feat(client): enhance knowledge file upload with process configuration support

- Updated the `CreateKnowledgeFromFile` function to accept an optional `processConfig` parameter for per-upload parse configuration overrides. - Introduced the `KnowledgeProcessOverrides` type to define various parsing settings, including parser engine rules and chunking configurations. - Modified the API client and related components to handle the new process configuration, improving flexibility in knowledge file uploads. - Updated documentation and API specifications to reflect the new `process_config` parameter in the upload process. - Added a new `UploadConfirmHost` component to facilitate user confirmation for uploads, integrating with the updated upload logic. - Enhanced user experience by providing detailed configuration options during the upload process.
2026-09-21 05:43:43 +08:00 · 2026-06-09 20:38:24 +08:00
parent 589d2a6e52
commit ed9e780c8c
41 changed files with 3546 additions and 596 deletions
@@ -71,7 +71,7 @@ func ExampleUsage() {
 			"source": "local",
 			"type":   "document",
 		}
-		knowledge, err := apiClient.CreateKnowledgeFromFile(context.Background(), createdKB.ID, filePath, metadata, nil, "", "")
+		knowledge, err := apiClient.CreateKnowledgeFromFile(context.Background(), createdKB.ID, filePath, metadata, nil, "", "", nil)
 		if err != nil {
 			fmt.Printf("Failed to upload knowledge file: %v\n", err)
 		} else {
@@ -83,6 +83,19 @@ var ErrDuplicateFile = errors.New("file already exists")
 // ErrDuplicateURL is returned when attempting to create a knowledge entry with a URL that already exists
 var ErrDuplicateURL = errors.New("URL already exists")

+// KnowledgeProcessOverrides stores per-upload parse config overrides sent as process_config.
+// When nil, the server uses the knowledge base defaults only.
+type KnowledgeProcessOverrides struct {
+	ParserEngineRules        []ParserEngineRule            `json:"parser_engine_rules,omitempty"`
+	ChunkingConfig           *ChunkingConfig               `json:"chunking_config,omitempty"`
+	EnableMultimodel         *bool                         `json:"enable_multimodel,omitempty"`
+	VLMConfig                *VLMConfig                    `json:"vlm_config,omitempty"`
+	ASRConfig                *ASRConfig                    `json:"asr_config,omitempty"`
+	QuestionGenerationConfig *QuestionGenerationConfig     `json:"question_generation_config,omitempty"`
+	GraphEnabled             *bool                         `json:"graph_enabled,omitempty"`
+	ExtractConfig            *ExtractConfig                `json:"extract_config,omitempty"`
+}
+
 // CreateKnowledgeFromFile creates a knowledge entry from a local file path
 // Parameters:
 //   - knowledgeBaseID: The ID of the knowledge base
@@ -91,8 +104,10 @@ var ErrDuplicateURL = errors.New("URL already exists")
 //   - enableMultimodel: Optional flag to enable multimodal processing
 //   - customFileName: Optional custom file name (useful for folder uploads with path)
 //   - channel: Optional ingestion channel (e.g. "web", "api", "wechat"); empty defaults to "web"
+//   - processConfig: Optional parse config overrides (serialized as process_config form field)
 func (c *Client) CreateKnowledgeFromFile(ctx context.Context,
 	knowledgeBaseID string, filePath string, metadata map[string]string, enableMultimodel *bool, customFileName string, channel string,
+	processConfig *KnowledgeProcessOverrides,
 ) (*Knowledge, error) {
 	// Open the local file
 	file, err := os.Open(filePath)
@@ -159,6 +174,16 @@ func (c *Client) CreateKnowledgeFromFile(ctx context.Context,
 		}
 	}

+	if processConfig != nil {
+		processConfigBytes, err := json.Marshal(processConfig)
+		if err != nil {
+			return nil, fmt.Errorf("failed to serialize process_config: %w", err)
+		}
+		if err := writer.WriteField("process_config", string(processConfigBytes)); err != nil {
+			return nil, fmt.Errorf("failed to write process_config field: %w", err)
+		}
+	}
+
 	// Close the multipart writer
 	err = writer.Close()
 	if err != nil {
@@ -209,6 +234,8 @@ type CreateKnowledgeFromURLRequest struct {
 	TagID string `json:"tag_id,omitempty"`
 	// Channel identifies the ingestion channel (e.g. "web", "browser_extension", "api")
 	Channel string `json:"channel,omitempty"`
+	// ProcessConfig is optional per-upload parse config overrides (KnowledgeProcessOverrides).
+	ProcessConfig *KnowledgeProcessOverrides `json:"process_config,omitempty"`
 }

 // CreateKnowledgeFromURL creates a knowledge entry from a URL.
@@ -108,6 +108,25 @@ type GraphRelation struct {
 	Type  string `json:"type"`
 }

+// ParserEngineRule maps a set of file types to a specific parser engine.
+type ParserEngineRule struct {
+	FileTypes []string `json:"file_types"`
+	Engine    string   `json:"engine"`
+}
+
+// QuestionGenerationConfig controls LLM-generated questions per chunk during parsing.
+type QuestionGenerationConfig struct {
+	Enabled         bool `json:"enabled"`
+	QuestionCount   int  `json:"question_count"`
+}
+
+// ASRConfig represents automatic speech recognition settings for audio files.
+type ASRConfig struct {
+	Enabled  bool   `json:"enabled"`
+	ModelID  string `json:"model_id"`
+	Language string `json:"language,omitempty"`
+}
+
 // UnmarshalJSON keeps backward compatibility for legacy responses that still
 // use `cos_config` instead of `storage_config`.
 func (kb *KnowledgeBase) UnmarshalJSON(data []byte) error {
@@ -52,9 +52,12 @@
 | `fileName`          | string  | 否   | 自定义文件名，用于"文件夹上传"时保留相对路径（如 `docs/intro.md`） |
 | `metadata`          | string  | 否   | JSON 字符串，会被反序列化为 `map[string]string`                     |
 | `enable_multimodel` | string  | 否   | `"true"` / `"false"`，是否启用图文多模态解析                         |
+| `process_config`    | string  | 否   | JSON 字符串，批次解析配置覆盖（`KnowledgeProcessOverrides`）；写入 `knowledge.metadata.process_overrides`。未传时行为与现网一致 |
 | `tag_id`            | string  | 否   | 标签 ID；传 `__untagged__` 或空字符串表示未分类                      |
 | `channel`           | string  | 否   | 来源渠道标识（写入 `channel` 字段，默认 `web`）                      |

+`process_config` 可选字段包括：`parser_engine_rules`、`chunking_config`、`enable_multimodel`、`vlm_config`、`asr_config`、`question_generation_config`、`graph_enabled`、`extract_config`。若同时传 `enable_multimodel` 与 `process_config.enable_multimodel`，以 `process_config` 为准。
+
 **请求**:

 ```curl
@@ -4775,6 +4775,12 @@ const docTemplate = `{
                        "description": "启用多模态处理",
                        "name": "enable_multimodel",
                        "in": "formData"
+                    },
+                    {
+                        "type": "string",
+                        "description": "处理配置JSON（KnowledgeProcessOverrides）",
+                        "name": "process_config",
+                        "in": "formData"
                    }
                ],
                "responses": {
@@ -4768,6 +4768,12 @@
                        "description": "启用多模态处理",
                        "name": "enable_multimodel",
                        "in": "formData"
+                    },
+                    {
+                        "type": "string",
+                        "description": "处理配置JSON（KnowledgeProcessOverrides）",
+                        "name": "process_config",
+                        "in": "formData"
                    }
                ],
                "responses": {
@@ -7746,6 +7746,10 @@ paths:
        in: formData
        name: enable_multimodel
        type: boolean
+      - description: 处理配置JSON（KnowledgeProcessOverrides）
+        in: formData
+        name: process_config
+        type: string
      produces:
      - application/json
      responses:
@@ -4,6 +4,7 @@ import { useRouter } from 'vue-router'
 import { useI18n } from 'vue-i18n'
 import { MessagePlugin, NotifyPlugin } from 'tdesign-vue-next'
 import ManualKnowledgeEditor from '@/components/manual-knowledge-editor.vue'
+import UploadConfirmHost from '@/components/UploadConfirmHost.vue'
 import { useAuthStore } from '@/stores/auth'
 import { useSettingsStore } from '@/stores/settings'
 import { getCurrentUser, userInfoFromApi } from '@/api/auth'
@@ -257,6 +258,7 @@ onUnmounted(() => {
    <div id="app">
      <RouterView />
      <ManualKnowledgeEditor />
+      <UploadConfirmHost />
    </div>
  </t-config-provider>
 </template>
@@ -1,4 +1,5 @@
 import { get, post, put, del, postUpload, getDown } from "../../utils/request";
+import type { KnowledgeProcessOverrides } from '@/types/knowledgeProcess';

 // 知识库管理 API（列表、创建、获取、更新、删除、复制）
 export function listKnowledgeBases(params?: {
@@ -160,23 +161,51 @@ export function togglePinKnowledgeBase(id: string) {

 // 知识文件 API（基于具体知识库）
 // data.tag_id: 可选，指定知识所属的分类ID
-export function uploadKnowledgeFile(kbId: string, data: { file: File; tag_id?: string; [key: string]: any } = { file: new File([], '') }, onProgress?: (progressEvent: any) => void) {
+export function uploadKnowledgeFile(
+  kbId: string,
+  data: {
+    file: File
+    tag_id?: string
+    fileName?: string
+    process_config?: KnowledgeProcessOverrides | string
+    [key: string]: any
+  } = { file: new File([], '') },
+  onProgress?: (progressEvent: any) => void,
