File input (#2270)

* doc * feat: file upload config * perf: chat box file params * feat: markdown show file * feat: chat file store and clear * perf: read file contentType * feat: llm vision config * feat: file url output * perf: plugin error text * perf: image load * feat: ai chat document * perf: file block ui * feat: read file node * feat: file read response field * feat: simple mode support read files * feat: tool call * feat: read file histories * perf: select file * perf: select file config * i18n * i18n * fix: ts; feat: tool response preview result
2025-10-21 11:30:06 +00:00 · 2024-08-06 10:00:22 +08:00
parent 10dcdb5491
commit e36d9d794f
121 changed files with 2600 additions and 1142 deletions
--- a/packages/service/core/workflow/dispatch/agent/classifyQuestion.ts
+++ b/packages/service/core/workflow/dispatch/agent/classifyQuestion.ts
@@ -16,6 +16,7 @@ import { formatModelChars2Points } from '../../../../support/wallet/usage/utils'
 import { DispatchNodeResultType } from '@fastgpt/global/core/workflow/runtime/type';
 import { chatValue2RuntimePrompt } from '@fastgpt/global/core/chat/adapt';
 import { getHandleId } from '@fastgpt/global/core/workflow/utils';
+import { loadRequestMessages } from '../../../chat/utils';

 type Props = ModuleDispatchProps<{
  [NodeInputKeyEnum.aiModel]: string;
@@ -113,6 +114,10 @@ const completions = async ({
      ]
    }
  ];
+  const requestMessages = await loadRequestMessages({
+    messages: chats2GPTMessages({ messages, reserveId: false }),
+    useVision: false
+  });

  const ai = getAIApi({
    userKey: user.openaiAccount,
@@ -122,7 +127,7 @@ const completions = async ({
  const data = await ai.chat.completions.create({
    model: cqModel.model,
    temperature: 0.01,
-    messages: chats2GPTMessages({ messages, reserveId: false }),
+    messages: requestMessages,
    stream: false
  });
  const answer = data.choices?.[0].message?.content || '';
--- a/packages/service/core/workflow/dispatch/agent/extract.ts
+++ b/packages/service/core/workflow/dispatch/agent/extract.ts
@@ -1,5 +1,5 @@
 import { chats2GPTMessages } from '@fastgpt/global/core/chat/adapt';
-import { filterGPTMessageByMaxTokens } from '../../../chat/utils';
+import { filterGPTMessageByMaxTokens, loadRequestMessages } from '../../../chat/utils';
 import type { ChatItemType } from '@fastgpt/global/core/chat/type.d';
 import {
  countMessagesTokens,
@@ -173,6 +173,10 @@ ${description ? `- ${description}` : ''}
    messages: adaptMessages,
    maxTokens: extractModel.maxContext
  });
+  const requestMessages = await loadRequestMessages({
+    messages: filterMessages,
+    useVision: false
+  });

  const properties: Record<
    string,
@@ -200,7 +204,7 @@ ${description ? `- ${description}` : ''}
  };

  return {
-    filterMessages,
+    filterMessages: requestMessages,
    agentFunction
  };
 };
@@ -338,6 +342,10 @@ Human: ${content}`
      ]
    }
  ];
+  const requestMessages = await loadRequestMessages({
+    messages: chats2GPTMessages({ messages, reserveId: false }),
+    useVision: false
+  });

  const ai = getAIApi({
    userKey: user.openaiAccount,
@@ -346,7 +354,7 @@ Human: ${content}`
  const data = await ai.chat.completions.create({
    model: extractModel.model,
    temperature: 0.01,
-    messages: chats2GPTMessages({ messages, reserveId: false }),
+    messages: requestMessages,
    stream: false
  });
  const answer = data.choices?.[0].message?.content || '';
--- a/packages/service/core/workflow/dispatch/agent/runTool/constants.ts
+++ b/packages/service/core/workflow/dispatch/agent/runTool/constants.ts
@@ -1,3 +1,5 @@
+import { replaceVariable } from '@fastgpt/global/common/string/tools';
+
 export const Prompt_Tool_Call = `<Instruction>
 你是一个智能机器人，除了可以回答用户问题外，你还掌握工具的使用能力。有时候，你可以依赖工具的运行结果，来更准确的回答用户。

@@ -32,6 +34,8 @@ TOOL_RESPONSE: """
 ANSWER: 0: 今天杭州是晴天，适合去西湖、灵隐寺、千岛湖等地玩。
 </Instruction>

+------
+
 现在，我们开始吧！下面是你本次可以使用的工具：

 """
@@ -42,3 +46,16 @@ ANSWER: 0: 今天杭州是晴天，适合去西湖、灵隐寺、千岛湖等地

 USER: {{question}}
 ANSWER: `;
+
+export const getMultiplePrompt = (obj: {
+  fileCount: number;
+  imgCount: number;
+  question: string;
+}) => {
+  const prompt = `Number of session file inputs：
+Document：{{fileCount}}
+Image：{{imgCount}}
+------
+{{question}}`;
+  return replaceVariable(prompt, obj);
+};
--- a/packages/service/core/workflow/dispatch/agent/runTool/functionCall.ts
+++ b/packages/service/core/workflow/dispatch/agent/runTool/functionCall.ts
@@ -9,7 +9,7 @@ import {
  ChatCompletionMessageFunctionCall,
  ChatCompletionFunctionMessageParam,
  ChatCompletionAssistantMessageParam
-} from '@fastgpt/global/core/ai/type';
+} from '@fastgpt/global/core/ai/type.d';
 import { NextApiResponse } from 'next';
 import {
  responseWrite,
@@ -24,10 +24,11 @@ import { DispatchToolModuleProps, RunToolResponse, ToolNodeItemType } from './ty
 import json5 from 'json5';
 import { DispatchFlowResponse } from '../../type';
 import { countGptMessagesTokens } from '../../../../../common/string/tiktoken/index';
-import { getNanoid } from '@fastgpt/global/common/string/tools';
+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 { updateToolInputValue } from './utils';
+import { computedMaxToken, computedTemperature } from '../../../../ai/utils';

