aiChatReasoning
    }
  } = workflowProps;
+  aiChatReasoning = !!aiChatReasoning && !!toolModel.reasoning;

  if (maxRunToolTimes <= 0 && response) {
    return response;
@@ -279,6 +287,7 @@ export const runToolWithToolChoice = async (
      messages: requestMessages,
      tools,
      tool_choice: 'auto',
+      parallel_tool_calls: true,
      temperature,
      max_tokens,
      top_p: aiChatTopP,
@@ -288,7 +297,7 @@ export const runToolWithToolChoice = async (
    },
    toolModel
  );
-  // console.log(JSON.stringify(filterMessages, null, 2), '==requestMessages');
+  // console.log(JSON.stringify(requestBody, null, 2), '==requestMessages');
  /* Run llm */
  const {
    response: aiResponse,
@@ -320,7 +329,9 @@ export const runToolWithToolChoice = async (
        res,
        workflowStreamResponse,
        toolNodes,
-        stream: aiResponse
+        stream: aiResponse,
+        aiChatReasoning,
+        retainDatasetCite
      });

      return {
@@ -335,11 +346,38 @@ export const runToolWithToolChoice = async (
      const finish_reason = result.choices?.[0]?.finish_reason as CompletionFinishReason;
      const calls = result.choices?.[0]?.message?.tool_calls || [];
      const answer = result.choices?.[0]?.message?.content || '';
+      // @ts-ignore
+      const reasoningContent = result.choices?.[0]?.message?.reasoning_content || '';
      const usage = result.usage;

-      // 加上name和avatar
+      if (aiChatReasoning && reasoningContent) {
+        workflowStreamResponse?.({
+          event: SseResponseEventEnum.fastAnswer,
+          data: textAdaptGptResponse({
+            reasoning_content: removeDatasetCiteText(reasoningContent, retainDatasetCite)
+          })
+        });
+      }
+
+      // 格式化 toolCalls
      const toolCalls = calls.map((tool) => {
        const toolNode = toolNodes.find((item) => item.nodeId === tool.function?.name);
+
+        // 不支持 stream 模式的模型的这里需要补一个响应给客户端
+        workflowStreamResponse?.({
+          event: SseResponseEventEnum.toolCall,
+          data: {
+            tool: {
+              id: tool.id,
+              toolName: toolNode?.name || '',
+              toolAvatar: toolNode?.avatar || '',
+              functionName: tool.function.name,
+              params: tool.function?.arguments ?? '',
+              response: ''
+            }
+          }
+        });
+
        return {
          ...tool,
          toolName: toolNode?.name || '',
@@ -347,27 +385,11 @@ export const runToolWithToolChoice = async (
        };
      });

-      // 不支持 stream 模式的模型的流失响应
-      toolCalls.forEach((tool) => {
-        workflowStreamResponse?.({
-          event: SseResponseEventEnum.toolCall,
-          data: {
-            tool: {
-              id: tool.id,
-              toolName: tool.toolName,
-              toolAvatar: tool.toolAvatar,
-              functionName: tool.function.name,
-              params: tool.function?.arguments ?? '',
-              response: ''
-            }
-          }
-        });
-      });
      if (answer) {
        workflowStreamResponse?.({
          event: SseResponseEventEnum.fastAnswer,
          data: textAdaptGptResponse({
-            text: answer
+            text: removeDatasetCiteText(answer, retainDatasetCite)
          })
        });
      }
@@ -627,12 +649,16 @@ async function streamResponse({
  res,
  toolNodes,
  stream,
-  workflowStreamResponse
+  workflowStreamResponse,
+  aiChatReasoning,
+  retainDatasetCite
 }: {
  res: NextApiResponse;
  toolNodes: ToolNodeItemType[];
  stream: StreamChatType;
  workflowStreamResponse?: WorkflowResponseType;
+  aiChatReasoning: boolean;
+  retainDatasetCite?: boolean;
 }) {
  const write = responseWriteController({
    res,
@@ -642,65 +668,89 @@ async function streamResponse({
  let textAnswer = '';
  let callingTool: { name: string; arguments: string } | null = null;
  let toolCalls: ChatCompletionMessageToolCall[] = [];
-  let finishReason: CompletionFinishReason = null;
+  let finish_reason: CompletionFinishReason = null;
  let usage = getLLMDefaultUsage();

+  const { parsePart } = parseLLMStreamResponse();
+
  for await (const part of stream) {
    usage = part.usage || usage;
    if (res.closed) {
      stream.controller?.abort();
-      finishReason = 'close';
+      finish_reason = 'close';
      break;
    }

+    const {
+      reasoningContent,
+      content: toolChoiceContent,
+      responseContent,
+      finishReason
+    } = parsePart({
+      part,
+      parseThinkTag: true,
+      retainDatasetCite
+    });
+    textAnswer += toolChoiceContent;
+    finish_reason = finishReason || finish_reason;
+
    const responseChoice = part.choices?.[0]?.delta;
-    const finish_reason = part.choices?.[0]?.finish_reason as CompletionFinishReason;
-    finishReason = finishReason || finish_reason;
-
-    if (responseChoice?.content) {
-      const content = responseChoice.content || '';
-      textAnswer += content;

+    // Reasoning response
+    if (aiChatReasoning && reasoningContent) {
      workflowStreamResponse?.({
        write,
        event: SseResponseEventEnum.answer,
        data: textAdaptGptResponse({
-          text: content
+          reasoning_content: reasoningContent
        })
      });
    }
-    if (responseChoice?.tool_calls?.[0]) {
-      // @ts-ignore
-      const toolCall: ChatCompletionMessageToolCall = responseChoice.tool_calls[0];
-      // In a stream response, only one tool is returned at a time.  If have id, description is executing a tool
+    if (responseContent) {
+      workflowStreamResponse?.({
+        write,
+        event: SseResponseEventEnum.answer,
+        data: textAdaptGptResponse({
+          text: responseContent
+        })
+      });
+    }
+    // Parse tool calls
+    if (responseChoice?.tool_calls?.length) {
+      responseChoice.tool_calls.forEach((toolCall) => {
+        const index = toolCall.index;
+
+        // Call new tool
        if (toolCall.id || callingTool) {
-        // Start call tool
+          // 有 id，代表新 call 工具
          if (toolCall.id) {
            callingTool = {
              name: toolCall.function?.name || '',
              arguments: toolCall.function?.arguments || ''
            };
          } else if (callingTool) {
-          // Continue call
-          callingTool.name += toolCall.function.name || '';
-          callingTool.arguments += toolCall.function.arguments || '';
+            // Continue call(Perhaps the name of the previous function was incomplete)
+            callingTool.name += toolCall.function?.name || '';
+            callingTool.arguments += toolCall.function?.arguments || '';
          }

-        const toolFunction = callingTool!;
+          if (!callingTool) {
+            return;
+          }

-        const toolNode = toolNodes.find((item) => item.nodeId === toolFunction.name);
+          const toolNode = toolNodes.find((item) => item.nodeId === callingTool!.name);

          if (toolNode) {
            // New tool, add to list.
            const toolId = getNanoid();
-          toolCalls.push({
+            toolCalls[index] = {
              ...toolCall,
              id: toolId,
              type: 'function',
-            function: toolFunction,
+              function: callingTool,
              toolName: toolNode.name,
              toolAvatar: toolNode.avatar
-          });
+            };

            workflowStreamResponse?.({
              event: SseResponseEventEnum.toolCall,
@@ -709,8 +759,8 @@ async function streamResponse({
                  id: toolId,
                  toolName: toolNode.name,
                  toolAvatar: toolNode.avatar,
-                functionName: toolFunction.name,
-                params: toolFunction?.arguments ?? '',
+                  functionName: callingTool.name,
+                  params: callingTool?.arguments ?? '',
                  response: ''
                }
              }
@@ -718,9 +768,9 @@ async function streamResponse({
            callingTool = null;
          }
        } else {
-        /* arg 插入最后一个工具的参数里 */
+          /* arg 追加到当前工具的参数里 */
          const arg: string = toolCall?.function?.arguments ?? '';
-        const currentTool = toolCalls[toolCalls.length - 1];
+          const currentTool = toolCalls[index];
          if (currentTool && arg) {
            currentTool.function.arguments += arg;