+) {
  const formData = new FormData();
  Object.keys(data).forEach(key => {
-    if (data[key] !== undefined) formData.append(key, data[key]);
+    const value = data[key];
+    if (value === undefined) return;
+    if (key === 'process_config' && value && typeof value !== 'string') {
+      formData.append(key, JSON.stringify(value));
+    } else {
+      formData.append(key, value);
+    }
  });
  return postUpload(`/api/v1/knowledge-bases/${kbId}/knowledge/file`, formData, onProgress);
 }

 // 从URL创建知识
 // data.tag_id: 可选，指定知识所属的分类ID
-export function createKnowledgeFromURL(kbId: string, data: { url: string; enable_multimodel?: boolean; tag_id?: string }) {
+export function createKnowledgeFromURL(
+  kbId: string,
+  data: { url: string; enable_multimodel?: boolean; tag_id?: string; process_config?: KnowledgeProcessOverrides },
+) {
  return post(`/api/v1/knowledge-bases/${kbId}/knowledge/url`, data);
 }

 // 手工创建知识
 // data.tag_id: 可选，指定知识所属的分类ID
-export function createManualKnowledge(kbId: string, data: { title: string; content: string; status: string; tag_id?: string }) {
+export function createManualKnowledge(
+  kbId: string,
+  data: {
+    title: string
+    content: string
+    status: string
+    tag_id?: string
+    process_config?: KnowledgeProcessOverrides
+  },
+) {
  return post(`/api/v1/knowledge-bases/${kbId}/knowledge/manual`, data);
 }

@@ -215,7 +244,10 @@ export function getKnowledgeDetails(id: string, options?: { agent_id?: string })
  return get(qs ? `/api/v1/knowledge/${id}?${qs}` : `/api/v1/knowledge/${id}`);
 }

-export function updateManualKnowledge(id: string, data: { title: string; content: string; status: string }) {
+export function updateManualKnowledge(
+  id: string,
+  data: { title: string; content: string; status: string; process_config?: KnowledgeProcessOverrides },
+) {
  return put(`/api/v1/knowledge/manual/${id}`, data);
 }

@@ -6,6 +6,7 @@
      :placeholder="placeholderText"
      :disabled="disabled"
      :loading="loading"
+      :status="status"
      filterable
      style="width: 100%;"
    >
@@ -51,13 +52,15 @@ interface Props {
  selectedModelId?: string
  disabled?: boolean
  placeholder?: string
+  status?: 'default' | 'success' | 'warning' | 'error'
  // 可选：外部传入的所有模型列表，如果提供则不调用API
  allModels?: ModelConfig[]
 }

 const props = withDefaults(defineProps<Props>(), {
  disabled: false,
-  placeholder: ''
+  placeholder: '',
+  status: 'default',
 })

 const emit = defineEmits<{
@@ -0,0 +1,29 @@
+<script setup lang="ts">
+import UploadConfirmDialog from '@/views/knowledge/components/UploadConfirmDialog.vue'
+import { useUploadConfirmStore } from '@/stores/uploadConfirm'
+
+const uploadConfirmStore = useUploadConfirmStore()
+
+const handleConfirm = (payload: Parameters<typeof uploadConfirmStore.resolveConfirm>[0]) => {
+  uploadConfirmStore.resolveConfirm(payload)
+}
+
+const handleCancel = () => {
+  uploadConfirmStore.rejectConfirm()
+}
+</script>
+
+<template>
+  <UploadConfirmDialog
+    v-model:visible="uploadConfirmStore.visible"
+    :mode="uploadConfirmStore.mode"
+    :kb-info="uploadConfirmStore.kbInfo"
+    :files="uploadConfirmStore.files"
+    :urls="uploadConfirmStore.urls"
+    :manual-preview="uploadConfirmStore.manual"
+    :accept-file-types="uploadConfirmStore.acceptFileTypes"
+    :supported-file-types="uploadConfirmStore.supportedFileTypes"
+    @confirm="handleConfirm"
+    @cancel="handleCancel"
+  />
+</template>
@@ -3,7 +3,15 @@ import { ref, reactive, computed, watch, nextTick, onBeforeUnmount } from 'vue'
 import { marked } from 'marked'
 import { MessagePlugin } from 'tdesign-vue-next'
 import { useUIStore } from '@/stores/ui'
-import { listKnowledgeBases, getKnowledgeDetails, createManualKnowledge, updateManualKnowledge } from '@/api/knowledge-base'
+import {
+  listKnowledgeBases,
+  getKnowledgeDetails,
+  getKnowledgeBaseById,
+  createManualKnowledge,
+  updateManualKnowledge,
+} from '@/api/knowledge-base'
+import { useUploadConfirmStore } from '@/stores/uploadConfirm'
+import type { KnowledgeProcessOverrides } from '@/types/knowledgeProcess'
 import { sanitizeHTML, safeMarkdownToHTML } from '@/utils/security'
 import { useI18n } from 'vue-i18n'

@@ -24,6 +32,7 @@ interface KnowledgeDetailResponse {
 type ManualStatus = 'draft' | 'publish'

 const uiStore = useUIStore()
+const uploadConfirmStore = useUploadConfirmStore()
 const { t } = useI18n()

 const visible = computed({
@@ -576,11 +585,53 @@ const handleSave = async (targetStatus: ManualStatus) => {
  saving.value = true
  savingAction.value = targetStatus
  try {
-    const payload: { title: string; content: string; status: string; tag_id?: string } = {
+    const tagIdToUpload = uiStore.selectedTagId !== '__untagged__' ? uiStore.selectedTagId : undefined
+    const payload: {
+      title: string
+      content: string
+      status: string
+      tag_id?: string
+      process_config?: KnowledgeProcessOverrides
+    } = {
      title: form.title.trim(),
      content: form.content,
      status: targetStatus,
    }
+    if (tagIdToUpload) {
+      payload.tag_id = tagIdToUpload
+    }
+
+    if (targetStatus === 'publish') {
+      let kbInfo: any
+      try {
+        const kbRes: any = await getKnowledgeBaseById(form.kbId)
+        kbInfo = kbRes?.data
+      } catch {
+        MessagePlugin.error(t('manualEditor.error.fetchDetailFailed'))
+        return
+      }
+      if (!kbInfo) {
+        MessagePlugin.error(t('manualEditor.error.fetchDetailFailed'))
+        return
+      }
+      try {
+        const confirmResult = await uploadConfirmStore.open({
+          mode: 'manual',
+          kbInfo,
+          manual: {
+            kbId: form.kbId,
+            knowledgeId: knowledgeId.value || undefined,
+            title: payload.title,
+            content: payload.content,
+            tagId: tagIdToUpload,
+          },
+        })
+        payload.process_config = confirmResult.processConfig
+      } catch {
+        return
+      }
+    }
+
    let response: any
    let knowledgeID = knowledgeId.value
    let kbId = form.kbId
@@ -588,11 +639,6 @@ const handleSave = async (targetStatus: ManualStatus) => {
    if (mode.value === 'edit' && knowledgeId.value) {
      response = await updateManualKnowledge(knowledgeId.value, payload)
    } else {
-      // 创建新知识时，从 store 获取当前选中的分类ID
-      const tagIdToUpload = uiStore.selectedTagId !== '__untagged__' ? uiStore.selectedTagId : undefined
-      if (tagIdToUpload) {
-        payload.tag_id = tagIdToUpload
-      }
      response = await createManualKnowledge(form.kbId, payload)
      knowledgeID = response?.data?.id || knowledgeID
      kbId = form.kbId
@@ -636,6 +636,74 @@ export default {
    chunkLoadFailed: 'Failed to load chunks',
  },

+  uploadConfirm: {
+    title: 'Confirm Upload',
+    fileList: 'Files to upload',
+    parseConfig: 'Parse settings',
+    configNav: 'Parse settings navigation',
+    tabOverview: 'Overview',
+    overviewTitle: 'Parse settings for this batch',
+    overviewDesc: 'Click any row to edit',
+    backToOverview: 'Back to overview',
+    summaryChunkOverlapShort: 'Overlap {overlap}',
+    summaryParentChildShort: 'Parent-child',
+    summaryParserMode: 'Mode',
+    summaryParserBuiltin: 'Built-in engine (default)',
+    summaryChunkSize: 'Chunk size',
+    summaryChunkOverlap: 'Overlap',
+    summaryStrategy: 'Strategy',
+    summaryStrategyDefault: 'Default',
+    summaryParentChild: 'Parent-child',
+    summaryParentChildOn: 'On (parent {parent} / child {child})',
+    summaryParentChildOff: 'Off',
+    summaryStatus: 'Status',
+    summaryModel: 'Model',
+    summaryQuestionCount: 'Questions per chunk',
+    summaryQuestionCountValue: '{count}',
+    summaryGraphTags: 'Relation types',
+    summaryGraphTagsValue: '{count}',
+    navChunkingSummary: 'Chunk {size}',
+    statusOn: 'On',
+    statusOff: 'Off',
+    notSet: 'Not set',
+    confirm: 'Upload and parse',
+    cancel: 'Cancel',
+    uploading: 'Uploading {current}/{total}',
+    tabParser: 'Parser',
+    tabChunking: 'Chunking',
+    tabMultimodal: 'Multimodal',
+    tabAsr: 'Audio',
+    tabQuestion: 'Question generation',
+    tabGraph: 'Knowledge graph',
+    noFiles: 'Keep at least one file to upload',
+    noItems: 'Add at least one file or URL',
+    urlItemLabel: 'URL',
+    urlAdded: 'URL added',
+    urlDuplicate: 'This URL is already in the list',
+    addUrl: 'Add',
+    vlmModelRequired: 'This batch includes images. Enable multimodal and select a VLM model.',
+    multimodalRequiredForImages: 'Off (batch includes images)',
+    asrModelRequired: 'This batch includes audio. Enable ASR and select a speech recognition model.',
+    asrRequiredForAudio: 'Off (batch includes audio)',
+    vlmModelSelectRequired: 'Multimodal is enabled. Please select a VLM model.',
+    asrModelSelectRequired: 'Speech recognition is enabled. Please select an ASR model.',
+    continueAdd: 'Add more',
+    addMoreFiles: 'Add more files',
+    addMoreFolder: 'Add more from folder',
+    filesAdded: 'Added {count} file(s)',
+    filesAllDuplicate: 'Selected files are already in the list',