 type FunctionRunResponseType = {
  toolRunResponse: DispatchFlowResponse;
@@ -42,7 +43,18 @@ export const runToolWithFunctionCall = async (
  },
  response?: RunToolResponse
 ): Promise<RunToolResponse> => {
-  const { toolModel, toolNodes, messages, res, runtimeNodes, detail = false, node, stream } = props;
+  const {
+    toolModel,
+    toolNodes,
+    messages,
+    res,
+    requestOrigin,
+    runtimeNodes,
+    detail = false,
+    node,
+    stream,
+    params: { temperature = 0, maxToken = 4000, aiChatVision }
+  } = props;
  const assistantResponses = response?.assistantResponses || [];

  const functions: ChatCompletionCreateParams.Function[] = toolNodes.map((item) => {
@@ -72,44 +84,60 @@ export const runToolWithFunctionCall = async (
    };
  });

-  const filterMessages = await filterGPTMessageByMaxTokens({
-    messages,
-    maxTokens: toolModel.maxContext - 500 // filter token. not response maxToken
-  });
-  const formativeMessages = filterMessages.map((item) => {
+  const filterMessages = (
+    await filterGPTMessageByMaxTokens({
+      messages,
+      maxTokens: toolModel.maxContext - 300 // filter token. not response maxToken
+    })
+  ).map((item) => {
    if (item.role === ChatCompletionRequestMessageRoleEnum.Assistant && item.function_call) {
      return {
        ...item,
        function_call: {
          name: item.function_call?.name,
          arguments: item.function_call?.arguments
-        }
+        },
+        content: ''
      };
    }
    return item;
  });
-  const requestMessages = await loadRequestMessages(formativeMessages);
+  const [requestMessages, max_tokens] = await Promise.all([
+    loadRequestMessages({
+      messages: filterMessages,
+      useVision: toolModel.vision && aiChatVision,
+      origin: requestOrigin
+    }),
+    computedMaxToken({
+      model: toolModel,
+      maxToken,
+      filterMessages
+    })
+  ]);
+  const requestBody: any = {
+    ...toolModel?.defaultConfig,
+    model: toolModel.model,
+    temperature: computedTemperature({
+      model: toolModel,
+      temperature
+    }),
+    max_tokens,
+    stream,
+    messages: requestMessages,
+    functions,
+    function_call: 'auto'
+  };

+  // console.log(JSON.stringify(requestBody, null, 2));
  /* Run llm */
  const ai = getAIApi({
    timeout: 480000
  });
-  const aiResponse = await ai.chat.completions.create(
-    {
-      ...toolModel?.defaultConfig,
-      model: toolModel.model,
-      temperature: 0,
-      stream,
-      messages: requestMessages,
-      functions,
-      function_call: 'auto'
-    },
-    {
-      headers: {
-        Accept: 'application/json, text/plain, */*'
-      }
+  const aiResponse = await ai.chat.completions.create(requestBody, {
+    headers: {
+      Accept: 'application/json, text/plain, */*'
    }
-  );
+  });

  const { answer, functionCalls } = await (async () => {
    if (res && stream) {
@@ -198,7 +226,7 @@ export const runToolWithFunctionCall = async (
                toolName: '',
                toolAvatar: '',
                params: '',
-                response: stringToolResponse
+                response: sliceStrStartEnd(stringToolResponse, 300, 300)
              }
            })
          });
@@ -222,7 +250,7 @@ export const runToolWithFunctionCall = async (
      function_call: functionCall
    };
    const concatToolMessages = [
-      ...filterMessages,
+      ...requestMessages,
      assistantToolMsgParams
    ] as ChatCompletionMessageParam[];
    const tokens = await countGptMessagesTokens(concatToolMessages, undefined, functions);
--- a/packages/service/core/workflow/dispatch/agent/runTool/index.ts
+++ b/packages/service/core/workflow/dispatch/agent/runTool/index.ts
@@ -8,7 +8,7 @@ import { ModelTypeEnum, getLLMModel } from '../../../../ai/model';
 import { filterToolNodeIdByEdges, getHistories } from '../../utils';
 import { runToolWithToolChoice } from './toolChoice';
 import { DispatchToolModuleProps, ToolNodeItemType } from './type.d';
-import { ChatItemType } from '@fastgpt/global/core/chat/type';
+import { ChatItemType, UserChatItemValueItemType } from '@fastgpt/global/core/chat/type';
 import { ChatRoleEnum } from '@fastgpt/global/core/chat/constants';
 import {
  GPTMessages2Chats,
@@ -22,12 +22,46 @@ import { getHistoryPreview } from '@fastgpt/global/core/chat/utils';
 import { runToolWithFunctionCall } from './functionCall';
 import { runToolWithPromptCall } from './promptCall';
 import { replaceVariable } from '@fastgpt/global/common/string/tools';
-import { Prompt_Tool_Call } from './constants';
+import { getMultiplePrompt, Prompt_Tool_Call } from './constants';
+import { filterToolResponseToPreview } from './utils';

 type Response = DispatchNodeResultType<{
  [NodeOutputKeyEnum.answerText]: string;
 }>;