@@ -739,8 +789,9 @@ async function streamResponse({
            });
          }
        }
+      });
    }
  }

-  return { answer: textAnswer, toolCalls, finish_reason: finishReason, usage };
+  return { answer: textAnswer, toolCalls: toolCalls.filter(Boolean), finish_reason, usage };
 }
--- a/packages/service/core/workflow/dispatch/chat/oneapi.ts
+++ b/packages/service/core/workflow/dispatch/chat/oneapi.ts
@@ -4,7 +4,11 @@ import type { ChatItemType, UserChatItemValueItemType } from '@fastgpt/global/co
 import { ChatRoleEnum } from '@fastgpt/global/core/chat/constants';
 import { SseResponseEventEnum } from '@fastgpt/global/core/workflow/runtime/constants';
 import { textAdaptGptResponse } from '@fastgpt/global/core/workflow/runtime/utils';
-import { parseReasoningContent, parseReasoningStreamContent } from '../../../ai/utils';
+import {
+  removeDatasetCiteText,
+  parseReasoningContent,
+  parseLLMStreamResponse
+} from '../../../ai/utils';
 import { createChatCompletion } from '../../../ai/config';
 import type {
  ChatCompletionMessageParam,
@@ -75,7 +79,7 @@ export const dispatchChatCompletion = async (props: ChatProps): Promise<ChatResp
    res,
    requestOrigin,
    stream = false,
-    parseQuote = true,
+    retainDatasetCite = true,
    externalProvider,
    histories,
    node: { name, version },
@@ -159,8 +163,7 @@ export const dispatchChatCompletion = async (props: ChatProps): Promise<ChatResp
      userChatInput,
      systemPrompt,
      userFiles,
-      documentQuoteText,
-      parseQuote
+      documentQuoteText
    }),
    // Censor = true and system key, will check content
    (() => {
@@ -223,7 +226,8 @@ export const dispatchChatCompletion = async (props: ChatProps): Promise<ChatResp
        aiChatReasoning,
        parseThinkTag: modelConstantsData.reasoning,
        isResponseAnswerText,
-        workflowStreamResponse
+        workflowStreamResponse,
+        retainDatasetCite
      });

      return {
@@ -258,12 +262,11 @@ export const dispatchChatCompletion = async (props: ChatProps): Promise<ChatResp
      })();

      // Some models do not support streaming
-      if (stream) {
      if (aiChatReasoning && reasoningContent) {
        workflowStreamResponse?.({
          event: SseResponseEventEnum.fastAnswer,
          data: textAdaptGptResponse({
-              reasoning_content: reasoningContent
+            reasoning_content: removeDatasetCiteText(reasoningContent, retainDatasetCite)
          })
        });
      }
@@ -271,11 +274,10 @@ export const dispatchChatCompletion = async (props: ChatProps): Promise<ChatResp
        workflowStreamResponse?.({
          event: SseResponseEventEnum.fastAnswer,
          data: textAdaptGptResponse({
-              text: content
+            text: removeDatasetCiteText(content, retainDatasetCite)
          })
        });
      }
-      }

      return {
        answerText: content,
@@ -452,8 +454,7 @@ async function getChatMessages({
  systemPrompt,
  userChatInput,
  userFiles,
-  documentQuoteText,
-  parseQuote = true
+  documentQuoteText
 }: {
  model: LLMModelItemType;
  maxTokens?: number;
@@ -470,14 +471,13 @@ async function getChatMessages({

  userFiles: UserChatItemValueItemType['file'][];
  documentQuoteText?: string; // document quote
-  parseQuote?: boolean;
 }) {
  // Dataset prompt ====>
  // User role or prompt include question
  const quoteRole =
    aiChatQuoteRole === 'user' || datasetQuotePrompt.includes('{{question}}') ? 'user' : 'system';

-  const defaultQuotePrompt = getQuotePrompt(version, quoteRole, parseQuote);
+  const defaultQuotePrompt = getQuotePrompt(version, quoteRole);

  const datasetQuotePromptTemplate = datasetQuotePrompt || defaultQuotePrompt;

@@ -539,7 +539,8 @@ async function streamResponse({
  workflowStreamResponse,
  aiChatReasoning,
  parseThinkTag,
-  isResponseAnswerText
+  isResponseAnswerText,
+  retainDatasetCite = true
 }: {
  res: NextApiResponse;
  stream: StreamChatType;
@@ -547,6 +548,7 @@ async function streamResponse({
  aiChatReasoning?: boolean;
  parseThinkTag?: boolean;
  isResponseAnswerText?: boolean;
+  retainDatasetCite: boolean;
 }) {
  const write = responseWriteController({
    res,
@@ -557,7 +559,7 @@ async function streamResponse({
  let finish_reason: CompletionFinishReason = null;
  let usage: CompletionUsage = getLLMDefaultUsage();

-  const { parsePart, getStartTagBuffer } = parseReasoningStreamContent();
+  const { parsePart } = parseLLMStreamResponse();

  for await (const part of stream) {
    usage = part.usage || usage;
@@ -568,7 +570,11 @@ async function streamResponse({
      break;
    }

-    const { reasoningContent, content, finishReason } = parsePart(part, parseThinkTag);
+    const { reasoningContent, content, responseContent, finishReason } = parsePart({
+      part,
+      parseThinkTag,
+      retainDatasetCite
+    });
    finish_reason = finish_reason || finishReason;
    answer += content;
    reasoning += reasoningContent;
@@ -583,26 +589,12 @@ async function streamResponse({
      });
    }

-    if (isResponseAnswerText && content) {
+    if (isResponseAnswerText && responseContent) {
      workflowStreamResponse?.({
        write,
        event: SseResponseEventEnum.answer,
        data: textAdaptGptResponse({
-          text: content
-        })
-      });
-    }
-  }
-
-  // if answer is empty, try to get value from startTagBuffer. (Cause: The response content is too short to exceed the minimum parse length)
-  if (answer === '') {
-    answer = getStartTagBuffer();
-    if (isResponseAnswerText && answer) {
-      workflowStreamResponse?.({
-        write,
-        event: SseResponseEventEnum.answer,
-        data: textAdaptGptResponse({
-          text: answer
+          text: responseContent
        })
      });
    }
--- a/packages/service/core/workflow/dispatch/dataset/concat.ts
+++ b/packages/service/core/workflow/dispatch/dataset/concat.ts
@@ -21,7 +21,7 @@ export async function dispatchDatasetConcat(
  props: DatasetConcatProps
 ): Promise<DatasetConcatResponse> {
  const {
-    params: { limit = 1500, ...quoteMap }
+    params: { limit = 6000, ...quoteMap }
  } = props as DatasetConcatProps;

  const quoteList = Object.values(quoteMap).filter((list) => Array.isArray(list));
--- a/packages/service/core/workflow/dispatch/dataset/search.ts
+++ b/packages/service/core/workflow/dispatch/dataset/search.ts
@@ -55,11 +55,10 @@ export async function dispatchDatasetSearch(
    runningUserInfo: { tmbId },
    histories,
    node,
-    parseQuote = true,
    params: {
      datasets = [],
      similarity,
-      limit = 1500,
+      limit = 5000,
      userChatInput = '',
      authTmbId = false,
      collectionFilterMatch,
@@ -114,7 +113,6 @@ export async function dispatchDatasetSearch(
  if (datasetIds.length === 0) {
    return emptyResult;
  }
-  // console.log(concatQueries, rewriteQuery, aiExtensionResult);