+    titleUrl: 'Confirm URL import',
+    titleManual: 'Confirm online publish',
+    overviewDescUrl: 'The URL will be fetched and parsed after you confirm the settings',
+    overviewDescManual: 'The document will be published and indexed after you confirm the settings',
+    confirmUrl: 'Import and parse',
+    confirmManual: 'Publish and parse',
+    urlSource: 'Import source',
+    manualSource: 'Document to publish',
+    editUrl: 'Edit URL',
+    manualCharCount: '{count} characters',
+  },
+
  knowledgeStages: {
    title: 'Processing pipeline',
    root: 'Knowledge processing',
@@ -642,6 +642,73 @@ export default {
    deleteSuccess: "지식이 성공적으로 삭제되었습니다!",
    chunkLoadFailed: "청크 로드 실패",
  },
+  uploadConfirm: {
+    title: "문서 업로드 확인",
+    fileList: "업로드할 파일",
+    parseConfig: "파싱 설정",
+    configNav: "파싱 설정 탐색",
+    tabOverview: "구성 개요",
+    overviewTitle: "이번 배치 파싱 설정",
+    overviewDesc: "항목을 클릭하여 설정을 수정하세요",
+    backToOverview: "구성 개요로 돌아가기",
+    summaryChunkOverlapShort: "겹침 {overlap}",
+    summaryParentChildShort: "부모-자식 청킹",
+    summaryParserMode: "모드",
+    summaryParserBuiltin: "내장 엔진(기본)",
+    summaryChunkSize: "청크 크기",
+    summaryChunkOverlap: "겹침",
+    summaryStrategy: "전략",
+    summaryStrategyDefault: "기본",
+    summaryParentChild: "부모-자식 청킹",
+    summaryParentChildOn: "켜짐(부모 {parent} / 자식 {child})",
+    summaryParentChildOff: "꺼짐",
+    summaryStatus: "상태",
+    summaryModel: "모델",
+    summaryQuestionCount: "청크당 질문 수",
+    summaryQuestionCountValue: "{count}개",
+    summaryGraphTags: "관계 유형",
+    summaryGraphTagsValue: "{count}개",
+    navChunkingSummary: "청크 {size}",
+    statusOn: "켜짐",
+    statusOff: "꺼짐",
+    notSet: "미설정",
+    confirm: "업로드 및 파싱 확인",
+    cancel: "취소",
+    uploading: "업로드 중 {current}/{total}",
+    tabParser: "파서",
+    tabChunking: "청킹",
+    tabMultimodal: "멀티모달",
+    tabAsr: "오디오",
+    tabQuestion: "질문 생성",
+    tabGraph: "지식 그래프",
+    noFiles: "업로드할 파일을 하나 이상 남겨 두세요",
+    noItems: "파일 또는 URL을 하나 이상 추가하세요",
+    urlItemLabel: "URL",
+    urlAdded: "URL이 추가되었습니다",
+    urlDuplicate: "이 URL은 이미 목록에 있습니다",
+    addUrl: "추가",
+    vlmModelRequired: "이 배치에 이미지가 포함되어 있습니다. 멀티모달을 활성화하고 VLM 모델을 선택하세요",
+    multimodalRequiredForImages: "꺼짐(이 배치에 이미지 포함)",
+    asrModelRequired: "이 배치에 오디오가 포함되어 있습니다. ASR을 활성화하고 음성 인식 모델을 선택하세요",
+    asrRequiredForAudio: "꺼짐(이 배치에 오디오 포함)",
+    vlmModelSelectRequired: "멀티모달이 활성화되었습니다. VLM 모델을 선택하세요",
+    asrModelSelectRequired: "음성 인식이 활성화되었습니다. ASR 모델을 선택하세요",
+    continueAdd: "계속 추가",
+    addMoreFiles: "파일 더 추가",
+    addMoreFolder: "폴더에서 더 추가",
+    filesAdded: "{count}개 파일이 추가되었습니다",
+    filesAllDuplicate: "선택한 파일이 이미 목록에 있습니다",
+    titleUrl: "URL 가져오기 확인",
+    titleManual: "온라인 편집 게시 확인",
+    overviewDescUrl: "설정 확인 후 URL을 가져와 파싱합니다",
+    overviewDescManual: "설정 확인 후 문서를 게시하고 인덱싱합니다",
+    confirmUrl: "가져오기 및 파싱",
+    confirmManual: "게시 및 파싱",
+    urlSource: "가져오기 소스",
+    manualSource: "게시할 문서",
+    editUrl: "URL 수정",
+    manualCharCount: "{count}자",
+  },
  knowledgeStages: {
    title: "처리 파이프라인",
    root: "지식 처리",
@@ -606,6 +606,73 @@ export default {
    cancelParseSubmitted: 'Разбор остановлен',
    cancelParseFailed: 'Не удалось остановить, попробуйте позже'
  },
+  uploadConfirm: {
+    title: 'Подтверждение загрузки',
+    fileList: 'Файлы для загрузки',
+    parseConfig: 'Настройки разбора',
+    configNav: 'Навигация по настройкам разбора',
+    tabOverview: 'Обзор',
+    overviewTitle: 'Настройки разбора для этой партии',
+    overviewDesc: 'Нажмите на строку, чтобы изменить',
+    backToOverview: 'Вернуться к обзору',
+    summaryChunkOverlapShort: 'Перекрытие {overlap}',
+    summaryParentChildShort: 'Родительско-дочернее',
+    summaryParserMode: 'Режим',
+    summaryParserBuiltin: 'Встроенный движок (по умолчанию)',
+    summaryChunkSize: 'Размер чанка',
+    summaryChunkOverlap: 'Перекрытие',
+    summaryStrategy: 'Стратегия',
+    summaryStrategyDefault: 'По умолчанию',
+    summaryParentChild: 'Родительско-дочернее',
+    summaryParentChildOn: 'Вкл. (родитель {parent} / дочерний {child})',
+    summaryParentChildOff: 'Выкл.',
+    summaryStatus: 'Статус',
+    summaryModel: 'Модель',
+    summaryQuestionCount: 'Вопросов на чанк',
+    summaryQuestionCountValue: '{count}',
+    summaryGraphTags: 'Типы связей',
+    summaryGraphTagsValue: '{count}',
+    navChunkingSummary: 'Чанк {size}',
+    statusOn: 'Вкл.',
+    statusOff: 'Выкл.',
+    notSet: 'Не задано',
+    confirm: 'Загрузить и обработать',
+    cancel: 'Отмена',
+    uploading: 'Загрузка {current}/{total}',
+    tabParser: 'Парсер',
+    tabChunking: 'Разбиение',
+    tabMultimodal: 'Мультимодальность',
+    tabAsr: 'Аудио',
+    tabQuestion: 'Генерация вопросов',
+    tabGraph: 'Граф знаний',
+    noFiles: 'Оставьте хотя бы один файл для загрузки',
+    noItems: 'Добавьте хотя бы один файл или URL',
+    urlItemLabel: 'URL',
+    urlAdded: 'URL добавлен',
+    urlDuplicate: 'Этот URL уже в списке',
+    addUrl: 'Добавить',
+    vlmModelRequired: 'В партии есть изображения. Включите мультимодальность и выберите модель VLM.',
+    multimodalRequiredForImages: 'Выкл. (в партии есть изображения)',
+    asrModelRequired: 'В партии есть аудио. Включите ASR и выберите модель распознавания речи.',
+    asrRequiredForAudio: 'Выкл. (в партии есть аудио)',
+    vlmModelSelectRequired: 'Мультимодальность включена. Выберите модель VLM.',
+    asrModelSelectRequired: 'Распознавание речи включено. Выберите модель ASR.',
+    continueAdd: 'Добавить ещё',
+    addMoreFiles: 'Добавить файлы',
+    addMoreFolder: 'Добавить из папки',
+    filesAdded: 'Добавлено файлов: {count}',
+    filesAllDuplicate: 'Выбранные файлы уже в списке',
+    titleUrl: 'Подтверждение импорта URL',
+    titleManual: 'Подтверждение публикации',
+    overviewDescUrl: 'После подтверждения настроек URL будет загружен и обработан',
+    overviewDescManual: 'После подтверждения настроек документ будет опубликован и проиндексирован',
+    confirmUrl: 'Импортировать и обработать',
+    confirmManual: 'Опубликовать и обработать',
+    urlSource: 'Источник импорта',
+    manualSource: 'Документ для публикации',
+    editUrl: 'Изменить URL',
+    manualCharCount: '{count} символов',
+  },
  knowledgeStages: {
    title: 'Конвейер обработки',
    root: 'Обработка знаний',
@@ -639,6 +639,73 @@ export default {
    deleteSuccess: "知识删除成功！",
    chunkLoadFailed: "分块加载失败",
  },
+  uploadConfirm: {
+    title: "上传文档确认",
+    fileList: "待上传文件",
+    parseConfig: "解析配置",
+    configNav: "解析配置导航",
+    tabOverview: "配置总览",
+    overviewTitle: "本批次解析配置",
+    overviewDesc: "点击任一项可修改配置",
+    backToOverview: "返回配置总览",
+    summaryChunkOverlapShort: "重叠 {overlap}",
+    summaryParentChildShort: "父子分块",
+    summaryParserMode: "模式",
+    summaryParserBuiltin: "内置引擎（默认）",
+    summaryChunkSize: "分块大小",
+    summaryChunkOverlap: "重叠",
+    summaryStrategy: "策略",
+    summaryStrategyDefault: "默认",
+    summaryParentChild: "父子分块",
+    summaryParentChildOn: "已开启（父 {parent} / 子 {child}）",
+    summaryParentChildOff: "未开启",
+    summaryStatus: "状态",
+    summaryModel: "模型",
+    summaryQuestionCount: "每块问题数",
+    summaryQuestionCountValue: "{count} 个",
+    summaryGraphTags: "关系类型",
+    summaryGraphTagsValue: "{count} 个",
+    navChunkingSummary: "分块 {size}",
+    statusOn: "已开启",
+    statusOff: "未开启",
+    notSet: "未设置",
+    confirm: "确认上传并解析",
+    cancel: "取消",
+    uploading: "正在上传 {current}/{total}",
+    tabParser: "解析引擎",
+    tabChunking: "分块",
+    tabMultimodal: "多模态",
+    tabAsr: "音频",
+    tabQuestion: "问题生成",
+    tabGraph: "图谱",
+    noFiles: "请至少保留一个待上传文件",
+    noItems: "请至少添加一个文件或 URL",
+    urlItemLabel: "URL",
+    urlAdded: "已添加 URL",
+    urlDuplicate: "该 URL 已在列表中",
+    addUrl: "添加",
+    vlmModelRequired: "本批次包含图片，请启用多模态并选择 VLM 模型",
+    multimodalRequiredForImages: "未开启（本批次含图片）",
+    asrModelRequired: "本批次包含音频，请启用 ASR 并选择语音识别模型",
+    asrRequiredForAudio: "未开启（本批次含音频）",
+    vlmModelSelectRequired: "已启用多模态，请选择 VLM 模型",
+    asrModelSelectRequired: "已启用语音识别，请选择 ASR 模型",
+    continueAdd: "继续添加",
+    addMoreFiles: "继续添加文件",
+    addMoreFolder: "继续添加文件夹",
+    filesAdded: "已添加 {count} 个文件",
+    filesAllDuplicate: "所选文件已在列表中",
+    titleUrl: "URL 导入确认",
+    titleManual: "在线编辑发布确认",
+    overviewDescUrl: "确认解析配置后将抓取并解析该 URL",
+    overviewDescManual: "确认解析配置后将发布并索引该文档",
+    confirmUrl: "确认导入并解析",
+    confirmManual: "确认发布并解析",
+    urlSource: "导入来源",
+    manualSource: "待发布文档",
+    editUrl: "修改 URL",
+    manualCharCount: "{count} 个字符",
+  },
  knowledgeStages: {
    title: "处理流水线",
    root: "知识处理",
@@ -0,0 +1,85 @@
+import { defineStore } from 'pinia'
+import type { KnowledgeProcessOverrides } from '@/types/knowledgeProcess'
+
+export type UploadConfirmMode = 'file' | 'url' | 'manual'
+
+export interface UploadConfirmManualSource {
+  kbId: string
+  knowledgeId?: string
+  title: string
+  content: string
+  tagId?: string
+}
+
+export interface UploadConfirmResult {
+  processConfig: KnowledgeProcessOverrides
+  mode: UploadConfirmMode
+  files?: File[]
+  urls?: string[]
+  manual?: UploadConfirmManualSource
+}
+
+export interface OpenUploadConfirmOptions {
