CherryHQ · lpffernando · Oct 31, 2025
diff --git a/src/main/knowledge/preprocess/MineruPreprocessProvider.ts b/src/main/knowledge/preprocess/MineruPreprocessProvider.ts
@@ -3,9 +3,12 @@ import path from 'node:path'
 
 import { loggerService } from '@logger'
 import { fileStorage } from '@main/services/FileStorage'
-import { FileMetadata, PreprocessProvider } from '@types'
+import { imageExts } from '@shared/config/constant'
+import { FileMetadata, FileTypes, PreprocessProvider } from '@types'
 import AdmZip from 'adm-zip'
 import { net } from 'electron'
+import { PDFDocument } from 'pdf-lib'
+import sharp from 'sharp'
 
 import BasePreprocessProvider from './BasePreprocessProvider'
 
@@ -63,33 +66,54 @@ export default class MineruPreprocessProvider extends BasePreprocessProvider {
     sourceId: string,
     file: FileMetadata
   ): Promise<{ processedFile: FileMetadata; quota: number }> {
+    let cleanupPath: string | null = null
     try {
-      const filePath = fileStorage.getFilePathById(file)
-      logger.info(`MinerU preprocess processing started: ${filePath}`)
-      await this.validateFile(filePath)
+      const originalPath = fileStorage.getFilePathById(file)
+      logger.info(`MinerU preprocess processing started: ${originalPath}`)
+
+      let workingFile: FileMetadata = file
+      let workingFilePath = originalPath
+
+      if (this.isImageFile(file)) {
+        const converted = await this.convertImageToPdf(file, originalPath)
+        workingFile = converted.metadata
+        workingFilePath = converted.path
+        cleanupPath = converted.path
+        logger.info(`Converted image to PDF for MinerU preprocessing: ${workingFilePath}`)
+      }
+
+      await this.validateFile(workingFilePath)
 
       // 1. 获取上传URL并上传文件
-      const batchId = await this.uploadFile(file)
+      const batchId = await this.uploadFile(workingFile, workingFilePath)
       logger.info(`MinerU file upload completed: batch_id=${batchId}`)
 
       // 2. 等待处理完成并获取结果
-      const extractResult = await this.waitForCompletion(sourceId, batchId, file.origin_name)
+      const extractResult = await this.waitForCompletion(sourceId, batchId, workingFile.origin_name)
       logger.info(`MinerU processing completed for batch: ${batchId}`)
 
       // 3. 下载并解压文件
-      const { path: outputPath } = await this.downloadAndExtractFile(extractResult.full_zip_url!, file)
+      const { path: outputPath } = await this.downloadAndExtractFile(extractResult.full_zip_url!, workingFile)
 
       // 4. check quota
       const quota = await this.checkQuota()
 
       // 5. 创建处理后的文件信息
       return {
-        processedFile: this.createProcessedFileInfo(file, outputPath),
+        processedFile: this.createProcessedFileInfo(workingFile, outputPath),
         quota
       }
     } catch (error: any) {
       logger.error(`MinerU preprocess processing failed for:`, error as Error)
       throw new Error(error.message)
+    } finally {
+      if (cleanupPath) {
+        try {
+          await fs.promises.unlink(cleanupPath)
+        } catch (cleanupError) {
+          logger.warn(`Failed to cleanup temporary MinerU conversion file ${cleanupPath}:`, cleanupError as Error)
+        }
+      }
     }
   }
 
@@ -207,12 +231,11 @@ export default class MineruPreprocessProvider extends BasePreprocessProvider {
     }
   }
 
-  private async uploadFile(file: FileMetadata): Promise<string> {
+  private async uploadFile(file: FileMetadata, filePath: string): Promise<string> {
     try {
       // 步骤1: 获取上传URL
       const { batchId, fileUrls } = await this.getBatchUploadUrls(file)
       // 步骤2: 上传文件到获取的URL
-      const filePath = fileStorage.getFilePathById(file)
       await this.putFileToUrl(filePath, fileUrls[0])
       logger.info(`File uploaded successfully: ${filePath}`, { batchId, fileUrls })
 
@@ -223,6 +246,67 @@ export default class MineruPreprocessProvider extends BasePreprocessProvider {
     }
   }
 