  // get vector
  const vectorModel = getEmbeddingModel(
@@ -267,7 +265,7 @@ export async function dispatchDatasetSearch(
    [DispatchNodeResponseKeyEnum.nodeResponse]: responseData,
    nodeDispatchUsages,
    [DispatchNodeResponseKeyEnum.toolResponses]: {
-      prompt: getDatasetSearchToolResponsePrompt(parseQuote),
+      prompt: getDatasetSearchToolResponsePrompt(),
      quotes: searchRes.map((item) => ({
        id: item.id,
        sourceName: item.sourceName,
--- a/packages/service/core/workflow/dispatch/index.ts
+++ b/packages/service/core/workflow/dispatch/index.ts
@@ -135,7 +135,7 @@ export async function dispatchWorkFlow(data: Props): Promise<DispatchFlowRespons
    timezone,
    externalProvider,
    stream = false,
-    parseQuote = true,
+    retainDatasetCite = true,
    version = 'v1',
    responseDetail = true,
    responseAllData = true,
@@ -607,7 +607,7 @@ export async function dispatchWorkFlow(data: Props): Promise<DispatchFlowRespons
      timezone,
      externalProvider,
      stream,
-      parseQuote,
+      retainDatasetCite,
      node,
      runtimeNodes,
      runtimeEdges,
--- a/packages/service/core/workflow/dispatch/utils.ts
+++ b/packages/service/core/workflow/dispatch/utils.ts
@@ -1,10 +1,7 @@
 import { getErrText } from '@fastgpt/global/common/error/utils';
 import { ChatRoleEnum } from '@fastgpt/global/core/chat/constants';
 import type { ChatItemType } from '@fastgpt/global/core/chat/type.d';
-import {
-  WorkflowIOValueTypeEnum,
-  NodeOutputKeyEnum
-} from '@fastgpt/global/core/workflow/constants';
+import { NodeOutputKeyEnum } from '@fastgpt/global/core/workflow/constants';
 import {
  RuntimeEdgeItemType,
  RuntimeNodeItemType,
@@ -34,31 +31,22 @@ export const getWorkflowResponseWrite = ({
  return ({
    write,
    event,
-    data,
-    stream
+    data
  }: {
    write?: (text: string) => void;
    event: SseResponseEventEnum;
    data: Record<string, any>;
-    stream?: boolean; // Focus set stream response
  }) => {
-    const useStreamResponse = stream ?? streamResponse;
+    const useStreamResponse = streamResponse;

    if (!res || res.closed || !useStreamResponse) return;

    // Forbid show detail
-    const detailEvent: Record<string, 1> = {
-      [SseResponseEventEnum.error]: 1,
-      [SseResponseEventEnum.flowNodeStatus]: 1,
-      [SseResponseEventEnum.flowResponses]: 1,
-      [SseResponseEventEnum.interactive]: 1,
-      [SseResponseEventEnum.toolCall]: 1,
-      [SseResponseEventEnum.toolParams]: 1,
-      [SseResponseEventEnum.toolResponse]: 1,
-      [SseResponseEventEnum.updateVariables]: 1,
-      [SseResponseEventEnum.flowNodeResponse]: 1
+    const notDetailEvent: Record<string, 1> = {
+      [SseResponseEventEnum.answer]: 1,
+      [SseResponseEventEnum.fastAnswer]: 1
    };
-    if (!detail && detailEvent[event]) return;
+    if (!detail && !notDetailEvent[event]) return;

    // Forbid show running status
    const statusEvent: Record<string, 1> = {
--- a/packages/templates/src/CQ/template.json
+++ b/packages/templates/src/CQ/template.json
@@ -308,7 +308,7 @@
            "key": "limit",
            "renderTypeList": ["hidden"],
            "label": "",
-            "value": 1500,
+            "value": 5000,
            "valueType": "number"
          },
          {
--- a/packages/templates/src/simpleDatasetChat/template.json
+++ b/packages/templates/src/simpleDatasetChat/template.json
@@ -211,7 +211,7 @@
            "key": "limit",
            "renderTypeList": ["hidden"],
            "label": "",
-            "value": 1500,
+            "value": 5000,
            "valueType": "number"
          },
          {
--- a/projects/app/src/components/Markdown/A.tsx
+++ b/projects/app/src/components/Markdown/A.tsx
@@ -21,16 +21,16 @@ import MyBox from '@fastgpt/web/components/common/MyBox';
 import { getCollectionSourceData } from '@fastgpt/global/core/dataset/collection/utils';
 import Markdown from '.';
 import { getSourceNameIcon } from '@fastgpt/global/core/dataset/utils';
+import { Types } from 'mongoose';

-const A = ({ children, chatAuthData, ...props }: any) => {
+const A = ({ children, chatAuthData, showAnimation, ...props }: any) => {
  const { t } = useTranslation();

  const { isOpen, onOpen, onClose } = useDisclosure();
+  const content = useMemo(() => String(children), [children]);

  // empty href link
  if (!props.href && typeof children?.[0] === 'string') {
-    const text = useMemo(() => String(children), [children]);
-
    return (
      <MyTooltip label={t('common:core.chat.markdown.Quick Question')}>
        <Button
@@ -38,16 +38,23 @@ const A = ({ children, chatAuthData, ...props }: any) => {
          size={'xs'}
          borderRadius={'md'}
          my={1}
-          onClick={() => eventBus.emit(EventNameEnum.sendQuestion, { text })}
+          onClick={() => eventBus.emit(EventNameEnum.sendQuestion, { text: content })}
        >
-          {text}
+          {content}
        </Button>
      </MyTooltip>
    );
  }

-  // Quote
-  if (props.href?.startsWith('QUOTE') && typeof children?.[0] === 'string') {
+  // Cite
+  if (
+    (props.href?.startsWith('CITE') || props.href?.startsWith('QUOTE')) &&
+    typeof children?.[0] === 'string'
+  ) {
+    if (!Types.ObjectId.isValid(content)) {
+      return <></>;
+    }
+
    const {
      data: quoteData,
      loading,
@@ -74,6 +81,7 @@ const A = ({ children, chatAuthData, ...props }: any) => {
        onClose={onClose}
        onOpen={() => {
          onOpen();
+          if (showAnimation) return;
          getQuoteDataById(String(children));
        }}
        trigger={'hover'}
@@ -90,7 +98,7 @@ const A = ({ children, chatAuthData, ...props }: any) => {
          </Button>
        </PopoverTrigger>
        <PopoverContent boxShadow={'lg'} w={'500px'} maxW={'90vw'} py={4}>
-          <MyBox isLoading={loading}>
+          <MyBox isLoading={loading || showAnimation}>
            <PopoverArrow />
            <PopoverBody py={0} px={0} fontSize={'sm'}>
              <Flex px={4} pb={1} justifyContent={'space-between'}>
--- a/projects/app/src/components/Markdown/index.tsx
+++ b/projects/app/src/components/Markdown/index.tsx
@@ -60,9 +60,9 @@ const MarkdownRender = ({
      img: Image,
      pre: RewritePre,
      code: Code,
-      a: (props: any) => <A {...props} chatAuthData={chatAuthData} />
+      a: (props: any) => <A {...props} showAnimation={showAnimation} chatAuthData={chatAuthData} />
    };
-  }, [chatAuthData]);
+  }, [chatAuthData, showAnimation]);

  const formatSource = useMemo(() => {
    if (showAnimation || forbidZhFormat) return source;
--- a/projects/app/src/components/Markdown/utils.ts
+++ b/projects/app/src/components/Markdown/utils.ts
@@ -27,14 +27,14 @@ export const mdTextFormat = (text: string) => {
    return match;
  });