+/* 
+  Tool call， auth add file prompt to question。
+  Guide the LLM to call tool.
+*/
+export const toolCallMessagesAdapt = ({
+  userInput
+}: {
+  userInput: UserChatItemValueItemType[];
+}) => {
+  const files = userInput.filter((item) => item.type === 'file');
+
+  if (files.length > 0) {
+    return userInput.map((item) => {
+      if (item.type === 'text') {
+        const filesCount = files.filter((file) => file.file?.type === 'file').length;
+        const imgCount = files.filter((file) => file.file?.type === 'image').length;
+        const text = item.text?.content || '';
+
+        return {
+          ...item,
+          text: {
+            content: getMultiplePrompt({ fileCount: filesCount, imgCount, question: text })
+          }
+        };
+      }
+
+      return item;
+    });
+  }
+
+  return userInput;
+};
+
 export const dispatchRunTools = async (props: DispatchToolModuleProps): Promise<Response> => {
  const {
    node: { nodeId, name },
@@ -62,16 +96,31 @@ export const dispatchRunTools = async (props: DispatchToolModuleProps): Promise<

  const messages: ChatItemType[] = [
    ...getSystemPrompt(systemPrompt),
-    ...chatHistories,
+    // Add file input prompt to histories
+    ...chatHistories.map((item) => {
+      if (item.obj === ChatRoleEnum.Human) {
+        return {
+          ...item,
+          value: toolCallMessagesAdapt({
+            userInput: item.value
+          })
+        };
+      }
+      return item;
+    }),
    {
      obj: ChatRoleEnum.Human,
-      value: runtimePrompt2ChatsValue({
-        text: userChatInput,
-        files: chatValue2RuntimePrompt(query).files
+      value: toolCallMessagesAdapt({
+        userInput: runtimePrompt2ChatsValue({
+          text: userChatInput,
+          files: chatValue2RuntimePrompt(query).files
+        })
      })
    }
  ];

+  // console.log(JSON.stringify(messages, null, 2));
+
  const {
    dispatchFlowResponse, // tool flow response
    totalTokens,
@@ -98,14 +147,24 @@ export const dispatchRunTools = async (props: DispatchToolModuleProps): Promise<
    }

    const lastMessage = adaptMessages[adaptMessages.length - 1];
-    if (typeof lastMessage.content !== 'string') {
-      return Promise.reject('暂时只支持纯文本');
+    if (typeof lastMessage.content === 'string') {
+      lastMessage.content = replaceVariable(Prompt_Tool_Call, {
+        question: lastMessage.content
+      });
+    } else if (Array.isArray(lastMessage.content)) {
+      // array, replace last element
+      const lastText = lastMessage.content[lastMessage.content.length - 1];
+      if (lastText.type === 'text') {
+        lastMessage.content = replaceVariable(Prompt_Tool_Call, {
+          question: lastText.text
+        });
+      } else {
+        return Promise.reject('Prompt call invalid input');
+      }
+    } else {
+      return Promise.reject('Prompt call invalid input');
    }

-    lastMessage.content = replaceVariable(Prompt_Tool_Call, {
-      question: userChatInput
-    });
-
    return runToolWithPromptCall({
      ...props,
      toolNodes,
@@ -132,12 +191,14 @@ export const dispatchRunTools = async (props: DispatchToolModuleProps): Promise<
    }, 0);
  const flatUsages = dispatchFlowResponse.map((item) => item.flowUsages).flat();

+  const previewAssistantResponses = filterToolResponseToPreview(assistantResponses);
+
  return {
-    [NodeOutputKeyEnum.answerText]: assistantResponses
+    [NodeOutputKeyEnum.answerText]: previewAssistantResponses
      .filter((item) => item.text?.content)
      .map((item) => item.text?.content || '')
      .join(''),
-    [DispatchNodeResponseKeyEnum.assistantResponses]: assistantResponses,
+    [DispatchNodeResponseKeyEnum.assistantResponses]: previewAssistantResponses,
    [DispatchNodeResponseKeyEnum.nodeResponse]: {
      totalPoints: totalPointsUsage,
      toolCallTokens: totalTokens,
--- a/packages/service/core/workflow/dispatch/agent/runTool/promptCall.ts
+++ b/packages/service/core/workflow/dispatch/agent/runTool/promptCall.ts
@@ -20,10 +20,16 @@ import { dispatchWorkFlow } from '../../index';
 import { DispatchToolModuleProps, RunToolResponse, ToolNodeItemType } from './type.d';
 import json5 from 'json5';
 import { countGptMessagesTokens } from '../../../../../common/string/tiktoken/index';
-import { getNanoid, replaceVariable, sliceJsonStr } from '@fastgpt/global/common/string/tools';
+import {
+  getNanoid,
+  replaceVariable,
+  sliceJsonStr,
+  sliceStrStartEnd
+} from '@fastgpt/global/common/string/tools';
 import { AIChatItemType } from '@fastgpt/global/core/chat/type';
 import { GPTMessages2Chats } from '@fastgpt/global/core/chat/adapt';
 import { updateToolInputValue } from './utils';
+import { computedMaxToken, computedTemperature } from '../../../../ai/utils';

 type FunctionCallCompletion = {
  id: string;
@@ -43,7 +49,18 @@ export const runToolWithPromptCall = async (
  },
  response?: RunToolResponse
 ): Promise<RunToolResponse> => {
-  const { toolModel, toolNodes, messages, res, runtimeNodes, detail = false, node, stream } = props;
+  const {
+    toolModel,
+    toolNodes,
+    messages,
+    res,
+    requestOrigin,
+    runtimeNodes,
+    detail = false,
+    node,
+    stream,
+    params: { temperature = 0, maxToken = 4000, aiChatVision }
+  } = props;
  const assistantResponses = response?.assistantResponses || [];

  const toolsPrompt = JSON.stringify(
@@ -77,7 +94,7 @@ export const runToolWithPromptCall = async (