+  mode: UploadConfirmMode
+  kbInfo: any
+  files?: File[]
+  urls?: string[]
+  manual?: UploadConfirmManualSource
+  acceptFileTypes?: string
+  supportedFileTypes?: string[]
+}
+
+export const useUploadConfirmStore = defineStore('uploadConfirm', {
+  state: () => ({
+    visible: false,
+    mode: 'file' as UploadConfirmMode,
+    kbInfo: null as any,
+    files: [] as File[],
+    urls: [] as string[],
+    manual: null as UploadConfirmManualSource | null,
+    acceptFileTypes: '',
+    supportedFileTypes: [] as string[],
+    pendingResolve: null as ((value: UploadConfirmResult) => void) | null,
+    pendingReject: null as (() => void) | null,
+  }),
+
+  actions: {
+    open(options: OpenUploadConfirmOptions): Promise<UploadConfirmResult> {
+      return new Promise((resolve, reject) => {
+        this.visible = true
+        this.mode = options.mode
+        this.kbInfo = options.kbInfo
+        this.files = options.files ? [...options.files] : []
+        this.urls = options.urls ? [...options.urls] : []
+        this.manual = options.manual || null
+        this.acceptFileTypes = options.acceptFileTypes || ''
+        this.supportedFileTypes = options.supportedFileTypes ? [...options.supportedFileTypes] : []
+        this.pendingResolve = resolve
+        this.pendingReject = reject
+      })
+    },
+
+    resolveConfirm(payload: UploadConfirmResult) {
+      this.pendingResolve?.(payload)
+      this.reset()
+    },
+
+    rejectConfirm() {
+      this.pendingReject?.()
+      this.reset()
+    },
+
+    reset() {
+      this.visible = false
+      this.mode = 'file'
+      this.kbInfo = null
+      this.files = []
+      this.urls = []
+      this.manual = null
+      this.acceptFileTypes = ''
+      this.supportedFileTypes = []
+      this.pendingResolve = null
+      this.pendingReject = null
+    },
+  },
+})
@@ -0,0 +1,66 @@
+/** Matches backend types.KnowledgeProcessOverrides (snake_case JSON). */
+
+export interface ParserEngineRule {
+  file_types: string[]
+  engine: string
+}
+
+export interface ChunkingConfigOverride {
+  chunk_size?: number
+  chunk_overlap?: number
+  separators?: string[]
+  enable_multimodal?: boolean
+  parser_engine_rules?: ParserEngineRule[]
+  enable_parent_child?: boolean
+  parent_chunk_size?: number
+  child_chunk_size?: number
+  strategy?: string
+  token_limit?: number
+  languages?: string[]
+}
+
+export interface VLMConfigOverride {
+  enabled?: boolean
+  model_id?: string
+}
+
+export interface ASRConfigOverride {
+  enabled?: boolean
+  model_id?: string
+  language?: string
+}
+
+export interface QuestionGenerationConfigOverride {
+  enabled?: boolean
+  question_count?: number
+}
+
+export interface GraphNodeOverride {
+  name: string
+  attributes?: string[]
+}
+
+export interface GraphRelationOverride {
+  node1: string
+  node2: string
+  type: string
+}
+
+export interface ExtractConfigOverride {
+  enabled?: boolean
+  text?: string
+  tags?: string[]
+  nodes?: GraphNodeOverride[]
+  relations?: GraphRelationOverride[]
+}
+
+export interface KnowledgeProcessOverrides {
+  parser_engine_rules?: ParserEngineRule[]
+  chunking_config?: ChunkingConfigOverride
+  enable_multimodel?: boolean
+  vlm_config?: VLMConfigOverride
+  asr_config?: ASRConfigOverride
+  question_generation_config?: QuestionGenerationConfigOverride
+  graph_enabled?: boolean
+  extract_config?: ExtractConfigOverride
+}
@@ -1,6 +1,6 @@
 <script setup lang="ts">
-import { ref, onMounted, onUnmounted, watch, reactive, computed, nextTick, h, type ComponentPublicInstance } from "vue";
-import { MessagePlugin, Icon as TIcon } from "tdesign-vue-next";
+import { ref, onMounted, onUnmounted, watch, reactive, computed, nextTick, type ComponentPublicInstance } from "vue";
+import { MessagePlugin } from "tdesign-vue-next";
 import DocContent from "@/components/doc-content.vue";
 import KnowledgeProcessingTimeline from "@/components/knowledge-processing-timeline.vue";
 import useKnowledgeBase from '@/hooks/useKnowledgeBase';
@@ -42,6 +42,9 @@ import { knowledgeSpansPayloadHasTrace } from '@/utils/knowledgeTrace';
 import FAQEntryManager from './components/FAQEntryManager.vue';
 import DocumentListView from './components/DocumentListView.vue';
 import DocumentBatchBar from './components/DocumentBatchBar.vue';
+import KbUploadSourceDropdown from './components/KbUploadSourceDropdown.vue';
+import type { KnowledgeProcessOverrides } from '@/types/knowledgeProcess';
+import { useUploadConfirmStore, type UploadConfirmResult } from '@/stores/uploadConfirm';
 import WikiBrowser from './wiki/WikiBrowser.vue';
 import { getWikiStats } from '@/api/wiki';
 import {
@@ -51,15 +54,14 @@ import {
 } from './wikiStatusRefresh';
 import { listMoveTargets, moveKnowledge, getKnowledgeMoveProgress } from '@/api/knowledge-base';
 import { useI18n } from 'vue-i18n';
-import { formatStringDate, kbFileTypeVerification } from '@/utils';
+import { formatStringDate } from '@/utils';
 import { formatFileSize } from '@/utils/files';
 import type { ParserEngineInfo } from '@/api/system';
 const route = useRoute();
 const { t } = useI18n();
 const kbId = computed(() => (route.params as any).kbId as string || '');
 const kbInfo = ref<any>(null);
-const uploadInputRef = ref<HTMLInputElement | null>(null);
-const folderUploadInputRef = ref<HTMLInputElement | null>(null);
+const uploadSourceRef = ref<InstanceType<typeof KbUploadSourceDropdown> | null>(null);
 const uploading = ref(false);
 const kbLoading = ref(false);
 const docListLoading = ref(true);
@@ -984,7 +986,9 @@ const handleOpenURLImportDialog = (event: CustomEvent) => {
  const eventKbId = event.detail.kbId;
  console.log('接收到URL导入对话框打开事件，知识库ID:', eventKbId, '当前知识库ID:', kbId.value);
  if (eventKbId && eventKbId === kbId.value && !isFAQ.value) {
-    urlDialogVisible.value = true;
+    if (ensureDocumentKbReady()) {
+      uploadSourceRef.value?.openUrlDialog();
+    }
  }
 };

@@ -1352,32 +1356,6 @@ const documentTitle = computed(() => {
  return t('knowledgeEditor.document.title');
 });

-// 文档操作下拉菜单选项
-const documentActionOptions = computed(() => [
-  { content: t('upload.uploadDocument'), value: 'upload', prefixIcon: () => h(TIcon, { name: 'upload', size: '16px' }) },
-  { content: t('upload.uploadFolder'), value: 'uploadFolder', prefixIcon: () => h(TIcon, { name: 'folder-add', size: '16px' }) },
-  { content: t('knowledgeBase.importURL'), value: 'importURL', prefixIcon: () => h(TIcon, { name: 'link', size: '16px' }) },
-  { content: t('upload.onlineEdit'), value: 'manualCreate', prefixIcon: () => h(TIcon, { name: 'edit', size: '16px' }) },
-]);
-
-// 处理文档操作下拉菜单选择
-const handleDocumentActionSelect = (data: { value: string }) => {
-  switch (data.value) {
-    case 'upload':
-      handleDocumentUploadClick();
-      break;
-    case 'uploadFolder':
-      handleFolderUploadClick();
-      break;
-    case 'importURL':
-      handleURLImportClick();
-      break;
-    case 'manualCreate':
-      handleManualCreate();
-      break;
-  }
-};
-
 const ensureDocumentKbReady = () => {
  if (isFAQ.value) {
    MessagePlugin.warning(t('knowledgeBase.operationNotSupportedForType'));
@@ -1406,338 +1384,147 @@ const ensureDocumentKbReady = () => {
 };


-const handleDocumentUploadClick = () => {
-  if (!ensureDocumentKbReady()) return;
-  uploadInputRef.value?.click();
+const IMAGE_EXTENSIONS = ['jpg', 'jpeg', 'png', 'gif', 'bmp', 'webp'];
+const AUDIO_EXTENSIONS = ['mp3', 'wav', 'm4a', 'flac', 'ogg'];
+
+const uploadConfirmStore = useUploadConfirmStore();
+
+const getFolderUploadFileName = (file: File) => {
+  const relativePath = (file as any).webkitRelativePath;
+  if (!relativePath) return undefined;
+  const pathParts = relativePath.split('/');
+  if (pathParts.length <= 2) return undefined;
+  const subPath = pathParts.slice(1, -1).join('/');
+  return `${subPath}/${file.name}`;
 };

-const handleFolderUploadClick = () => {
-  if (!ensureDocumentKbReady()) return;
-  folderUploadInputRef.value?.click();
-};
-
-const resetUploadInput = () => {
-  if (uploadInputRef.value) {
-    uploadInputRef.value.value = '';
-  }
-};
-
-const handleDocumentUpload = async (event: Event) => {
-  const input = event.target as HTMLInputElement;
-  const files = input?.files;
-  if (!files || files.length === 0) return;
-
-  const targetKbId = kbId.value;
-  if (!targetKbId) {
-    MessagePlugin.error(t('error.missingKbId'));
-    resetUploadInput();
-    return;
-  }
-
-  const vlmEnabled = kbInfo.value?.vlm_config?.enabled || false;
-  const asrEnabled = kbInfo.value?.asr_config?.enabled || false;
-  const dynamicTypes = supportedFileTypes.value.size > 0 ? supportedFileTypes.value : undefined
-  const validFiles: File[] = [];
-  let skippedCount = 0;
-  let imageFilteredCount = 0;
-  let videoFilteredCount = 0;
-  let audioFilteredCount = 0;
-
-  for (let i = 0; i < files.length; i++) {
-    const file = files[i];
-    const fileExt = file.name.substring(file.name.lastIndexOf('.') + 1).toLowerCase();
-    const imageTypes = ['jpg', 'jpeg', 'png', 'gif', 'bmp', 'webp'];
-    const videoTypes = ['mp4', 'mov', 'avi', 'mkv', 'webm', 'wmv', 'flv'];