-  // 处理 [quote:id] 格式引用，将 [quote:675934a198f46329dfc6d05a] 转换为 [675934a198f46329dfc6d05a](QUOTE)
+  // 处理 [quote:id] 格式引用，将 [quote:675934a198f46329dfc6d05a] 转换为 [675934a198f46329dfc6d05a](CITE)
  text = text
    // .replace(
    //   /([\u4e00-\u9fa5\u3000-\u303f])([a-zA-Z0-9])|([a-zA-Z0-9])([\u4e00-\u9fa5\u3000-\u303f])/g,
    //   '$1$3 $2$4'
    // )
-    // 处理 格式引用，将 [675934a198f46329dfc6d05a] 转换为 [675934a198f46329dfc6d05a](QUOTE)
-    .replace(/\[([a-f0-9]{24})\](?!\()/g, '[$1](QUOTE)');
+    // 处理 格式引用，将 [675934a198f46329dfc6d05a] 转换为 [675934a198f46329dfc6d05a](CITE)
+    .replace(/\[([a-f0-9]{24})\](?!\()/g, '[$1](CITE)');

  // 处理链接后的中文标点符号，增加空格
  text = text.replace(/(https?:\/\/[^\s，。！？；：、]+)([，。！？；：、])/g, '$1 $2');
--- a/projects/app/src/components/core/chat/ChatContainer/ChatBox/components/ChatItem.tsx
+++ b/projects/app/src/components/core/chat/ChatContainer/ChatBox/components/ChatItem.tsx
@@ -240,11 +240,6 @@ const ChatItem = (props: Props) => {
      quoteId?: string;
    }) => {
      if (!setQuoteData) return;
-      if (isChatting)
-        return toast({
-          title: t('chat:chat.waiting_for_response'),
-          status: 'info'
-        });

      const collectionIdList = collectionId
        ? [collectionId]
@@ -277,18 +272,7 @@ const ChatItem = (props: Props) => {
              }
      });
    },
-    [
-      setQuoteData,
-      isChatting,
-      toast,
-      t,
-      quoteList,
-      isShowReadRawSource,
-      appId,
-      chatId,
-      chat.dataId,
-      outLinkAuthData
-    ]
+    [setQuoteData, quoteList, isShowReadRawSource, appId, chatId, chat.dataId, outLinkAuthData]
  );

  useEffect(() => {
--- a/projects/app/src/components/core/chat/components/AIResponseBox.tsx
+++ b/projects/app/src/components/core/chat/components/AIResponseBox.tsx
@@ -96,8 +96,6 @@ const RenderText = React.memo(function RenderText({
  text: string;
  chatItemDataId: string;
 }) {
-  const isResponseDetail = useContextSelector(ChatItemContext, (v) => v.isResponseDetail);
-
  const appId = useContextSelector(ChatBoxContext, (v) => v.appId);
  const chatId = useContextSelector(ChatBoxContext, (v) => v.chatId);
  const outLinkAuthData = useContextSelector(ChatBoxContext, (v) => v.outLinkAuthData);
@@ -106,10 +104,8 @@ const RenderText = React.memo(function RenderText({
    if (!text) return '';

    // Remove quote references if not showing response detail
-    return isResponseDetail
-      ? text
-      : text.replace(/\[([a-f0-9]{24})\]\(QUOTE\)/g, '').replace(/\[([a-f0-9]{24})\](?!\()/g, '');
-  }, [text, isResponseDetail]);
+    return text;
+  }, [text]);

  const chatAuthData = useCreation(() => {
    return { appId, chatId, chatItemDataId, ...outLinkAuthData };
--- a/projects/app/src/components/core/dataset/SearchParamsTip.tsx
+++ b/projects/app/src/components/core/dataset/SearchParamsTip.tsx
@@ -12,7 +12,7 @@ import { getWebLLMModel } from '@/web/common/system/utils';
 const SearchParamsTip = ({
  searchMode,
  similarity = 0,
-  limit = 1500,
+  limit = 5000,
  responseEmptyText,
  usingReRank = false,
  datasetSearchUsingExtensionQuery,
--- a/projects/app/src/pageComponents/account/info/UpdatePswModal.tsx
+++ b/projects/app/src/pageComponents/account/info/UpdatePswModal.tsx
@@ -5,8 +5,8 @@ import { useTranslation } from 'next-i18next';
 import { useForm } from 'react-hook-form';
 import { useRequest2 } from '@fastgpt/web/hooks/useRequest';
 import { updatePasswordByOld } from '@/web/support/user/api';
-import { checkPasswordRule } from '@/web/support/user/login/constants';
 import { useToast } from '@fastgpt/web/hooks/useToast';
+import { checkPasswordRule } from '@fastgpt/global/common/string/password';

 type FormType = {
  oldPsw: string;
--- a/projects/app/src/pageComponents/login/ForgetPasswordForm.tsx
+++ b/projects/app/src/pageComponents/login/ForgetPasswordForm.tsx
@@ -1,7 +1,7 @@
 import React, { Dispatch } from 'react';
 import { FormControl, Box, Input, Button } from '@chakra-ui/react';
 import { useForm } from 'react-hook-form';
-import { LoginPageTypeEnum, checkPasswordRule } from '@/web/support/user/login/constants';
+import { LoginPageTypeEnum } from '@/web/support/user/login/constants';
 import { postFindPassword } from '@/web/support/user/api';
 import { useSendCode } from '@/web/support/user/hooks/useSendCode';
 import type { ResLogin } from '@/global/support/api/userRes.d';
@@ -9,6 +9,7 @@ import { useToast } from '@fastgpt/web/hooks/useToast';
 import { useSystemStore } from '@/web/common/system/useSystemStore';
 import { useTranslation } from 'next-i18next';
 import { useRequest2 } from '@fastgpt/web/hooks/useRequest';
+import { checkPasswordRule } from '@fastgpt/global/common/string/password';

 interface Props {
  setPageType: Dispatch<`${LoginPageTypeEnum}`>;
--- a/projects/app/src/pageComponents/login/RegisterForm.tsx
+++ b/projects/app/src/pageComponents/login/RegisterForm.tsx
@@ -1,7 +1,7 @@
 import React, { Dispatch } from 'react';
 import { FormControl, Box, Input, Button } from '@chakra-ui/react';
 import { useForm } from 'react-hook-form';
-import { LoginPageTypeEnum, checkPasswordRule } from '@/web/support/user/login/constants';
+import { LoginPageTypeEnum } from '@/web/support/user/login/constants';
 import { postRegister } from '@/web/support/user/api';
 import { useSendCode } from '@/web/support/user/hooks/useSendCode';
 import type { ResLogin } from '@/global/support/api/userRes';
@@ -19,6 +19,7 @@ import {
  getSourceDomain,
  removeFastGPTSem
 } from '@/web/support/marketing/utils';
+import { checkPasswordRule } from '@fastgpt/global/common/string/password';

 interface Props {
  loginSuccess: (e: ResLogin) => void;
--- a/projects/app/src/pages/api/core/ai/model/test.ts
+++ b/projects/app/src/pages/api/core/ai/model/test.ts
@@ -16,7 +16,7 @@ import { reRankRecall } from '@fastgpt/service/core/ai/rerank';
 import { aiTranscriptions } from '@fastgpt/service/core/ai/audio/transcriptions';
 import { isProduction } from '@fastgpt/global/common/system/constants';
 import * as fs from 'fs';
-import { llmCompletionsBodyFormat, llmResponseToAnswerText } from '@fastgpt/service/core/ai/utils';
+import { llmCompletionsBodyFormat, formatLLMResponse } from '@fastgpt/service/core/ai/utils';

 export type testQuery = { model: string; channelId?: number };