  const lastMessage = messages[messages.length - 1];
  if (typeof lastMessage.content !== 'string') {
-    return Promise.reject('暂时只支持纯文本');
+    return Promise.reject('Prompt call invalid input');
  }
  lastMessage.content = replaceVariable(lastMessage.content, {
    toolsPrompt
@@ -87,27 +104,40 @@ export const runToolWithPromptCall = async (
    messages,
    maxTokens: toolModel.maxContext - 500 // filter token. not response maxToken
  });
-  const requestMessages = await loadRequestMessages(filterMessages);
+  const [requestMessages, max_tokens] = await Promise.all([
+    loadRequestMessages({
+      messages: filterMessages,
+      useVision: toolModel.vision && aiChatVision,
+      origin: requestOrigin
+    }),
+    computedMaxToken({
+      model: toolModel,
+      maxToken,
+      filterMessages
+    })
+  ]);
+  const requestBody = {
+    ...toolModel?.defaultConfig,
+    model: toolModel.model,
+    temperature: computedTemperature({
+      model: toolModel,
+      temperature
+    }),
+    max_tokens,
+    stream,
+    messages: requestMessages
+  };

-  // console.log(JSON.stringify(filterMessages, null, 2));
+  // console.log(JSON.stringify(requestBody, null, 2));
  /* Run llm */
  const ai = getAIApi({
    timeout: 480000
  });
-  const aiResponse = await ai.chat.completions.create(
-    {
-      ...toolModel?.defaultConfig,
-      model: toolModel.model,
-      temperature: 0,
-      stream,
-      messages: requestMessages
-    },
-    {
-      headers: {
-        Accept: 'application/json, text/plain, */*'
-      }
+  const aiResponse = await ai.chat.completions.create(requestBody, {
+    headers: {
+      Accept: 'application/json, text/plain, */*'
    }
-  );
+  });

  const answer = await (async () => {
    if (res && stream) {
@@ -225,7 +255,7 @@ export const runToolWithPromptCall = async (
            toolName: '',
            toolAvatar: '',
            params: '',
-            response: stringToolResponse
+            response: sliceStrStartEnd(stringToolResponse, 300, 300)
          }
        })
      });
@@ -250,7 +280,7 @@ export const runToolWithPromptCall = async (
    function_call: toolJson
  };
  const concatToolMessages = [
-    ...filterMessages,
+    ...requestMessages,
    assistantToolMsgParams
  ] as ChatCompletionMessageParam[];
  const tokens = await countGptMessagesTokens(concatToolMessages, undefined);
--- a/packages/service/core/workflow/dispatch/agent/runTool/toolChoice.ts
+++ b/packages/service/core/workflow/dispatch/agent/runTool/toolChoice.ts
@@ -28,6 +28,8 @@ import { countGptMessagesTokens } from '../../../../../common/string/tiktoken/in
 import { GPTMessages2Chats } from '@fastgpt/global/core/chat/adapt';
 import { AIChatItemType } from '@fastgpt/global/core/chat/type';
 import { updateToolInputValue } from './utils';
+import { computedMaxToken, computedTemperature } from '../../../../ai/utils';
+import { sliceStrStartEnd } from '@fastgpt/global/common/string/tools';

 type ToolRunResponseType = {
  toolRunResponse: DispatchFlowResponse;
@@ -49,7 +51,18 @@ export const runToolWithToolChoice = async (
  },
  response?: RunToolResponse
 ): Promise<RunToolResponse> => {
-  const { toolModel, toolNodes, messages, res, runtimeNodes, detail = false, node, stream } = props;
+  const {
+    toolModel,
+    toolNodes,
+    messages,
+    res,
+    requestOrigin,
+    runtimeNodes,
+    detail = false,
+    node,
+    stream,
+    params: { temperature = 0, maxToken = 4000, aiChatVision }
+  } = props;
  const assistantResponses = response?.assistantResponses || [];

  const tools: ChatCompletionTool[] = toolNodes.map((item) => {
@@ -81,12 +94,13 @@ export const runToolWithToolChoice = async (
      }
    };
  });
-
-  const filterMessages = await filterGPTMessageByMaxTokens({
-    messages,
-    maxTokens: toolModel.maxContext - 300 // filter token. not response maxToken
-  });
-  const formativeMessages = filterMessages.map((item) => {
+  // Filter histories by maxToken
+  const filterMessages = (
+    await filterGPTMessageByMaxTokens({
+      messages,
+      maxTokens: toolModel.maxContext - 300 // filter token. not response maxToken
+    })
+  ).map((item) => {
    if (item.role === 'assistant' && item.tool_calls) {
      return {
        ...item,
@@ -99,43 +113,43 @@ export const runToolWithToolChoice = async (
    }
    return item;
  });
-  const requestMessages = await loadRequestMessages(formativeMessages);

-  // console.log(
-  //   JSON.stringify(
-  //     {
-  //       ...toolModel?.defaultConfig,
-  //       model: toolModel.model,
-  //       temperature: 0,
-  //       stream,
-  //       messages: requestMessages,
-  //       tools,
-  //       tool_choice: 'auto'
-  //     },
-  //     null,
-  //     2
-  //   )
-  // );
+  const [requestMessages, max_tokens] = await Promise.all([
+    loadRequestMessages({
+      messages: filterMessages,
+      useVision: toolModel.vision && aiChatVision,
+      origin: requestOrigin
+    }),
+    computedMaxToken({
+      model: toolModel,
+      maxToken,
+      filterMessages
+    })
+  ]);
+  const requestBody: any = {
+    ...toolModel?.defaultConfig,
+    model: toolModel.model,
+    temperature: computedTemperature({
+      model: toolModel,
+      temperature
+    }),
+    max_tokens,
+    stream,
+    messages: requestMessages,
+    tools,
+    tool_choice: 'auto'
+  };
+
+  // console.log(JSON.stringify(requestBody, null, 2));
  /* Run llm */
  const ai = getAIApi({
    timeout: 480000
  });
-  const aiResponse = await ai.chat.completions.create(
-    {
-      ...toolModel?.defaultConfig,
-      model: toolModel.model,
-      temperature: 0,
-      stream,
-      messages: requestMessages,
-      tools,
-      tool_choice: 'auto'
-    },
-    {
-      headers: {
-        Accept: 'application/json, text/plain, */*'
-      }
+  const aiResponse = await ai.chat.completions.create(requestBody, {
+    headers: {
+      Accept: 'application/json, text/plain, */*'
    }
-  );
+  });