-    const audioTypes = ['mp3', 'wav', 'm4a', 'flac', 'ogg'];
-
-    if (videoTypes.includes(fileExt)) {
-      videoFilteredCount++;
-      continue;
-    }
-
-    if (!vlmEnabled) {
-      if (imageTypes.includes(fileExt)) {
-        imageFilteredCount++;
-        continue;
-      }
-    }
-
-    if (!asrEnabled && audioTypes.includes(fileExt)) {
-      audioFilteredCount++;
-      continue;
-    }
-
-    if (!kbFileTypeVerification(file, files.length > 1, dynamicTypes)) {
-      validFiles.push(file);
+const showUploadResultMessages = (
+  successCount: number,
+  failCount: number,
+  totalCount: number,
+  mode: 'document' | 'folder',
+) => {
+  if (mode === 'folder') {
+    if (failCount === 0) {
+      MessagePlugin.success(t('knowledgeBase.uploadAllSuccess', { count: successCount }));
+    } else if (successCount > 0) {
+      MessagePlugin.warning(t('knowledgeBase.uploadPartialSuccess', { success: successCount, fail: failCount }));
    } else {
-      skippedCount++;
+      MessagePlugin.error(t('knowledgeBase.uploadAllFailed'));
    }
-  }
-
-  if (imageFilteredCount > 0) {
-    MessagePlugin.warning(t('knowledgeBase.imagesFilteredNoVLM', { count: imageFilteredCount }));
-  }
-  if (videoFilteredCount > 0) {
-    MessagePlugin.warning(t('knowledgeBase.videosFilteredNoVLM', { count: videoFilteredCount }));
-  }
-  if (audioFilteredCount > 0) {
-    MessagePlugin.warning(t('knowledgeBase.audiosFilteredNoASR', { count: audioFilteredCount }));
-  }
-
-  if (validFiles.length === 0) {
-    if (skippedCount > 0) {
-      MessagePlugin.warning(t('knowledgeBase.allFilesSkippedNoEngine'));
-    }
-    resetUploadInput();
    return;
  }
-  if (skippedCount > 0) {
-    MessagePlugin.warning(t('knowledgeBase.filesSkippedNoEngine', { count: skippedCount }));
-  }
-
-  let successCount = 0;
-  let failCount = 0;
-  const totalCount = validFiles.length;
-
-  // 获取当前选中的分类ID（如果不是"未分类"则传递）
-  const tagIdToUpload = selectedTagId.value !== '__untagged__' ? selectedTagId.value : undefined;
-
-  for (const file of validFiles) {
-    try {
-      const responseData: any = await uploadKnowledgeFile(targetKbId, { file, tag_id: tagIdToUpload });
-      const isSuccess = responseData?.success || responseData?.code === 200 || responseData?.status === 'success' || (!responseData?.error && responseData);
-      if (isSuccess) {
-        successCount++;
-      } else {
-        failCount++;
-        let errorMessage = t('knowledgeBase.uploadFailed');
-        if (responseData?.error?.message) {
-          errorMessage = responseData.error.message;
-        } else if (responseData?.message) {
-          errorMessage = responseData.message;
-        }
-        if (responseData?.code === 'duplicate_file' || responseData?.error?.code === 'duplicate_file') {
-          errorMessage = t('knowledgeBase.fileExists');
-        }
-        if (totalCount === 1) {
-          MessagePlugin.error(errorMessage);
-        }
-      }
-    } catch (error: any) {
-      failCount++;
-      let errorMessage = error?.error?.message || error?.message || t('knowledgeBase.uploadFailed');
-      if (error?.code === 'duplicate_file') {
-        errorMessage = t('knowledgeBase.fileExists');
-      }
-      if (totalCount === 1) {
-        MessagePlugin.error(errorMessage);
-      }
-    }
-  }
-
-  // 显示上传结果
-  if (successCount > 0) {
-    window.dispatchEvent(new CustomEvent('knowledgeFileUploaded', {
-      detail: { kbId: targetKbId }
-    }));
-  }

  if (totalCount === 1) {
    if (successCount === 1) {
      MessagePlugin.success(t('knowledgeBase.uploadSuccess'));
    }
-  } else {
-    if (failCount === 0) {
-      MessagePlugin.success(t('knowledgeBase.allUploadSuccess', { count: successCount }));
-    } else if (successCount > 0) {
-      MessagePlugin.warning(t('knowledgeBase.partialUploadSuccess', { success: successCount, fail: failCount }));
-    } else {
-      MessagePlugin.error(t('knowledgeBase.allUploadFailed', { count: failCount }));
-    }
+    return;
  }

-  resetUploadInput();
+  if (failCount === 0) {
+    MessagePlugin.success(t('knowledgeBase.allUploadSuccess', { count: successCount }));
+  } else if (successCount > 0) {
+    MessagePlugin.warning(t('knowledgeBase.partialUploadSuccess', { success: successCount, fail: failCount }));
+  } else {
+    MessagePlugin.error(t('knowledgeBase.allUploadFailed', { count: failCount }));
+  }
 };

-// 处理文件夹上传
-const handleFolderUpload = async (event: Event) => {
-  const input = event.target as HTMLInputElement;
-  const files = input?.files;
-  if (!files || files.length === 0) return;
-
+const executeUploadBatch = async (
+  files: File[],
+  options: { processConfig?: KnowledgeProcessOverrides } = {},
+) => {
  const targetKbId = kbId.value;
-  if (!targetKbId) {
-    MessagePlugin.error(t('error.missingKbId'));
-    if (input) input.value = '';
-    return;
+  if (!targetKbId || files.length === 0) {
+    return { successCount: 0, failCount: files.length };
  }

-  const vlmEnabled = kbInfo.value?.vlm_config?.enabled || false;
-  const asrEnabled = kbInfo.value?.asr_config?.enabled || false;
-  const dynamicTypes = supportedFileTypes.value.size > 0 ? supportedFileTypes.value : undefined
-
-  const validFiles: File[] = [];
-  let hiddenFileCount = 0;
-  let imageFilteredCount = 0;
-  let videoFilteredCount = 0;
-  let audioFilteredCount = 0;
-
-  for (let i = 0; i < files.length; i++) {
-    const file = files[i];
-    const relativePath = (file as any).webkitRelativePath || file.name;
-
-    const pathParts = relativePath.split('/');
-    const hasHiddenComponent = pathParts.some((part: string) => part.startsWith('.'));
-    if (hasHiddenComponent) {
-      hiddenFileCount++;
-      continue;
-    }
-
-    const fileExt = file.name.substring(file.name.lastIndexOf('.') + 1).toLowerCase();
-    const imageTypes = ['jpg', 'jpeg', 'png', 'gif', 'bmp', 'webp'];
-    const videoTypes = ['mp4', 'mov', 'avi', 'mkv', 'webm', 'wmv', 'flv'];
-    const audioTypes = ['mp3', 'wav', 'm4a', 'flac', 'ogg'];
-
-    if (videoTypes.includes(fileExt)) {
-      videoFilteredCount++;
-      continue;
-    }
-
-    if (!vlmEnabled) {
-      if (imageTypes.includes(fileExt)) {
-        imageFilteredCount++;
-        continue;
-      }
-    }
-
-    if (!asrEnabled && audioTypes.includes(fileExt)) {
-      audioFilteredCount++;
-      continue;
-    }
-
-    if (!kbFileTypeVerification(file, true, dynamicTypes)) {
-      validFiles.push(file);
-    }
-  }
-
-  if (imageFilteredCount > 0) {
-    MessagePlugin.warning(t('knowledgeBase.imagesFilteredNoVLM', { count: imageFilteredCount }));
-  }
-  if (videoFilteredCount > 0) {
-    MessagePlugin.warning(t('knowledgeBase.videosFilteredNoVLM', { count: videoFilteredCount }));
-  }
-  if (audioFilteredCount > 0) {
-    MessagePlugin.warning(t('knowledgeBase.audiosFilteredNoASR', { count: audioFilteredCount }));
-  }
-
-  if (validFiles.length === 0) {
-    MessagePlugin.warning(t('knowledgeBase.noValidFilesInFolder', { total: files.length }));
-    if (input) input.value = '';
-    return;
-  }
-  MessagePlugin.info(t('knowledgeBase.uploadingFolder', { total: validFiles.length }));
-
-  // 批量上传
+  const tagIdToUpload = selectedTagId.value !== '__untagged__' ? selectedTagId.value : undefined;
  let successCount = 0;
  let failCount = 0;
-  const tagIdToUpload = selectedTagId.value !== '__untagged__' ? selectedTagId.value : undefined;
-
-  for (const file of validFiles) {
-    const relativePath = (file as any).webkitRelativePath;
-    let fileName = file.name;
-    if (relativePath) {
-      const pathParts = relativePath.split('/');
-      if (pathParts.length > 2) {
-        const subPath = pathParts.slice(1, -1).join('/');
-        fileName = `${subPath}/${file.name}`;
-      }
-    }
+  const totalCount = files.length;
+  const hasFolderPaths = files.some((file) => {
+    const relativePath = (file as File & { webkitRelativePath?: string }).webkitRelativePath;
+    return !!relativePath && relativePath.split('/').length > 2;
+  });

+  for (const file of files) {
    try {
-      await uploadKnowledgeFile(targetKbId, { file, fileName, tag_id: tagIdToUpload });
-      successCount++;
+      const uploadData: {
+        file: File
+        tag_id?: string
+        fileName?: string
+        process_config?: KnowledgeProcessOverrides
+      } = { file, tag_id: tagIdToUpload };
+
+      const fileName = getFolderUploadFileName(file);
+      if (fileName) uploadData.fileName = fileName;
+      if (options.processConfig) {
+        uploadData.process_config = options.processConfig;
+      }
+
+      const responseData: any = await uploadKnowledgeFile(targetKbId, uploadData);
+      const isSuccess = responseData?.success || responseData?.code === 200 || responseData?.status === 'success' || (!responseData?.error && responseData);
+      if (isSuccess) {
+        successCount++;
+      } else {
+        failCount++;
+        if (totalCount === 1) {
+          let errorMessage = t('knowledgeBase.uploadFailed');
+          if (responseData?.error?.message) {
+            errorMessage = responseData.error.message;
+          } else if (responseData?.message) {