@@ -78,7 +78,7 @@ const testLLMModel = async (model: LLMModelItemType, headers: Record<string, str
    model
  );

-  const { response, isStreamResponse } = await createChatCompletion({
+  const { response } = await createChatCompletion({
    modelData: model,
    body: requestBody,
    options: {
@@ -88,7 +88,7 @@ const testLLMModel = async (model: LLMModelItemType, headers: Record<string, str
      }
    }
  });
-  const { text: answer } = await llmResponseToAnswerText(response);
+  const { text: answer } = await formatLLMResponse(response);

  if (answer) {
    return answer;
--- a/projects/app/src/pages/api/core/chat/getPaginationRecords.ts
+++ b/projects/app/src/pages/api/core/chat/getPaginationRecords.ts
@@ -9,7 +9,10 @@ import { authChatCrud } from '@/service/support/permission/auth/chat';
 import { MongoApp } from '@fastgpt/service/core/app/schema';
 import { AppErrEnum } from '@fastgpt/global/common/error/code/app';
 import { ChatItemValueTypeEnum, ChatRoleEnum } from '@fastgpt/global/core/chat/constants';
-import { filterPublicNodeResponseData } from '@fastgpt/global/core/chat/utils';
+import {
+  filterPublicNodeResponseData,
+  removeAIResponseCite
+} from '@fastgpt/global/core/chat/utils';
 import { GetChatTypeEnum } from '@/global/core/chat/constants';
 import { PaginationProps, PaginationResponse } from '@fastgpt/web/common/fetch/type';
 import { ChatItemType } from '@fastgpt/global/core/chat/type';
@@ -83,6 +86,13 @@ async function handler(
      }
    });
  }
+  if (!responseDetail) {
+    histories.forEach((item) => {
+      if (item.obj === ChatRoleEnum.AI) {
+        item.value = removeAIResponseCite(item.value, false);
+      }
+    });
+  }

  return {
    list: isPlugin ? histories : transformPreviewHistories(histories, responseDetail),
--- a/projects/app/src/pages/api/core/dataset/searchTest.ts
+++ b/projects/app/src/pages/api/core/dataset/searchTest.ts
@@ -19,7 +19,7 @@ async function handler(req: ApiRequestProps<SearchTestProps>): Promise<SearchTes
  const {
    datasetId,
    text,
-    limit = 1500,
+    limit = 5000,
    similarity,
    searchMode,
    embeddingWeight,
--- a/projects/app/src/pages/api/v1/chat/completions.ts
+++ b/projects/app/src/pages/api/v1/chat/completions.ts
@@ -30,6 +30,7 @@ import {
  concatHistories,
  filterPublicNodeResponseData,
  getChatTitleFromChatMessage,
+  removeAIResponseCite,
  removeEmptyUserInput
 } from '@fastgpt/global/core/chat/utils';
 import { updateApiKeyUsage } from '@fastgpt/service/support/openapi/tools';
@@ -74,7 +75,7 @@ export type Props = ChatCompletionCreateParams &
    responseChatItemId?: string;
    stream?: boolean;
    detail?: boolean;
-    parseQuote?: boolean;
+    retainDatasetCite?: boolean;
    variables: Record<string, any>; // Global variables or plugin inputs
  };

@@ -107,7 +108,7 @@ async function handler(req: NextApiRequest, res: NextApiResponse) {

    stream = false,
    detail = false,
-    parseQuote = false,
+    retainDatasetCite = false,
    messages = [],
    variables = {},
    responseChatItemId = getNanoid(),
@@ -187,6 +188,7 @@ async function handler(req: NextApiRequest, res: NextApiResponse) {
        chatId
      });
    })();
+    retainDatasetCite = retainDatasetCite && !!responseDetail;
    const isPlugin = app.type === AppTypeEnum.plugin;

    // Check message type
@@ -291,7 +293,7 @@ async function handler(req: NextApiRequest, res: NextApiResponse) {
            chatConfig,
            histories: newHistories,
            stream,
-            parseQuote,
+            retainDatasetCite,
            maxRunTimes: WORKFLOW_MAX_RUN_TIMES,
            workflowStreamResponse: workflowResponseWrite
          });
@@ -406,17 +408,18 @@ async function handler(req: NextApiRequest, res: NextApiResponse) {

        return assistantResponses;
      })();
+      const formatResponseContent = removeAIResponseCite(responseContent, retainDatasetCite);
      const error = flowResponses[flowResponses.length - 1]?.error;

      res.json({
        ...(detail ? { responseData: feResponseData, newVariables } : {}),
        error,
-        id: chatId || '',
+        id: saveChatId,
        model: '',
        usage: { prompt_tokens: 1, completion_tokens: 1, total_tokens: 1 },
        choices: [
          {
-            message: { role: 'assistant', content: responseContent },
+            message: { role: 'assistant', content: formatResponseContent },
            finish_reason: 'stop',
            index: 0
          }
--- a/projects/app/src/pages/api/v2/chat/completions.ts
+++ b/projects/app/src/pages/api/v2/chat/completions.ts
@@ -30,6 +30,7 @@ import {
  concatHistories,
  filterPublicNodeResponseData,
  getChatTitleFromChatMessage,
+  removeAIResponseCite,
  removeEmptyUserInput
 } from '@fastgpt/global/core/chat/utils';
 import { updateApiKeyUsage } from '@fastgpt/service/support/openapi/tools';
@@ -74,7 +75,7 @@ export type Props = ChatCompletionCreateParams &
    responseChatItemId?: string;
    stream?: boolean;
    detail?: boolean;
-    parseQuote?: boolean;
+    retainDatasetCite?: boolean;
    variables: Record<string, any>; // Global variables or plugin inputs
  };

@@ -107,7 +108,7 @@ async function handler(req: NextApiRequest, res: NextApiResponse) {

    stream = false,
    detail = false,
-    parseQuote = false,
+    retainDatasetCite = false,
    messages = [],
    variables = {},
    responseChatItemId = getNanoid(),
@@ -187,6 +188,7 @@ async function handler(req: NextApiRequest, res: NextApiResponse) {
        chatId
      });
    })();
+    retainDatasetCite = retainDatasetCite && !!responseDetail;
    const isPlugin = app.type === AppTypeEnum.plugin;