+  private isImageFile(file: FileMetadata): boolean {
+    const ext = (file.ext || '').toLowerCase()
+    return imageExts.includes(ext)
+  }
+
+  private async convertImageToPdf(
+    file: FileMetadata,
+    sourcePath: string
+  ): Promise<{ metadata: FileMetadata; path: string }> {
+    try {
+      const ext = (file.ext || '').toLowerCase()
+      const imageBuffer = await fs.promises.readFile(sourcePath)
+
+      let convertedBuffer: Buffer
+      let embedType: 'jpg' | 'png'
+
+      if (ext === '.jpg' || ext === '.jpeg') {
+        convertedBuffer = imageBuffer
+        embedType = 'jpg'
+      } else {
+        convertedBuffer = await sharp(imageBuffer).png().toBuffer()
+        embedType = 'png'
+      }
+
+      const pdfDoc = await PDFDocument.create()
+      const embeddedImage =
+        embedType === 'jpg' ? await pdfDoc.embedJpg(convertedBuffer) : await pdfDoc.embedPng(convertedBuffer)
+      const { width, height } = embeddedImage
+
+      const page = pdfDoc.addPage([width, height])
+      page.drawImage(embeddedImage, {
+        x: 0,
+        y: 0,
+        width,
+        height
+      })
+
+      const pdfBytes = await pdfDoc.save()
+      const pdfPath = path.join(this.storageDir, `${file.id}-mineru.pdf`)
+      await fs.promises.writeFile(pdfPath, pdfBytes)
+
+      const originNameBase = path.parse(file.origin_name).name || file.origin_name
+
+      const metadata: FileMetadata = {
+        ...file,
+        path: pdfPath,
+        size: pdfBytes.length,
+        ext: '.pdf',
+        origin_name: `${originNameBase}.pdf`,
+        name: `${file.id}.pdf`,
+        type: FileTypes.DOCUMENT,
+        created_at: new Date().toISOString()
+      }
+
+      return { metadata, path: pdfPath }
+    } catch (error: any) {
+      logger.error(`Failed to convert image ${file.origin_name} to PDF: ${error.message}`)
+      throw new Error(`Failed to convert image to PDF: ${error.message}`)
+    }
+  }
+
   private async getBatchUploadUrls(file: FileMetadata): Promise<{ batchId: string; fileUrls: string[] }> {
     const endpoint = `${this.provider.apiHost}/api/v4/file-urls/batch`
 

diff --git a/src/main/services/KnowledgeService.ts b/src/main/services/KnowledgeService.ts
@@ -13,7 +13,7 @@
 * @see {@link ../../../docs/technical/KnowledgeService.md}
 */

 import * as fs from 'node:fs'
 import path from 'node:path'

 import { RAGApplication, RAGApplicationBuilder } from '@cherrystudio/embedjs'
@@ -31,10 +31,10 @@
 import { getDataPath } from '@main/utils'
 import { getAllFiles, sanitizeFilename } from '@main/utils/file'
 import { TraceMethod } from '@mcp-trace/trace-core'
-import { MB } from '@shared/config/constant'
+import { MB, imageExts } from '@shared/config/constant'
 import type { LoaderReturn } from '@shared/config/types'
 import { IpcChannel } from '@shared/IpcChannel'
-import { FileMetadata, KnowledgeBaseParams, KnowledgeItem, KnowledgeSearchResult } from '@types'
+import { FileMetadata, FileTypes, KnowledgeBaseParams, KnowledgeItem, KnowledgeSearchResult } from '@types'
 import { v4 as uuidv4 } from 'uuid'
 
 const logger = loggerService.withContext('MainKnowledgeService')
@@ -298,6 +298,12 @@
       this.workload >= KnowledgeService.MAXIMUM_WORKLOAD
     )
   }
+
+  private isImageFile(file: FileMetadata): boolean {
+    const ext = (file.ext || '').toLowerCase()
+    return file.type === FileTypes.IMAGE || imageExts.some((imageExt) => imageExt === ext)
+  }
+
   private fileTask(
     ragApplication: RAGApplication,
     options: KnowledgeBaseAddItemOptionsNonNullableAttribute
@@ -313,23 +319,20 @@
             try {
               // Add preprocessing logic
               const fileToProcess: FileMetadata = await this.preprocessing(file, base, item, userId)
-
-              // Use processed file for loading
-              return addFileLoader(ragApplication, fileToProcess, base, forceReload)
-                .then((result) => {
-                  loaderTask.loaderDoneReturn = result
-                  return result
-                })
-                .catch((e) => {
-                  logger.error(`Error in addFileLoader for ${file.name}: ${e}`)
-                  const errorResult: LoaderReturn = {
-                    ...KnowledgeService.ERROR_LOADER_RETURN,
-                    message: e.message,
-                    messageSource: 'embedding'
-                  }
-                  loaderTask.loaderDoneReturn = errorResult
-                  return errorResult
-                })
+              try {
+                const result = await addFileLoader(ragApplication, fileToProcess, base, forceReload)
+                loaderTask.loaderDoneReturn = result
+                return result
+              } catch (e: any) {
+                logger.error(`Error in addFileLoader for ${file.name}: ${e}`)
+                const errorResult: LoaderReturn = {
+                  ...KnowledgeService.ERROR_LOADER_RETURN,
+                  message: e.message,
+                  messageSource: 'embedding'
+                }
+                loaderTask.loaderDoneReturn = errorResult
+                return errorResult
+              }
             } catch (e: any) {
               logger.error(`Preprocessing failed for ${file.name}: ${e}`)
               const errorResult: LoaderReturn = {
@@ -692,9 +695,16 @@
     userId: string
   ): Promise<FileMetadata> => {
     let fileToProcess: FileMetadata = file
-    if (base.preprocessProvider && file.ext.toLowerCase() === '.pdf') {
+    const ext = (file.ext || '').toLowerCase()
+    const preprocessConfig = base.preprocessProvider
+    const providerId = preprocessConfig?.provider.id
+    const supportsImagePreprocess =
+      this.isImageFile(file) && Boolean(providerId && ['mineru', 'open-mineru'].includes(providerId))
+    const shouldUsePreprocess = Boolean(preprocessConfig && (ext === '.pdf' || supportsImagePreprocess))
+
+    if (shouldUsePreprocess && preprocessConfig) {
       try {
-        const provider = new PreprocessProvider(base.preprocessProvider.provider, userId)
+        const provider = new PreprocessProvider(preprocessConfig.provider, userId)
         const filePath = fileStorage.getFilePathById(file)
         // Check if file has already been preprocessed
         const alreadyProcessed = await provider.checkIfAlreadyProcessed(file)