+            errorMessage = responseData.message;
+          }
+          if (responseData?.code === 'duplicate_file' || responseData?.error?.code === 'duplicate_file') {
+            errorMessage = t('knowledgeBase.fileExists');
+          }
+          MessagePlugin.error(errorMessage);
+        }
+      }
    } catch (error: any) {
      failCount++;
+      if (totalCount === 1) {
+        let errorMessage = error?.error?.message || error?.message || t('knowledgeBase.uploadFailed');
+        if (error?.code === 'duplicate_file') {
+          errorMessage = t('knowledgeBase.fileExists');
+        }
+        MessagePlugin.error(errorMessage);
+      }
    }
  }

  if (successCount > 0) {
    window.dispatchEvent(new CustomEvent('knowledgeFileUploaded', {
-      detail: { kbId: targetKbId }
+      detail: { kbId: targetKbId },
    }));
  }

-  if (failCount === 0) {
-    MessagePlugin.success(t('knowledgeBase.uploadAllSuccess', { count: successCount }));
-  } else if (successCount > 0) {
-    MessagePlugin.warning(t('knowledgeBase.uploadPartialSuccess', { success: successCount, fail: failCount }));
-  } else {
-    MessagePlugin.error(t('knowledgeBase.uploadAllFailed'));
-  }
-
-  if (input) input.value = '';
+  showUploadResultMessages(successCount, failCount, totalCount, hasFolderPaths ? 'folder' : 'document');
+  return { successCount, failCount };
 };

-const handleManualCreate = () => {
-  if (!ensureDocumentKbReady()) return;
-  uiStore.openManualEditor({
-    mode: 'create',
-    kbId: kbId.value,
-    status: 'draft',
-    onSuccess: manualEditorSuccess,
-  });
-};
-
-// URL 导入相关
-const urlDialogVisible = ref(false);
-const urlInputValue = ref('');
-const urlImporting = ref(false);
-
-const handleURLImportClick = () => {
-  if (!ensureDocumentKbReady()) return;
-  urlInputValue.value = '';
-  urlDialogVisible.value = true;
-};
-
-const handleURLImportCancel = () => {
-  urlDialogVisible.value = false;
-  urlInputValue.value = '';
-};
-
-const handleURLImportConfirm = async () => {
-  const url = urlInputValue.value.trim();
-  if (!url) {
-    MessagePlugin.warning(t('knowledgeBase.urlRequired'));
-    return;
-  }
-
-  // 简单的URL格式验证
-  try {
-    new URL(url);
-  } catch (error) {
-    MessagePlugin.warning(t('knowledgeBase.invalidURL'));
-    return;
-  }
-
+const executeUrlImport = async (url: string, processConfig?: KnowledgeProcessOverrides) => {
  const targetKbId = kbId.value;
  if (!targetKbId) {
    MessagePlugin.error(t('error.missingKbId'));
    return;
  }

-  urlImporting.value = true;
+  const tagIdToUpload = selectedTagId.value !== '__untagged__' ? selectedTagId.value : undefined;
  try {
-    // 获取当前选中的分类ID
-    const tagIdToUpload = selectedTagId.value !== '__untagged__' ? selectedTagId.value : undefined;
-    const responseData: any = await createKnowledgeFromURL(targetKbId, { url, tag_id: tagIdToUpload });
+    const responseData: any = await createKnowledgeFromURL(targetKbId, {
+      url,
+      tag_id: tagIdToUpload,
+      process_config: processConfig,
+    });
    window.dispatchEvent(new CustomEvent('knowledgeFileUploaded', {
-      detail: { kbId: targetKbId }
+      detail: { kbId: targetKbId },
    }));
    const isSuccess = responseData?.success || responseData?.code === 200 || responseData?.status === 'success' || (!responseData?.error && responseData);
    if (isSuccess) {
      MessagePlugin.success(t('knowledgeBase.urlImportSuccess'));
-      urlDialogVisible.value = false;
-      urlInputValue.value = '';
    } else {
      let errorMessage = t('knowledgeBase.urlImportFailed');
      if (responseData?.error?.message) {
@@ -1756,11 +1543,73 @@ const handleURLImportConfirm = async () => {
      errorMessage = t('knowledgeBase.urlExists');
    }
    MessagePlugin.error(errorMessage);
-  } finally {
-    urlImporting.value = false;
  }
 };

+const handleUploadConfirmResult = async (result: UploadConfirmResult) => {
+  if (result.mode === 'manual') {
+    return;
+  }
+
+  const files = result.files || [];
+  const urls = result.urls || [];
+  const processConfig = result.processConfig;
+
+  if (files.length > 0) {
+    const hasFolderPaths = files.some((file) => {
+      const relativePath = (file as File & { webkitRelativePath?: string }).webkitRelativePath;
+      return !!relativePath && relativePath.split('/').length > 2;
+    });
+    if (hasFolderPaths) {
+      MessagePlugin.info(t('knowledgeBase.uploadingFolder', { total: files.length }));
+    }
+    await executeUploadBatch(files, { processConfig });
+  }
+
+  for (const url of urls) {
+    await executeUrlImport(url, processConfig);
+  }
+};
+
+const openUploadConfirmDialog = async (files: File[], urls: string[] = []) => {
+  if (!kbInfo.value) return;
+  if (files.length === 0 && urls.length === 0) return;
+  try {
+    const result = await uploadConfirmStore.open({
+      mode: 'file',
+      kbInfo: kbInfo.value,
+      files,
+      urls,
+      acceptFileTypes: acceptFileTypes.value,
+      supportedFileTypes: [...supportedFileTypes.value],
+    });
+    await handleUploadConfirmResult(result);
+  } catch {
+    // cancelled
+  }
+};
+
+const handleUploadSourceFiles = (files: File[]) => {
+  if (!ensureDocumentKbReady()) return;
+  if (files.length === 0) return;
+  openUploadConfirmDialog(files);
+};
+
+const handleUploadSourceUrl = (url: string) => {
+  if (!ensureDocumentKbReady()) return;
+  openUploadConfirmDialog([], [url]);
+};
+
+const handleManualCreate = () => {
+  if (!ensureDocumentKbReady()) return;
+  uiStore.openManualEditor({
+    mode: 'create',
+    kbId: kbId.value,
+    status: 'draft',
+    onSuccess: manualEditorSuccess,
+  });
+};
+
 const handleOpenKBSettings = () => {
  if (!kbId.value) {
    MessagePlugin.warning(t('knowledgeEditor.messages.missingId'));
@@ -2158,11 +2007,6 @@ async function createNewSession(value: string): Promise<void> {
      </div>

      <template v-if="activeKbTab === 'documents' || !isWiki">
-        <input ref="uploadInputRef" type="file" class="document-upload-input"
-          :accept="acceptFileTypes || '.pdf,.docx,.doc,.txt,.md,.json,.jpg,.jpeg,.png,.csv,.xlsx,.xls,.pptx,.ppt,.mp3,.wav,.m4a,.flac,.ogg'"
-          multiple @change="handleDocumentUpload" />
-        <input ref="folderUploadInputRef" type="file" class="document-upload-input" webkitdirectory
-          @change="handleFolderUpload" />
        <div class="knowledge-main">
          <aside class="tag-sidebar">
            <div class="sidebar-header">
@@ -2319,15 +2163,20 @@ async function createNewSession(value: string): Promise<void> {
                  </t-tooltip>
                </div>
                <div v-if="canEdit" class="doc-filter-actions">
-                  <t-tooltip :content="$t('knowledgeBase.addDocument')" placement="top">
-                    <t-dropdown :options="documentActionOptions" trigger="click" placement="bottom-right"
-                      @click="handleDocumentActionSelect">
-                      <t-button variant="text" theme="default" class="content-bar-icon-btn" size="small"
-                        data-guide="kb-detail-add-doc">
-                        <template #icon><t-icon name="file-add" size="16px" /></template>
-                      </t-button>
-                    </t-dropdown>
-                  </t-tooltip>
+                  <KbUploadSourceDropdown
+                    ref="uploadSourceRef"
+                    :accept-file-types="acceptFileTypes"
+                    :supported-file-types="[...supportedFileTypes]"
+                    include-manual
+                    trigger-icon="file-add"
+                    trigger-class="content-bar-icon-btn"
+                    data-guide="kb-detail-add-doc"
+                    :tooltip="t('knowledgeBase.addDocument')"
+                    placement="bottom-right"
+                    @files="handleUploadSourceFiles"
+                    @url="handleUploadSourceUrl"
+                    @manual="handleManualCreate"
+                  />
                </div>
              </div>
              <div class="doc-scroll-container" :class="{ 'is-empty': !cardList.length && !docListLoading }" ref="knowledgeScroll"
@@ -2687,21 +2536,6 @@ async function createNewSession(value: string): Promise<void> {
                  @delete="confirmBatchDelete" />
              </div>
            </div>
-            <!-- URL 导入对话框 -->
-            <t-dialog v-model:visible="urlDialogVisible" :header="$t('knowledgeBase.importURLTitle')" :confirm-btn="{
-              content: $t('common.confirm'),
-              theme: 'primary',
-              loading: urlImporting,
-            }" :cancel-btn="{ content: $t('common.cancel') }" @confirm="handleURLImportConfirm"
-              @cancel="handleURLImportCancel" width="500px">
-              <div class="url-import-form">
-                <div class="url-input-label">{{ $t('knowledgeBase.urlLabel') }}</div>
-                <t-input v-model="urlInputValue" :placeholder="$t('knowledgeBase.urlPlaceholder')" clearable autofocus
-                  @enter="handleURLImportConfirm" />
-                <div class="url-input-tip">{{ $t('knowledgeBase.urlTip') }}</div>
-              </div>