    // Check message type
@@ -290,7 +292,7 @@ async function handler(req: NextApiRequest, res: NextApiResponse) {
            chatConfig,
            histories: newHistories,
            stream,
-            parseQuote,
+            retainDatasetCite,
            maxRunTimes: WORKFLOW_MAX_RUN_TIMES,
            workflowStreamResponse: workflowResponseWrite,
            version: 'v2',
@@ -401,6 +403,7 @@ async function handler(req: NextApiRequest, res: NextApiResponse) {

        return assistantResponses;
      })();
+      const formatResponseContent = removeAIResponseCite(responseContent, retainDatasetCite);
      const error = flowResponses[flowResponses.length - 1]?.error;

      res.json({
@@ -411,7 +414,7 @@ async function handler(req: NextApiRequest, res: NextApiResponse) {
        usage: { prompt_tokens: 1, completion_tokens: 1, total_tokens: 1 },
        choices: [
          {
-            message: { role: 'assistant', content: responseContent },
+            message: { role: 'assistant', content: formatResponseContent },
            finish_reason: 'stop',
            index: 0
          }
--- a/projects/app/src/pages/chat/share.tsx
+++ b/projects/app/src/pages/chat/share.tsx
@@ -87,6 +87,7 @@ const OutLink = (props: Props) => {
  const setChatBoxData = useContextSelector(ChatItemContext, (v) => v.setChatBoxData);
  const quoteData = useContextSelector(ChatItemContext, (v) => v.quoteData);
  const setQuoteData = useContextSelector(ChatItemContext, (v) => v.setQuoteData);
+  const isResponseDetail = useContextSelector(ChatItemContext, (v) => v.isResponseDetail);

  const chatRecords = useContextSelector(ChatRecordContext, (v) => v.chatRecords);
  const totalRecordsCount = useContextSelector(ChatRecordContext, (v) => v.totalRecordsCount);
@@ -162,7 +163,8 @@ const OutLink = (props: Props) => {
          },
          responseChatItemId,
          chatId: completionChatId,
-          ...outLinkAuthData
+          ...outLinkAuthData,
+          retainDatasetCite: isResponseDetail
        },
        onMessage: generatingMessage,
        abortCtrl: controller
@@ -200,6 +202,7 @@ const OutLink = (props: Props) => {
      chatId,
      customVariables,
      outLinkAuthData,
+      isResponseDetail,
      onUpdateHistoryTitle,
      setChatBoxData,
      forbidLoadChat,
--- a/projects/app/src/service/events/generateQA.ts
+++ b/projects/app/src/service/events/generateQA.ts
@@ -17,7 +17,7 @@ import {
 } from '@fastgpt/service/common/string/tiktoken/index';
 import { pushDataListToTrainingQueueByCollectionId } from '@fastgpt/service/core/dataset/training/controller';
 import { loadRequestMessages } from '@fastgpt/service/core/chat/utils';
-import { llmCompletionsBodyFormat, llmResponseToAnswerText } from '@fastgpt/service/core/ai/utils';
+import { llmCompletionsBodyFormat, formatLLMResponse } from '@fastgpt/service/core/ai/utils';
 import { LLMModelItemType } from '@fastgpt/global/core/ai/model.d';
 import {
  chunkAutoChunkSize,
@@ -140,7 +140,7 @@ ${replaceVariable(Prompt_AgentQA.fixedText, { text })}`;
        modelData
      )
    });
-    const { text: answer, usage } = await llmResponseToAnswerText(chatResponse);
+    const { text: answer, usage } = await formatLLMResponse(chatResponse);
    const inputTokens = usage?.prompt_tokens || (await countGptMessagesTokens(messages));
    const outputTokens = usage?.completion_tokens || (await countPromptTokens(answer));

--- a/projects/app/src/service/support/mcp/utils.ts
+++ b/projects/app/src/service/support/mcp/utils.ts
@@ -37,6 +37,7 @@ import { saveChat } from '@fastgpt/service/core/chat/saveChat';
 import { DispatchNodeResponseKeyEnum } from '@fastgpt/global/core/workflow/runtime/constants';
 import { createChatUsage } from '@fastgpt/service/support/wallet/usage/controller';
 import { UsageSourceEnum } from '@fastgpt/global/support/wallet/usage/constants';
+import { removeDatasetCiteText } from '@fastgpt/service/core/ai/utils';

 export const pluginNodes2InputSchema = (
  nodes: { flowNodeType: FlowNodeTypeEnum; inputs: FlowNodeInputItemType[] }[]
@@ -288,7 +289,7 @@ export const callMcpServerTool = async ({ key, toolName, inputs }: toolCallProps
    })();

    // Format response content
-    responseContent = responseContent.trim().replace(/\[\w+\]\(QUOTE\)/g, '');
+    responseContent = removeDatasetCiteText(responseContent.trim(), false);

    return responseContent;
  };
--- a/projects/app/src/web/common/api/fetch.ts
+++ b/projects/app/src/web/common/api/fetch.ts
@@ -132,7 +132,7 @@ export const streamFetch = ({
          variables,
          detail: true,
          stream: true,
-          parseQuote: true
+          retainDatasetCite: data.retainDatasetCite ?? true
        })
      };

--- a/projects/app/src/web/support/user/login/constants.ts
+++ b/projects/app/src/web/support/user/login/constants.ts
@@ -4,22 +4,3 @@ export enum LoginPageTypeEnum {
  forgetPassword = 'forgetPassword',
  wechat = 'wechat'
 }
-
-export const checkPasswordRule = (password: string) => {
-  const patterns = [
-    /\d/, // Contains digits
-    /[a-z]/, // Contains lowercase letters
-    /[A-Z]/, // Contains uppercase letters
-    /[!@#$%^&*()_+=-]/ // Contains special characters
-  ];
-  const validChars = /^[\dA-Za-z!@#$%^&*()_+=-]{6,100}$/;
-
-  // Check length and valid characters
-  if (!validChars.test(password)) return false;
-
-  // Count how many patterns are satisfied
-  const matchCount = patterns.filter((pattern) => pattern.test(password)).length;
-
-  // Must satisfy at least 2 patterns
-  return matchCount >= 2;
-};
--- a/test/cases/components/Markdown/utils.test.ts
+++ b/test/cases/components/Markdown/utils.test.ts
@@ -16,7 +16,7 @@ describe('Markdown utils', () => {

    it('should convert quote references to proper markdown links', () => {
      const input = '[123456789012345678901234]';
-      const expected = '[123456789012345678901234](QUOTE)';
+      const expected = '[123456789012345678901234](CITE)';
      expect(mdTextFormat(input)).toBe(expected);
    });