  const { answer, toolCalls } = await (async () => {
    if (res && stream) {
@@ -221,7 +235,7 @@ export const runToolWithToolChoice = async (
                toolName: '',
                toolAvatar: '',
                params: '',
-                response: stringToolResponse
+                response: sliceStrStartEnd(stringToolResponse, 300, 300)
              }
            })
          });
@@ -243,7 +257,7 @@ export const runToolWithToolChoice = async (
      tool_calls: toolCalls
    };
    const concatToolMessages = [
-      ...filterMessages,
+      ...requestMessages,
      assistantToolMsgParams
    ] as ChatCompletionMessageParam[];
    const tokens = await countGptMessagesTokens(concatToolMessages, tools);
--- a/packages/service/core/workflow/dispatch/agent/runTool/type.d.ts
+++ b/packages/service/core/workflow/dispatch/agent/runTool/type.d.ts
@@ -11,9 +11,13 @@ import { AIChatItemValueItemType, ChatItemValueItemType } from '@fastgpt/global/

 export type DispatchToolModuleProps = ModuleDispatchProps<{
  [NodeInputKeyEnum.history]?: ChatItemType[];
+  [NodeInputKeyEnum.userChatInput]: string;
+
  [NodeInputKeyEnum.aiModel]: string;
  [NodeInputKeyEnum.aiSystemPrompt]: string;
-  [NodeInputKeyEnum.userChatInput]: string;
+  [NodeInputKeyEnum.aiChatTemperature]: number;
+  [NodeInputKeyEnum.aiChatMaxToken]: number;
+  [NodeInputKeyEnum.aiChatVision]?: boolean;
 }>;

 export type RunToolResponse = {
--- a/packages/service/core/workflow/dispatch/agent/runTool/utils.ts
+++ b/packages/service/core/workflow/dispatch/agent/runTool/utils.ts
@@ -1,3 +1,6 @@
+import { sliceStrStartEnd } from '@fastgpt/global/common/string/tools';
+import { ChatItemValueTypeEnum } from '@fastgpt/global/core/chat/constants';
+import { AIChatItemValueItemType } from '@fastgpt/global/core/chat/type';
 import { FlowNodeInputItemType } from '@fastgpt/global/core/workflow/type/io';

 export const updateToolInputValue = ({
@@ -12,3 +15,22 @@ export const updateToolInputValue = ({
    value: params[input.key] ?? input.value
  }));
 };
+
+export const filterToolResponseToPreview = (response: AIChatItemValueItemType[]) => {
+  return response.map((item) => {
+    if (item.type === ChatItemValueTypeEnum.tool) {
+      const formatTools = item.tools?.map((tool) => {
+        return {
+          ...tool,
+          response: sliceStrStartEnd(tool.response, 500, 500)
+        };
+      });
+      return {
+        ...item,
+        tools: formatTools
+      };
+    }
+
+    return item;
+  });
+};
--- a/packages/service/core/workflow/dispatch/chat/oneapi.ts
+++ b/packages/service/core/workflow/dispatch/chat/oneapi.ts
@@ -1,9 +1,5 @@
 import type { NextApiResponse } from 'next';
-import {
-  filterGPTMessageByMaxTokens,
-  formatGPTMessagesInRequestBefore,
-  loadRequestMessages
-} from '../../../chat/utils';
+import { filterGPTMessageByMaxTokens, loadRequestMessages } from '../../../chat/utils';
 import type { ChatItemType, UserChatItemValueItemType } from '@fastgpt/global/core/chat/type.d';
 import { ChatRoleEnum } from '@fastgpt/global/core/chat/constants';
 import { SseResponseEventEnum } from '@fastgpt/global/core/workflow/runtime/constants';
@@ -19,10 +15,7 @@ import type { LLMModelItemType } from '@fastgpt/global/core/ai/model.d';
 import { postTextCensor } from '../../../../common/api/requestPlusApi';
 import { ChatCompletionRequestMessageRoleEnum } from '@fastgpt/global/core/ai/constants';
 import type { DispatchNodeResultType } from '@fastgpt/global/core/workflow/runtime/type';
-import {
-  countGptMessagesTokens,
-  countMessagesTokens
-} from '../../../../common/string/tiktoken/index';
+import { countMessagesTokens } from '../../../../common/string/tiktoken/index';
 import {
  chats2GPTMessages,
  chatValue2RuntimePrompt,
@@ -31,6 +24,7 @@ import {
  runtimePrompt2ChatsValue
 } from '@fastgpt/global/core/chat/adapt';
 import {
+  Prompt_DocumentQuote,
  Prompt_QuotePromptList,
  Prompt_QuoteTemplateList
 } from '@fastgpt/global/core/ai/prompt/AIChat';
@@ -46,6 +40,7 @@ import { getHistories } from '../utils';
 import { filterSearchResultsByMaxChars } from '../../utils';
 import { getHistoryPreview } from '@fastgpt/global/core/chat/utils';
 import { addLog } from '../../../../common/system/log';
+import { computedMaxToken, computedTemperature } from '../../../ai/utils';

 export type ChatProps = ModuleDispatchProps<
  AIChatNodeProps & {
@@ -63,6 +58,7 @@ export type ChatResponse = DispatchNodeResultType<{
 export const dispatchChatCompletion = async (props: ChatProps): Promise<ChatResponse> => {
  let {
    res,
+    requestOrigin,
    stream = false,
    detail = false,
    user,
@@ -79,7 +75,9 @@ export const dispatchChatCompletion = async (props: ChatProps): Promise<ChatResp
      isResponseAnswerText = true,
      systemPrompt = '',
      quoteTemplate,
-      quotePrompt
+      quotePrompt,
+      aiChatVision,
+      stringQuoteText
    }
  } = props;
  const { files: inputFiles } = chatValue2RuntimePrompt(query);
@@ -91,54 +89,43 @@ export const dispatchChatCompletion = async (props: ChatProps): Promise<ChatResp

  const chatHistories = getHistories(history, histories);