-            </t-dialog>
-
          </div>
        </div>
      </template>
@@ -0,0 +1,251 @@
+<template>
+  <div class="kb-upload-source-dropdown">
+    <input
+      ref="fileInputRef"
+      type="file"
+      class="hidden-file-input"
+      multiple
+      :accept="acceptFileTypes || undefined"
+      @change="(e) => handleFilesChange(e, false)"
+    />
+    <input
+      ref="folderInputRef"
+      type="file"
+      class="hidden-file-input"
+      webkitdirectory
+      multiple
+      @change="(e) => handleFilesChange(e, true)"
+    />
+
+    <t-tooltip :content="tooltipText" placement="top">
+      <t-dropdown
+        :options="dropdownOptions"
+        trigger="click"
+        :placement="placement"
+        @click="handleActionSelect"
+      >
+        <t-button
+          variant="text"
+          theme="default"
+          :class="['kb-upload-source-trigger', triggerClass]"
+          :data-guide="dataGuide || undefined"
+          size="small"
+        >
+          <template #icon><t-icon :name="triggerIcon" size="16px" /></template>
+        </t-button>
+      </t-dropdown>
+    </t-tooltip>
+
+    <t-dialog
+      v-model:visible="urlDialogVisible"
+      :header="t('knowledgeBase.importURLTitle')"
+      :confirm-btn="{ content: t('common.confirm'), theme: 'primary' }"
+      :cancel-btn="{ content: t('common.cancel') }"
+      width="500px"
+      @confirm="handleUrlDialogConfirm"
+      @cancel="handleUrlDialogCancel"
+    >
+      <div class="url-import-form">
+        <div class="url-input-label">{{ t('knowledgeBase.urlLabel') }}</div>
+        <t-input
+          v-model="urlInputValue"
+          :placeholder="t('knowledgeBase.urlPlaceholder')"
+          clearable
+          autofocus
+          @enter="handleUrlDialogConfirm"
+        />
+        <div class="url-input-tip">{{ t('knowledgeBase.urlTip') }}</div>
+      </div>
+    </t-dialog>
+  </div>
+</template>
+
+<script setup lang="ts">
+import { ref, computed, h, withDefaults } from 'vue'
+import { useI18n } from 'vue-i18n'
+import { MessagePlugin, Icon as TIcon } from 'tdesign-vue-next'
+import { filterUploadFiles } from '../utils/uploadSources'
+
+const props = withDefaults(defineProps<{
+  acceptFileTypes?: string
+  supportedFileTypes?: string[]
+  includeManual?: boolean
+  triggerIcon?: string
+  triggerClass?: string
+  dataGuide?: string
+  tooltip?: string
+  placement?: 'top' | 'bottom' | 'bottom-right' | 'bottom-left'
+}>(), {
+  acceptFileTypes: '',
+  supportedFileTypes: () => [],
+  includeManual: false,
+  triggerIcon: 'file-add',
+  triggerClass: '',
+  dataGuide: '',
+  tooltip: '',
+  placement: 'bottom-right',
+})
+
+const emit = defineEmits<{
+  files: [files: File[]]
+  url: [url: string]
+  manual: []
+}>()
+
+const { t } = useI18n()
+
+const fileInputRef = ref<HTMLInputElement | null>(null)
+const folderInputRef = ref<HTMLInputElement | null>(null)
+const urlDialogVisible = ref(false)
+const urlInputValue = ref('')
+
+const tooltipText = computed(() => props.tooltip || t('knowledgeBase.addDocument'))
+
+const dropdownOptions = computed(() => {
+  const options = [
+    {
+      content: t('upload.uploadDocument'),
+      value: 'upload',
+      prefixIcon: () => h(TIcon, { name: 'upload', size: '16px' }),
+    },
+    {
+      content: t('upload.uploadFolder'),
+      value: 'uploadFolder',
+      prefixIcon: () => h(TIcon, { name: 'folder-add', size: '16px' }),
+    },
+    {
+      content: t('knowledgeBase.importURL'),
+      value: 'importURL',
+      prefixIcon: () => h(TIcon, { name: 'link', size: '16px' }),
+    },
+  ]
+  if (props.includeManual) {
+    options.push({
+      content: t('upload.onlineEdit'),
+      value: 'manualCreate',
+      prefixIcon: () => h(TIcon, { name: 'edit', size: '16px' }),
+    })
+  }
+  return options
+})
+
+const handleActionSelect = (data: { value: string }) => {
+  switch (data.value) {
+    case 'upload':
+      fileInputRef.value?.click()
+      break
+    case 'uploadFolder':
+      folderInputRef.value?.click()
+      break
+    case 'importURL':
+      urlInputValue.value = ''
+      urlDialogVisible.value = true
+      break
+    case 'manualCreate':
+      emit('manual')
+      break
+    default:
+      break
+  }
+}
+
+const notifyFilterResult = (result: ReturnType<typeof filterUploadFiles>, emptyAllSkippedKey: string) => {
+  const { validFiles, skippedCount, videoFilteredCount } = result
+  if (validFiles.length === 0) {
+    if (skippedCount > 0) {
+      MessagePlugin.warning(t(emptyAllSkippedKey))
+    }
+    return false
+  }
+  if (videoFilteredCount > 0) {
+    MessagePlugin.warning(t('knowledgeBase.videosFilteredNoVLM', { count: videoFilteredCount }))
+  }
+  if (skippedCount > 0) {
+    MessagePlugin.warning(t('knowledgeBase.filesSkippedNoEngine', { count: skippedCount }))
+  }
+  return true
+}
+
+const handleFilesChange = (event: Event, fromFolder: boolean) => {
+  const input = event.target as HTMLInputElement
+  const files = input.files
+  if (!files || files.length === 0) return
+
+  const result = filterUploadFiles(files, {
+    supportedFileTypes: props.supportedFileTypes,
+    fromFolder,
+    multiFile: files.length > 1,
+  })
+
+  if (!notifyFilterResult(result, 'knowledgeBase.allFilesSkippedNoEngine')) {
+    input.value = ''
+    return
+  }
+
+  emit('files', result.validFiles)
+  input.value = ''
+}
+
+const handleUrlDialogConfirm = () => {
+  const url = urlInputValue.value.trim()
+  if (!url) {
+    MessagePlugin.warning(t('knowledgeBase.urlRequired'))
+    return
+  }
+  try {
+    new URL(url)
+  } catch {
+    MessagePlugin.warning(t('knowledgeBase.invalidURL'))
+    return
+  }
+  urlDialogVisible.value = false
+  urlInputValue.value = ''
+  emit('url', url)
+}
+
+const handleUrlDialogCancel = () => {
+  urlDialogVisible.value = false
+  urlInputValue.value = ''
+}
+
+const openUrlDialog = () => {
+  urlInputValue.value = ''
+  urlDialogVisible.value = true
+}
+
+defineExpose({ openUrlDialog })
+</script>
+
+<style lang="less" scoped>
+.hidden-file-input {
+  position: absolute;
+  width: 0;
+  height: 0;
+  opacity: 0;
+  pointer-events: none;
+}
+
+.kb-upload-source-trigger {
+  color: var(--td-text-color-secondary);
+
+  &:hover {
+    color: var(--td-brand-color);
+  }
+}
+
+.url-import-form {
+  .url-input-label {
+    margin-bottom: 8px;
+    font-size: 14px;
+    font-weight: 500;
+    color: var(--td-text-color-primary);
+  }
+
+  .url-input-tip {
+    margin-top: 8px;
+    font-size: 12px;
+    line-height: 1.5;
+    color: var(--td-text-color-placeholder);
+  }
+}
+</style>
@@ -1,10 +1,9 @@
 <template>
-  <div class="graph-settings">
-    <div class="section-header">
+  <div class="graph-settings" :class="{ 'graph-settings--embedded': embedded }">
+    <div v-if="!embedded" class="section-header">
      <h2>{{ t('graphSettings.title') }}</h2>
      <p class="section-description">{{ t('graphSettings.description') }}</p>
-      
-      <!-- Warning message when graph database is not enabled -->
+
      <t-alert
        v-if="!isGraphDatabaseEnabled"
        theme="warning"
@@ -18,6 +17,15 @@
        </template>
      </t-alert>
    </div>
+    <t-alert
+      v-else-if="!isGraphDatabaseEnabled"
+      theme="warning"
+      class="embedded-graph-alert"
+    >
+      <template #message>
+        <div>{{ t('graphSettings.disabledWarning') }}</div>
+      </template>
+    </t-alert>

    <div v-if="isGraphDatabaseEnabled" class="settings-group">
      <!-- 启用实体关系提取 -->
@@ -296,7 +304,7 @@
 </template>

 <script setup lang="ts">
-import { ref, watch, onMounted, computed } from 'vue'
+import { ref, watch, onMounted, computed, withDefaults } from 'vue'
 import { MessagePlugin } from 'tdesign-vue-next'
 import { useI18n } from 'vue-i18n'
 import { extractTextRelations, fabriText, fabriTag, type Node, type Relation } from '@/api/initialization'
@@ -323,9 +331,12 @@ interface Props {
  graphExtract: GraphExtractConfig
  modelId: string
  allModels?: any[]
+  embedded?: boolean
 }

-const props = defineProps<Props>()
+const props = withDefaults(defineProps<Props>(), {
+  embedded: false,
+})

 const emit = defineEmits<{
  'update:graphExtract': [value: GraphExtractConfig]
@@ -766,4 +777,27 @@ onMounted(async () => {
  gap: 12px;
  flex-wrap: wrap;
 }
+
+.graph-settings--embedded {
+  .embedded-graph-alert {
+    margin-bottom: 12px;
+  }
+
+  .setting-row:not(.vertical) {
+    flex-direction: column;
+    align-items: stretch;
+    gap: 8px;
+    padding: 12px 0;
+  }
+
+  .setting-row:not(.vertical) .setting-info {
+    flex: none;
+    max-width: none;
+    padding-right: 0;
+  }
+
+  .setting-row:not(.vertical) .setting-control {
+    align-self: flex-start;
+  }
+}
 </style>
@@ -1,6 +1,6 @@
 <template>
-  <div class="kb-advanced-settings">
-    <div class="section-header">
+  <div class="kb-advanced-settings" :class="{ 'kb-advanced-settings--embedded': embedded }">
+    <div v-if="!embedded" class="section-header">
      <h2>{{ $t('knowledgeEditor.advanced.title') }}</h2>
      <p class="section-description">{{ $t('knowledgeEditor.advanced.description') }}</p>
    </div>
@@ -49,7 +49,7 @@
 </template>