diff --git a/src/renderer/src/pages/knowledge/items/KnowledgeFiles.tsx b/src/renderer/src/pages/knowledge/items/KnowledgeFiles.tsx
@@ -8,7 +8,7 @@ import FileManager from '@renderer/services/FileManager'
 import { getProviderName } from '@renderer/services/ProviderService'
 import { FileMetadata, FileTypes, isKnowledgeFileItem, KnowledgeBase, KnowledgeItem } from '@renderer/types'
 import { formatFileSize, uuid } from '@renderer/utils'
-import { bookExts, documentExts, textExts, thirdPartyApplicationExts } from '@shared/config/constant'
+import { bookExts, documentExts, imageExts, textExts, thirdPartyApplicationExts } from '@shared/config/constant'
 import { Button, Tooltip, Upload } from 'antd'
 import dayjs from 'dayjs'
 import { FC, useCallback, useEffect, useState } from 'react'
@@ -40,7 +40,7 @@ interface KnowledgeContentProps {
   preprocessMap: Map<string, boolean>
 }
 
-const fileTypes = [...bookExts, ...thirdPartyApplicationExts, ...documentExts, ...textExts]
+const fileTypes = [...bookExts, ...thirdPartyApplicationExts, ...documentExts, ...textExts, ...imageExts]
 
 const getDisplayTime = (item: KnowledgeItem) => {
   const timestamp = item.updated_at && item.updated_at > item.created_at ? item.updated_at : item.created_at
@@ -165,7 +165,7 @@ const KnowledgeFiles: FC<KnowledgeContentProps> = ({ selectedBase, progressMap,
             openFileDialogOnClick={false}>
             <p className="ant-upload-text">{t('knowledge.drag_file')}</p>
             <p className="ant-upload-hint">
-              {t('knowledge.file_hint', { file_types: 'TXT, MD, HTML, PDF, DOCX, PPTX, XLSX, EPUB...' })}
+              {t('knowledge.file_hint', { file_types: 'TXT, MD, HTML, PDF, DOCX, PPTX, XLSX, EPUB, PNG, JPG...' })}
             </p>
           </Dragger>
         </div>

diff --git a/src/renderer/src/queue/KnowledgeQueue.ts b/src/renderer/src/queue/KnowledgeQueue.ts
@@ -1,4 +1,4 @@
 import { loggerService } from '@logger'
 import db from '@renderer/databases'
 import { getStoreSetting } from '@renderer/hooks/useSettings'
 import { getKnowledgeBaseParams } from '@renderer/services/KnowledgeService'
@@ -10,7 +10,7 @@
   updateBaseItemUniqueId,
   updateItemProcessingStatus
 } from '@renderer/store/knowledge'
-import { KnowledgeItem } from '@renderer/types'
+import { FileTypes, KnowledgeItem, isKnowledgeFileItem } from '@renderer/types'
 import { uuid } from '@renderer/utils'
 import type { LoaderReturn } from '@shared/config/types'
 import { t } from 'i18next'
@@ -127,11 +127,16 @@
         throw new Error(`[KnowledgeQueue] Source item ${item.id} not found in base ${baseId}`)
       }
 
-      let result: LoaderReturn | null = null
-      let note, content
+  let result: LoaderReturn | null = null
+  let note, content
 
       logger.info(`Processing item: ${sourceItem.content}`)
 
+      const isImageItem = isKnowledgeFileItem(sourceItem) && sourceItem.content.type === FileTypes.IMAGE
+      const preprocessProviderId = base.preprocessProvider?.provider.id
+      const shouldUsePreprocessForImage =
+        isImageItem && Boolean(preprocessProviderId && ['mineru', 'open-mineru'].includes(preprocessProviderId))
+
       switch (item.type) {
         case 'note':
           note = await db.knowledge_notes.get(item.id)
@@ -201,7 +206,7 @@
           updateBaseItemIsPreprocessed({
             baseId,
             itemId: item.id,
-            isPreprocessed: !!base.preprocessProvider
+            isPreprocessed: shouldUsePreprocessForImage ? true : !!base.preprocessProvider
           })
         )
       }