-  // temperature adapt
  const modelConstantsData = getLLMModel(model);
-
  if (!modelConstantsData) {
    return Promise.reject('The chat model is undefined, you need to select a chat model.');
  }

-  const { quoteText } = await filterQuote({
+  const { datasetQuoteText } = await filterDatasetQuote({
    quoteQA,
    model: modelConstantsData,
    quoteTemplate
  });

-  // censor model and system key
-  if (modelConstantsData.censor && !user.openaiAccount?.key) {
-    await postTextCensor({
-      text: `${systemPrompt}
-      ${quoteText}
-      ${userChatInput}
-      `
-    });
-  }
-
-  const { filterMessages } = await getChatMessages({
-    model: modelConstantsData,
-    histories: chatHistories,
-    quoteQA,
-    quoteText,
-    quotePrompt,
-    userChatInput,
-    inputFiles,
-    systemPrompt
-  });
-
-  const { max_tokens } = await getMaxTokens({
-    model: modelConstantsData,
-    maxToken,
-    filterMessages
-  });
-
-  // FastGPT temperature range: 1~10
-  temperature = +(modelConstantsData.maxTemperature * (temperature / 10)).toFixed(2);
-  temperature = Math.max(temperature, 0.01);
-  const ai = getAIApi({
-    userKey: user.openaiAccount,
-    timeout: 480000
-  });
+  const [{ filterMessages }] = await Promise.all([
+    getChatMessages({
+      model: modelConstantsData,
+      histories: chatHistories,
+      useDatasetQuote: quoteQA !== undefined,
+      datasetQuoteText,
+      datasetQuotePrompt: quotePrompt,
+      userChatInput,
+      inputFiles,
+      systemPrompt,
+      stringQuoteText
+    }),
+    async () => {
+      // censor model and system key
+      if (modelConstantsData.censor && !user.openaiAccount?.key) {
+        await postTextCensor({
+          text: `${systemPrompt}
+            ${datasetQuoteText}
+            ${userChatInput}
+          `
+        });
+      }
+    }
+  ]);

+  // Get the request messages
  const concatMessages = [
    ...(modelConstantsData.defaultSystemChatPrompt
      ? [
@@ -148,20 +135,39 @@ export const dispatchChatCompletion = async (props: ChatProps): Promise<ChatResp
          }
        ]
      : []),
-    ...formatGPTMessagesInRequestBefore(filterMessages)
+    ...filterMessages
  ] as ChatCompletionMessageParam[];

-  const requestMessages = await loadRequestMessages(concatMessages);
+  const [requestMessages, max_tokens] = await Promise.all([
+    loadRequestMessages({
+      messages: concatMessages,
+      useVision: modelConstantsData.vision && aiChatVision,
+      origin: requestOrigin
+    }),
+    computedMaxToken({
+      model: modelConstantsData,
+      maxToken,
+      filterMessages
+    })
+  ]);

  const requestBody = {
    ...modelConstantsData?.defaultConfig,
    model: modelConstantsData.model,
-    temperature,
+    temperature: computedTemperature({
+      model: modelConstantsData,
+      temperature
+    }),
    max_tokens,
    stream,
    messages: requestMessages
  };
+  // console.log(JSON.stringify(requestBody, null, 2), '===');
  try {
+    const ai = getAIApi({
+      userKey: user.openaiAccount,
+      timeout: 480000
+    });
    const response = await ai.chat.completions.create(requestBody, {
      headers: {
        Accept: 'application/json, text/plain, */*'
@@ -194,7 +200,7 @@ export const dispatchChatCompletion = async (props: ChatProps): Promise<ChatResp
      }
    })();

-    const completeMessages = filterMessages.concat({
+    const completeMessages = requestMessages.concat({
      role: ChatCompletionRequestMessageRoleEnum.Assistant,
      content: answerText
    });
@@ -243,7 +249,7 @@ export const dispatchChatCompletion = async (props: ChatProps): Promise<ChatResp
  }
 };

-async function filterQuote({
+async function filterDatasetQuote({
  quoteQA = [],
  model,
  quoteTemplate
@@ -265,44 +271,52 @@ async function filterQuote({
  // slice filterSearch
  const filterQuoteQA = await filterSearchResultsByMaxChars(quoteQA, model.quoteMaxToken);