@@ -35,7 +35,7 @@ describe('Markdown utils', () => {
      const input =
        'Math \\[x^2\\] with link https://test.com，and quote [123456789012345678901234]';
      const expected =
-        'Math $$x^2$$ with link https://test.com ，and quote [123456789012345678901234](QUOTE)';
+        'Math $$x^2$$ with link https://test.com ，and quote [123456789012345678901234](CITE)';
      expect(mdTextFormat(input)).toBe(expected);
    });
  });
--- a/test/cases/function/packages/global/common/string/chunks.json
+++ b/test/cases/function/packages/global/common/string/chunks.json
@@ -1,5 +0,0 @@
-[
-  "测试的呀,第一个表格\n\n| 序号 | 姓名 | 年龄 | 职业 | 城市 |\n| --- | --- | --- | --- | --- |\n| 1 | 张三 | 25 | 工程师 | 北京 |\n| 2 | 李四 | 30 | 教师 | 上海 |\n| 3 | 王五 | 28 | 医生 | 广州 |\n| 6 | 周八 | 32 | 会计 | 成都 |\n| 4 | 赵六 | 35 | 律师 | 深圳 |\n| 5 | 孙七 | 27 | 设计师 | 杭州 |\n| 6 | 周八 | 32 | 会计 | 成都 |\n| 6 | 周八 | 32 | 会计 | 成都 |\n| 7 | 吴九 | 29 | 销售 | 武汉 |\n| 8 | 郑十 | 31 | 记者 | 南京 |\n| 9 | 刘一 | 33 | 建筑师 | 天津 |\n| 10 | 陈二 | 26 | 程序员 | 重庆 |\n| 1000 | 黄末 | 28 | 作家 | 厦门 |\n| 1001 | 杨一 | 34 | 程序员 | 厦门 |\n| 1002 | 杨二 | 34 | 程序员 | 厦门 |\n| 1003 | 杨三 | 34 | 程序员 | 厦门 |",
-  "| 序号 | 姓名 | 年龄 | 职业 | 城市 |\n| --- | --- | --- | --- | --- |\n| 6 | 周八 | 32 | 会计 | 成都 |\n| 1004 | 杨四 | 34 | 程序员 | 厦门 |\n| 1005 | 杨五 | 34 | 程序员 | 厦门 |\n| 1000 | 黄末 | 28 | 作家 | 厦门 |\n| 1000 | 黄末 | 28 | 作家 | 厦门 |\n| 1000 | 黄末 | 28 | 作家 | 厦门 |\n| 9 | 刘一 | 33 | 建筑师 | 天津 |\n| 10 | 陈二 | 26 | 程序员 | 重庆 |\n| 1000 | 黄末 | 28 | 作家 | 厦门 |\n| 1001 | 杨一 | 34 | 程序员 | 厦门 |\n| 1002 | 杨二 | 34 | 程序员 | 厦门 |\n| 1003 | 杨三 | 34 | 程序员 | 厦门 |\n| 1004 | 杨四 | 34 | 程序员 | 厦门 |\n| 1005 | 杨五 | 34 | 程序员 | 厦门 |\n\n| 序号 | 姓名 | 年龄 | 职业 | 城市 |\n| --- | --- | --- | --- | --- |\n| 6 | 周八 | 32 | 会计 | 成都 |\n| 1000 | 黄末 | 28 | 作家 | 厦门 |\n| 1000 | 黄末 | 28 | 作家 | 厦门 |\n| 1000 | 黄末 | 28 | 作家 | 厦门 |",
-  "这是第二段了，第二表格\n\n| 序号 | 姓名 | 年龄 | 职业 | 城市 |\n| --- | --- | --- | --- | --- |\n| 1 | 张三 | 25 | 工程师 | 北京 |\n| 6 | 周八 | 32 | 会计 | 成都 |\n| 2 | 李四 | 30 | 教师 | 上海 |\n| 3 | 王五 | 28 | 医生 | 广州 |\n| 4 | 赵六 | 35 | 律师 | 深圳 |\n| 5 | 孙七 | 27 | 设计师 | 杭州 |\n| 6 | 周八 | 32 | 会计 | 成都 |\n| 7 | 吴九 | 29 | 销售 | 武汉 |\n| 8 | 郑十 | 31 | 记者 | 南京 |\n| 9 | 刘一 | 33 | 建筑师 | 天津 |\n| 10 | 陈二 | 26 | 程序员 | 重庆 |\n| 10004 | 黄末 | 28 | 作家 | 厦门 |\n| 10013 | 杨一 | 34 | 程序员 | 厦门 |\n\n\n结束了\n\n| 序号22 | 姓名 | 年龄 | 职业 | 城市 |\n| --- | --- | --- | --- | --- |\n| 1 | 张三 | 25 | 工程师 | 北京 |\n| 2 | 李四 | 30 | 教师 | 上海 |\n| 3 | 王五 | 28 | 医生 | 广州 |\n| 4 | 赵六 | 35 | 律师 | 深圳 |\n| 5 | 孙七 | 27 | 设计师 | 杭州 |\n| 6 | 周八 | 32 | 会计 | 成都 |\n| 6 | 周八 | 32 | 会计 | 成都 |\n| 7 | 吴九 | 29 | 销售 | 武汉 |\n| 8 | 郑十 | 31 | 记者 | 南京 |\n| 9 | 刘一 | 33 | 建筑师 | 天津 |\n| 10 | 陈二 | 26 | 程序员 | 重庆 |\n| 10002 | 黄末 | 28 | 作家 | 厦门 |\n| 10012 | 杨一 | 34 | 程序员 | 厦门 |"
-]
--- a/test/cases/function/packages/global/common/string/password.test.ts
+++ b/test/cases/function/packages/global/common/string/password.test.ts
@@ -1,5 +1,5 @@
 import { describe, expect, it } from 'vitest';
-import { checkPasswordRule } from '@/web/support/user/login/constants';
+import { checkPasswordRule } from '@fastgpt/global/common/string/password';

 describe('PasswordRule', () => {
  it('should be a valid password', () => {
--- a/test/cases/function/packages/global/common/string/textSplitter.test.ts
+++ b/test/cases/function/packages/global/common/string/textSplitter.test.ts
@@ -1,6 +1,5 @@
 import { it, expect } from 'vitest'; // 必须显式导入
 import { splitText2Chunks } from '@fastgpt/global/common/string/textSplitter';
-import * as fs from 'fs';