-  const quoteText =
+  const datasetQuoteText =
    filterQuoteQA.length > 0
      ? `${filterQuoteQA.map((item, index) => getValue(item, index).trim()).join('\n------\n')}`
      : '';

  return {
-    quoteText
+    datasetQuoteText
  };
 }
 async function getChatMessages({
-  quotePrompt,
-  quoteText,
-  quoteQA,
+  datasetQuotePrompt,
+  datasetQuoteText,
+  useDatasetQuote,
  histories = [],
  systemPrompt,
  userChatInput,
  inputFiles,
-  model
+  model,
+  stringQuoteText
 }: {
-  quotePrompt?: string;
-  quoteText: string;
-  quoteQA: ChatProps['params']['quoteQA'];
+  datasetQuotePrompt?: string;
+  datasetQuoteText: string;
+  useDatasetQuote: boolean;
  histories: ChatItemType[];
  systemPrompt: string;
  userChatInput: string;
  inputFiles: UserChatItemValueItemType['file'][];
  model: LLMModelItemType;
+  stringQuoteText?: string;
 }) {
-  const replaceInputValue =
-    quoteQA !== undefined
-      ? replaceVariable(quotePrompt || Prompt_QuotePromptList[0].value, {
-          quote: quoteText,
-          question: userChatInput
-        })
-      : userChatInput;
+  const replaceInputValue = useDatasetQuote
+    ? replaceVariable(datasetQuotePrompt || Prompt_QuotePromptList[0].value, {
+        quote: datasetQuoteText,
+        question: userChatInput
+      })
+    : userChatInput;

  const messages: ChatItemType[] = [
    ...getSystemPrompt(systemPrompt),
+    ...(stringQuoteText
+      ? getSystemPrompt(
+          replaceVariable(Prompt_DocumentQuote, {
+            quote: stringQuoteText
+          })
+        )
+      : []),
    ...histories,
    {
      obj: ChatRoleEnum.Human,
@@ -323,29 +337,6 @@ async function getChatMessages({
    filterMessages
  };
 }
-async function getMaxTokens({
-  maxToken,
-  model,
-  filterMessages = []
-}: {
-  maxToken: number;
-  model: LLMModelItemType;
-  filterMessages: ChatCompletionMessageParam[];
-}) {
-  maxToken = Math.min(maxToken, model.maxResponse);
-  const tokensLimit = model.maxContext;
-
-  /* count response max token */
-  const promptsToken = await countGptMessagesTokens(filterMessages);
-  maxToken = promptsToken + maxToken > tokensLimit ? tokensLimit - promptsToken : maxToken;
-
-  if (maxToken <= 0) {
-    maxToken = 200;
-  }
-  return {
-    max_tokens: maxToken
-  };
-}

 async function streamResponse({
  res,
--- a/packages/service/core/workflow/dispatch/index.ts
+++ b/packages/service/core/workflow/dispatch/index.ts
@@ -55,6 +55,7 @@ import { surrenderProcess } from '../../../common/system/tools';
 import { dispatchRunCode } from './code/run';
 import { dispatchTextEditor } from './tools/textEditor';
 import { dispatchCustomFeedback } from './tools/customFeedback';
+import { dispatchReadFiles } from './tools/readFiles';

 const callbackMap: Record<FlowNodeTypeEnum, Function> = {
  [FlowNodeTypeEnum.workflowStart]: dispatchWorkflowStart,
@@ -78,6 +79,7 @@ const callbackMap: Record<FlowNodeTypeEnum, Function> = {
  [FlowNodeTypeEnum.code]: dispatchRunCode,
  [FlowNodeTypeEnum.textEditor]: dispatchTextEditor,
  [FlowNodeTypeEnum.customFeedback]: dispatchCustomFeedback,
+  [FlowNodeTypeEnum.readFiles]: dispatchReadFiles,

  // none
  [FlowNodeTypeEnum.systemConfig]: dispatchSystemConfig,
--- a/packages/service/core/workflow/dispatch/init/workflowStart.tsx
+++ b/packages/service/core/workflow/dispatch/init/workflowStart.tsx
@@ -1,13 +1,16 @@
 import { chatValue2RuntimePrompt } from '@fastgpt/global/core/chat/adapt';
-import { UserChatItemValueItemType } from '@fastgpt/global/core/chat/type';
-import { NodeInputKeyEnum } from '@fastgpt/global/core/workflow/constants';
+import { NodeInputKeyEnum, NodeOutputKeyEnum } from '@fastgpt/global/core/workflow/constants';
 import type { ModuleDispatchProps } from '@fastgpt/global/core/workflow/runtime/type';
+
 export type UserChatInputProps = ModuleDispatchProps<{
  [NodeInputKeyEnum.userChatInput]: string;
-  [NodeInputKeyEnum.inputFiles]: UserChatItemValueItemType['file'][];
 }>;
+type Response = {
+  [NodeOutputKeyEnum.userChatInput]: string;
+  [NodeOutputKeyEnum.userFiles]: string[];
+};

-export const dispatchWorkflowStart = (props: Record<string, any>) => {
+export const dispatchWorkflowStart = (props: Record<string, any>): Response => {
  const {
    query,
    params: { userChatInput }
@@ -17,6 +20,11 @@ export const dispatchWorkflowStart = (props: Record<string, any>) => {