 const simpleChunks = (chunks: string[]) => {
  return chunks.map((chunk) => chunk.replace(/\s+/g, ''));
--- a/test/cases/function/packages/service/core/ai/parseStreamResponse.test.ts
+++ b/test/cases/function/packages/service/core/ai/parseStreamResponse.test.ts
@@ -0,0 +1,340 @@
+import { CompletionFinishReason } from '@fastgpt/global/core/ai/type';
+import { parseLLMStreamResponse } from '@fastgpt/service/core/ai/utils';
+import { describe, expect, it } from 'vitest';
+
+describe('Parse reasoning stream content test', async () => {
+  const partList = [
+    {
+      data: [{ content: '你好1' }, { content: '你好2' }, { content: '你好3' }],
+      correct: { answer: '你好1你好2你好3', reasoning: '' }
+    },
+    {
+      data: [
+        { reasoning_content: '这是' },
+        { reasoning_content: '思考' },
+        { reasoning_content: '过程' },
+        { content: '你好1' },
+        { content: '你好2' },
+        { content: '你好3' }
+      ],
+      correct: { answer: '你好1你好2你好3', reasoning: '这是思考过程' }
+    },
+    {
+      data: [
+        { content: '<t' },
+        { content: 'hink>' },
+        { content: '这是' },
+        { content: '思考' },
+        { content: '过程' },
+        { content: '</think>' },
+        { content: '你好1' },
+        { content: '你好2' },
+        { content: '你好3' }
+      ],
+      correct: { answer: '你好1你好2你好3', reasoning: '这是思考过程' }
+    },
+    {
+      data: [
+        { content: '<think>' },
+        { content: '这是' },
+        { content: '思考' },
+        { content: '过程' },
+        { content: '</think>' },
+        { content: '你好1' },
+        { content: '你好2' },
+        { content: '你好3' }
+      ],
+      correct: { answer: '你好1你好2你好3', reasoning: '这是思考过程' }
+    },
+    {
+      data: [
+        { content: '<think>这是' },
+        { content: '思考' },
+        { content: '过程' },
+        { content: '</think>' },
+        { content: '你好1' },
+        { content: '你好2' },
+        { content: '你好3' }
+      ],
+      correct: { answer: '你好1你好2你好3', reasoning: '这是思考过程' }
+    },
+    {
+      data: [
+        { content: '<think>这是' },
+        { content: '思考' },
+        { content: '过程</' },
+        { content: 'think>' },
+        { content: '你好1' },
+        { content: '你好2' },
+        { content: '你好3' }
+      ],
+      correct: { answer: '你好1你好2你好3', reasoning: '这是思考过程' }
+    },
+    {
+      data: [
+        { content: '<think>这是' },
+        { content: '思考' },
+        { content: '过程</think>' },
+        { content: '你好1' },
+        { content: '你好2' },
+        { content: '你好3' }
+      ],
+      correct: { answer: '你好1你好2你好3', reasoning: '这是思考过程' }
+    },
+    {
+      data: [
+        { content: '<think>这是' },
+        { content: '思考' },
+        { content: '过程</think>你好1' },
+        { content: '你好2' },
+        { content: '你好3' }
+      ],
+      correct: { answer: '你好1你好2你好3', reasoning: '这是思考过程' }
+    },
+    {
+      data: [
+        { content: '<think>这是' },
+        { content: '思考' },
+        { content: '过程</th' },
+        { content: '假的' },
+        { content: '你好2' },
+        { content: '你好3' },
+        { content: '过程</think>你好1' },
+        { content: '你好2' },
+        { content: '你好3' }
+      ],
+      correct: { answer: '你好1你好2你好3', reasoning: '这是思考过程</th假的你好2你好3过程' }
+    },
+    {
+      data: [
+        { content: '<think>这是' },
+        { content: '思考' },
+        { content: '过程</th' },
+        { content: '假的' },
+        { content: '你好2' },
+        { content: '你好3' }
+      ],
+      correct: { answer: '', reasoning: '这是思考过程</th假的你好2你好3' }
+    }
+  ];
+
+  // Remove think
+  partList.forEach((part, index) => {
+    it(`Reasoning test:${index}`, () => {
+      const { parsePart } = parseLLMStreamResponse();
+
+      let answer = '';
+      let reasoning = '';
+      part.data.forEach((item) => {
+        const formatPart = {
+          choices: [
+            {
+              delta: {
+                role: 'assistant',
+                content: item.content,
+                reasoning_content: item.reasoning_content
+              }
+            }
+          ]
+        };
+        const { reasoningContent, content } = parsePart({
+          part: formatPart,
+          parseThinkTag: true,
+          retainDatasetCite: false
+        });
+        answer += content;
+        reasoning += reasoningContent;
+      });
+      expect(answer).toBe(part.correct.answer);
+      expect(reasoning).toBe(part.correct.reasoning);
+    });
+  });
+});
+
+describe('Parse dataset cite content test', async () => {
+  const partList = [
+    {
+      // 完整的
+      data: [
+        { content: '知识库' },
+        { content: '问答系统' },
+        { content: '[67e517e747' },
+        { content: '67063e882d' },
+        { content: '6861](CITE)' }
+      ],
+      correct: {
+        content: '知识库问答系统[67e517e74767063e882d6861](CITE)',
+        responseContent: '知识库问答系统'
+      }
+    },
+    {
+      // 缺失结尾
+      data: [
+        { content: '知识库问答系统' },
+        { content: '[67e517e747' },
+        { content: '67063e882d' },
+        { content: '6861](CITE' }
+      ],
+      correct: {
+        content: '知识库问答系统[67e517e74767063e882d6861](CITE',
+        responseContent: '知识库问答系统[67e517e74767063e882d6861](CITE'
+      }
+    },
+    {
+      // ObjectId 不正确
+      data: [
+        { content: '知识库问答系统' },
+        { content: '[67e517e747' },
+        { content: '67882d' },
+        { content: '6861](CITE)' }
+      ],
+      correct: {
+        content: '知识库问答系统[67e517e74767882d6861](CITE)',
+        responseContent: '知识库问答系统[67e517e74767882d6861](CITE)'
+      }
+    },
+    {
+      // 其他链接
+      data: [{ content: '知识库' }, { content: '问答系统' }, { content: '[](https://fastgpt.cn)' }],
+      correct: {
+        content: '知识库问答系统[](https://fastgpt.cn)',
+        responseContent: '知识库问答系统[](https://fastgpt.cn)'
+      }
+    },
+    {
+      // 不完整的其他链接
+      data: [{ content: '知识库' }, { content: '问答系统' }, { content: '[](https://fastgp' }],
+      correct: {
+        content: '知识库问答系统[](https://fastgp',
+        responseContent: '知识库问答系统[](https://fastgp'
+      }
+    },
+    {
+      // 开头
+      data: [{ content: '[知识库' }, { content: '问答系统' }, { content: '[](https://fastgp' }],
+      correct: {
+        content: '[知识库问答系统[](https://fastgp',
+        responseContent: '[知识库问答系统[](https://fastgp'
+      }
+    },
+    {
+      // 结尾
+      data: [{ content: '知识库' }, { content: '问答系统' }, { content: '[' }],
+      correct: {
+        content: '知识库问答系统[',
+        responseContent: '知识库问答系统['
+      }
+    },
+    {
+      // 中间
+      data: [
+        { content: '知识库' },
+        { content: '问答系统' },
+        { content: '[' },
+        { content: '问答系统]' }
+      ],
+      correct: {
+        content: '知识库问答系统[问答系统]',
+        responseContent: '知识库问答系统[问答系统]'
+      }
+    },
+    {
+      // 双链接
+      data: [
+        { content: '知识库' },
+        { content: '问答系统' },
+        { content: '[](https://fastgpt.cn)' },
+        { content: '[67e517e747' },
+        { content: '67063e882d' },
+        { content: '6861](CITE)' }
+      ],
+      correct: {
+        content: '知识库问答系统[](https://fastgpt.cn)[67e517e74767063e882d6861](CITE)',
+        responseContent: '知识库问答系统[](https://fastgpt.cn)'
+      }
+    },
+    {
+      // 双链接缺失部分
+      data: [
+        { content: '知识库' },
+        { content: '问答系统' },
+        { content: '[](https://fastgpt.cn)' },
+        { content: '[67e517e747' },
+        { content: '67063e882d' },
+        { content: '6861](CIT' }
+      ],
+      correct: {
+        content: '知识库问答系统[](https://fastgpt.cn)[67e517e74767063e882d6861](CIT',
+        responseContent: '知识库问答系统[](https://fastgpt.cn)[67e517e74767063e882d6861](CIT'
+      }
+    },
+    {
+      // 双Cite
+      data: [
+        { content: '知识库' },
+        { content: '问答系统' },
+        { content: '[67e517e747' },
+        { content: '67063e882d' },
+        { content: '6861](CITE)' },
+        { content: '[67e517e747' },
+        { content: '67063e882d' },
+        { content: '6861](CITE)' }
+      ],
+      correct: {
+        content: '知识库问答系统[67e517e74767063e882d6861](CITE)[67e517e74767063e882d6861](CITE)',
+        responseContent: '知识库问答系统'
+      }
+    },
+    {
+      // 双Cite-第一个假Cite
+      data: [
+        { content: '知识库' },
+        { content: '问答系统' },
+        { content: '[67e517e747' },
+        { content: '6861](CITE)' },
+        { content: '[67e517e747' },
+        { content: '67063e882d' },
+        { content: '6861](CITE)' }
+      ],
+      correct: {
+        content: '知识库问答系统[67e517e7476861](CITE)[67e517e74767063e882d6861](CITE)',
+        responseContent: '知识库问答系统[67e517e7476861](CITE)'
+      }
+    }
+  ];
+
+  partList.forEach((part, index) => {
+    it(`Dataset cite test: ${index}`, () => {
+      const { parsePart } = parseLLMStreamResponse();
+
+      let answer = '';
+      let responseContent = '';
+      part.data.forEach((item, index) => {
+        const formatPart = {
+          choices: [
+            {
+              delta: {
+                role: 'assistant',
+                content: item.content,
+                reasoning_content: ''
+              },
+              finish_reason: (index === part.data.length - 1
+                ? 'stop'
+                : null) as CompletionFinishReason
+            }
+          ]
+        };
+        const { content, responseContent: newResponseContent } = parsePart({
+          part: formatPart,
+          parseThinkTag: false,
+          retainDatasetCite: false
+        });
+        answer += content;
+        responseContent += newResponseContent;
+      });
+
+      expect(answer).toEqual(part.correct.content);
+      expect(responseContent).toEqual(part.correct.responseContent);
+    });
+  });
+});