  return {
    [NodeInputKeyEnum.userChatInput]: text || userChatInput,
-    [NodeInputKeyEnum.inputFiles]: files
+    [NodeOutputKeyEnum.userFiles]: files
+      .map((item) => {
+        return item?.url ?? '';
+      })
+      .filter(Boolean)
+    // [NodeInputKeyEnum.inputFiles]: files
  };
 };
--- a/packages/service/core/workflow/dispatch/tools/readFiles.ts
+++ b/packages/service/core/workflow/dispatch/tools/readFiles.ts
@@ -0,0 +1,196 @@
+import { DispatchNodeResponseKeyEnum } from '@fastgpt/global/core/workflow/runtime/constants';
+import type { ModuleDispatchProps } from '@fastgpt/global/core/workflow/runtime/type';
+import { NodeInputKeyEnum, NodeOutputKeyEnum } from '@fastgpt/global/core/workflow/constants';
+import { DispatchNodeResultType } from '@fastgpt/global/core/workflow/runtime/type';
+import { documentFileType } from '@fastgpt/global/common/file/constants';
+import axios from 'axios';
+import { serverRequestBaseUrl } from '../../../../common/api/serverRequest';
+import { MongoRawTextBuffer } from '../../../../common/buffer/rawText/schema';
+import { readFromSecondary } from '../../../../common/mongo/utils';
+import { getErrText } from '@fastgpt/global/common/error/utils';
+import { detectFileEncoding } from '@fastgpt/global/common/file/tools';
+import { readRawContentByFileBuffer } from '../../../../common/file/read/utils';
+import { ChatRoleEnum } from '@fastgpt/global/core/chat/constants';
+import { UserChatItemValueItemType } from '@fastgpt/global/core/chat/type';
+
+type Props = ModuleDispatchProps<{
+  [NodeInputKeyEnum.fileUrlList]: string[];
+}>;
+type Response = DispatchNodeResultType<{
+  [NodeOutputKeyEnum.text]: string;
+}>;
+
+const formatResponseObject = ({
+  filename,
+  url,
+  content
+}: {
+  filename: string;
+  url: string;
+  content: string;
+}) => ({
+  filename,
+  url,
+  text: `File: ${filename}
+<Content>
+${content}
+</Content>`,
+  nodeResponsePreviewText: `File: ${filename}
+<Content>
+${content.slice(0, 100)}${content.length > 100 ? '......' : ''}
+</Content>`
+});
+
+export const dispatchReadFiles = async (props: Props): Promise<Response> => {
+  const {
+    requestOrigin,
+    teamId,
+    histories,
+    chatConfig,
+    params: { fileUrlList = [] }
+  } = props;
+  const maxFiles = chatConfig?.fileSelectConfig?.maxFiles || 0;
+
+  // Get files from histories
+  const filesFromHistories = histories
+    .filter((item) => {
+      if (item.obj === ChatRoleEnum.Human) {
+        return item.value.filter((value) => value.type === 'file');
+      }
+      return false;
+    })
+    .map((item) => {
+      const value = item.value as UserChatItemValueItemType[];
+      const files = value
+        .map((item) => {
+          return item.file?.url;
+        })
+        .filter(Boolean) as string[];
+      return files;
+    })
+    .flat();
+
+  const parseUrlList = [...fileUrlList, ...filesFromHistories].slice(0, maxFiles);
+
+  const readFilesResult = await Promise.all(
+    parseUrlList
+      .map(async (url) => {
+        // System file
+        if (url.startsWith('/') || (requestOrigin && url.startsWith(requestOrigin))) {
+          // Parse url, get filename query. Keep only documents that can be parsed
+          const parseUrl = new URL(url);
+          const filenameQuery = parseUrl.searchParams.get('filename');
+          if (filenameQuery) {
+            const extensionQuery = filenameQuery.split('.').pop()?.toLowerCase() || '';
+            if (!documentFileType.includes(extensionQuery)) {
+              return;
+            }
+          }
+
+          //  Remove the origin(Make intranet requests directly)
+          if (requestOrigin && url.startsWith(requestOrigin)) {
+            url = url.replace(requestOrigin, '');
+          }
+        }
+
+        // Get from buffer
+        const fileBuffer = await MongoRawTextBuffer.findOne({ sourceId: url }, undefined, {
+          ...readFromSecondary
+        }).lean();
+        if (fileBuffer) {
+          return formatResponseObject({
+            filename: fileBuffer.metadata?.filename || url,
+            url,
+            content: fileBuffer.rawText
+          });
+        }
+
+        try {
+          // Get file buffer
+          const response = await axios.get(url, {
+            baseURL: serverRequestBaseUrl,
+            responseType: 'arraybuffer'
+          });
+
+          const buffer = Buffer.from(response.data, 'binary');
+
+          // Get file name
+          const filename = (() => {
+            const contentDisposition = response.headers['content-disposition'];
+            if (contentDisposition) {
+              const filenameRegex = /filename[^;=\n]*=((['"]).*?\2|[^;\n]*)/;
+              const matches = filenameRegex.exec(contentDisposition);
+              if (matches != null && matches[1]) {
+                return decodeURIComponent(matches[1].replace(/['"]/g, ''));
+              }
+            }
+
+            return url;
+          })();
+          // Extension
+          const extension = filename.split('.').pop()?.toLowerCase() || '';
+          // Get encoding
+          const encoding = (() => {
+            const contentType = response.headers['content-type'];
+            if (contentType) {
+              const charsetRegex = /charset=([^;]*)/;
+              const matches = charsetRegex.exec(contentType);
+              if (matches != null && matches[1]) {
+                return matches[1];
+              }
+            }
+
+            return detectFileEncoding(buffer);
+          })();
+
+          // Read file
+          const { rawText } = await readRawContentByFileBuffer({
+            extension,
+            isQAImport: false,
+            teamId,
+            buffer,
+            encoding
+          });
+
+          // Add to buffer
+          try {
+            if (buffer.length < 14 * 1024 * 1024 && rawText.trim()) {
+              MongoRawTextBuffer.create({
+                sourceId: url,
+                rawText,
+                metadata: {
+                  filename: filename
+                }
+              });
+            }
+          } catch (error) {}
+
+          return formatResponseObject({ filename, url, content: rawText });
+        } catch (error) {
+          return formatResponseObject({
+            filename: '',
+            url,
+            content: getErrText(error, 'Load file error')
+          });
+        }
+      })
+      .filter(Boolean)
+  );
+  const text = readFilesResult.map((item) => item?.text ?? '').join('\n******\n');
+
+  return {
+    [NodeOutputKeyEnum.text]: text,
+    [DispatchNodeResponseKeyEnum.nodeResponse]: {
+      readFiles: readFilesResult.map((item) => ({
+        name: item?.filename || '',
+        url: item?.url || ''
+      })),
+      readFilesResult: readFilesResult
+        .map((item) => item?.nodeResponsePreviewText ?? '')
+        .join('\n******\n')
+    },
+    [DispatchNodeResponseKeyEnum.toolResponses]: {
+      fileContent: text
+    }
+  };
+};