 <script setup lang="ts">
-import { ref, watch } from 'vue'
+import { ref, watch, withDefaults } from 'vue'

 interface QuestionGenerationConfig {
  enabled: boolean
@@ -60,9 +60,12 @@ interface Props {
  questionGeneration?: QuestionGenerationConfig
  ragEnabled?: boolean
  allModels?: any[]
+  embedded?: boolean
 }

-const props = defineProps<Props>()
+const props = withDefaults(defineProps<Props>(), {
+  embedded: false,
+})

 const emit = defineEmits<{
  'update:questionGeneration': [value: QuestionGenerationConfig]
@@ -182,4 +185,51 @@ const handleQuestionGenerationChange = () => {
  font-weight: 500;
 }

+.kb-advanced-settings--embedded {
+  .setting-row {
+    padding: 12px 0;
+  }
+
+  .setting-row:has(.t-switch) {
+    flex-direction: row;
+    align-items: center;
+    gap: 16px;
+
+    .setting-info {
+      flex: 1;
+      min-width: 0;
+      max-width: none;
+      padding-right: 0;
+    }
+
+    .setting-control {
+      flex: none;
+      align-self: center;
+    }
+  }
+
+  .subsection .setting-row {
+    flex-direction: column;
+    align-items: stretch;
+    gap: 8px;
+
+    .setting-info {
+      flex: none;
+      max-width: none;
+      padding-right: 0;
+    }
+
+    .setting-control {
+      align-self: flex-start;
+    }
+  }
+
+  .subsection {
+    margin-top: 0;
+    padding: 0;
+    border: none;
+    background: none;
+  }
+}
+
 </style>
@@ -1,6 +1,6 @@
 <template>
-  <div class="kb-chunking-settings">
-    <div class="section-header">
+  <div class="kb-chunking-settings" :class="{ 'kb-chunking-settings--embedded': embedded }">
+    <div v-if="!embedded" class="section-header">
      <div class="section-header-text">
        <h2>{{ $t('knowledgeEditor.chunking.title') }}</h2>
        <p class="section-description">{{ $t('knowledgeEditor.chunking.description') }}</p>
@@ -21,12 +21,12 @@
            :placeholder="$t('knowledgeEditor.chunking.strategyPlaceholder')"
            :clearable="true"
            @change="handleStrategyChange"
-            style="width: 280px;"
+            :style="selectStyle"
          />
          <!-- Test trigger sits right next to the strategy picker so users
               discover it exactly when they're deciding which strategy to
               use on their content. -->
-          <KBChunkingDebug :config="debugConfig" />
+          <KBChunkingDebug v-if="!embedded" :config="debugConfig" />
        </div>
      </div>

@@ -51,9 +51,9 @@
              :min="100"
              :max="4000"
              :step="50"
-              :marks="{ 100: '100', 1000: '1000', 2000: '2000', 4000: '4000' }"
+              :marks="embedded ? undefined : chunkSizeMarks"
              @change="handleChunkSizeChange"
-              style="width: 200px;"
+              :style="sliderStyle"
            />
            <span class="value-display">{{ localChunkSize }} {{ $t('knowledgeEditor.chunking.characters') }}</span>
          </div>
@@ -74,9 +74,9 @@
              :min="0"
              :max="500"
              :step="20"
-              :marks="{ 0: '0', 250: '250', 500: '500' }"
+              :marks="embedded ? undefined : chunkOverlapMarks"
              @change="handleChunkOverlapChange"
-              style="width: 200px;"
+              :style="sliderStyle"
            />
            <span class="value-display">{{ localChunkOverlap }} {{ $t('knowledgeEditor.chunking.characters') }}</span>
          </div>
@@ -84,7 +84,7 @@
      </div>

      <!-- Separators -->
-      <div class="setting-row">
+      <div class="setting-row setting-row--separators">
        <div class="setting-info">
          <label>{{ $t('knowledgeEditor.chunking.separatorsLabel') }}</label>
          <p class="desc">{{ $t('knowledgeEditor.chunking.separatorsDescription') }}</p>
@@ -98,13 +98,13 @@
            filterable
            :placeholder="$t('knowledgeEditor.chunking.separatorsPlaceholder')"
            @change="handleSeparatorsChange"
-            style="width: 280px;"
+            :style="selectStyle"
          />
        </div>
      </div>

      <!-- Parent-Child Chunking -->
-      <div class="setting-row">
+      <div class="setting-row setting-row--toggle">
        <div class="setting-info">
          <label>{{ $t('knowledgeEditor.chunking.parentChildLabel') }}</label>
          <p class="desc">{{ $t('knowledgeEditor.chunking.parentChildDescription') }}</p>
@@ -130,9 +130,9 @@
              :min="512"
              :max="8192"
              :step="64"
-              :marks="{ 512: '512', 2048: '2048', 4096: '4096', 8192: '8192' }"
+              :marks="embedded ? undefined : parentChunkSizeMarks"
              @change="handleParentChunkSizeChange"
-              style="width: 200px;"
+              :style="sliderStyle"
            />
            <span class="value-display">{{ localParentChunkSize }} {{ $t('knowledgeEditor.chunking.characters') }}</span>
          </div>
@@ -152,9 +152,9 @@
              :min="64"
              :max="2048"
              :step="32"
-              :marks="{ 64: '64', 384: '384', 1024: '1024', 2048: '2048' }"
+              :marks="embedded ? undefined : childChunkSizeMarks"
              @change="handleChildChunkSizeChange"
-              style="width: 200px;"
+              :style="sliderStyle"
            />
            <span class="value-display">{{ localChildChunkSize }} {{ $t('knowledgeEditor.chunking.characters') }}</span>
          </div>
@@ -201,7 +201,7 @@
              :disabled="advancedDisabled"
              :placeholder="$t('knowledgeEditor.chunking.languagesPlaceholder')"
              @change="handleLanguagesChange"
-              style="width: 280px;"
+              :style="selectStyle"
            />
          </div>
        </div>
@@ -212,7 +212,7 @@
 </template>

 <script setup lang="ts">
-import { ref, watch, computed } from 'vue'
+import { ref, watch, computed, withDefaults } from 'vue'
 import { useI18n } from 'vue-i18n'
 import { ChevronRightIcon } from 'tdesign-icons-vue-next'
 import KBChunkingDebug from './KBChunkingDebug.vue'
@@ -253,9 +253,20 @@ interface ChunkingConfig {

 interface Props {
  config: ChunkingConfig
+  embedded?: boolean
 }

-const props = defineProps<Props>()
+const props = withDefaults(defineProps<Props>(), {
+  embedded: false,
+})
+
+const selectStyle = computed(() => (props.embedded ? { width: '100%' } : { width: '280px' }))
+const sliderStyle = computed(() => (props.embedded ? { width: '100%' } : { width: '200px' }))
+
+const chunkSizeMarks = { 100: '100', 1000: '1000', 2000: '2000', 4000: '4000' }
+const chunkOverlapMarks = { 0: '0', 250: '250', 500: '500' }
+const parentChunkSizeMarks = { 512: '512', 2048: '2048', 4096: '4096', 8192: '8192' }
+const childChunkSizeMarks = { 64: '64', 384: '384', 1024: '1024', 2048: '2048' }

 const emit = defineEmits<{
  'update:config': [value: ChunkingConfig]
@@ -563,4 +574,102 @@ const emitUpdate = () => {
  // panel edge and looks detached from the rest of the form.
  margin-top: 4px;
 }
+
+.kb-chunking-settings--embedded {
+  .setting-row {
+    flex-direction: column;
+    align-items: stretch;
+    gap: 10px;
+    padding: 14px 0;
+  }
+
+  .setting-row--toggle {
+    flex-direction: row;
+    align-items: center;
+    gap: 16px;
+
+    .setting-info {
+      flex: 1;
+      min-width: 0;
+    }
+
+    .setting-control {
+      flex: none;
+      width: auto;
+      justify-content: flex-end;
+    }
+  }
+
+  .setting-row--separators {
+    padding-bottom: 18px;
+  }
+
+  .setting-info {
+    flex: none;
+    max-width: none;
+    padding-right: 0;
+
+    label {
+      font-size: 14px;
+    }
+
+    .desc {
+      font-size: 12px;
+    }
+  }
+
+  .setting-control {
+    flex: none;
+    max-width: none;
+    justify-content: flex-start;
+    align-items: stretch;
+    width: 100%;
+  }
+
+  .strategy-control {
+    flex-direction: column;
+    align-items: stretch;
+    width: 100%;
+  }
+
+  .strategy-info-panel {
+    margin: -4px 0 10px;
+    padding: 8px 12px;
+    font-size: 12px;
+  }
+
+  .slider-container {
+    flex-direction: column;
+    align-items: stretch;
+    gap: 8px;
+    width: 100%;
+    justify-content: flex-start;
+  }
+
+  .value-display {
+    order: -1;
+    text-align: left;
+    min-width: 0;
+  }
+
+  :deep(.t-slider) {
+    padding-bottom: 0;
+  }
+
+  :deep(.t-select__wrap) {
+    max-width: 100%;
+  }
+
+  :deep(.t-tag) {
+    max-width: 100%;
+  }
+
+  .advanced-toggle {
+    padding-top: 10px;
+  }
+
+  .advanced-section .setting-row {
+    padding: 14px 0;
+  }
+}
 </style>
@@ -1,6 +1,6 @@
 <template>
-  <div class="kb-parser-settings">
-    <div class="section-header">
+  <div class="kb-parser-settings" :class="{ 'kb-parser-settings--embedded': embedded }">
+    <div v-if="!embedded" class="section-header">
      <h2>{{ $t('kbSettings.parser.title') }}</h2>
      <p class="section-description">{{ $t('kbSettings.parser.description') }}</p>
    </div>
@@ -33,7 +33,7 @@
          <t-select
            :value="getEngineForGroup(group.extensions) || undefined"
            @change="(val: string) => handleEngineChange(group.extensions, val)"
-            style="width: 280px;"
+            :style="embedded ? { width: '100%' } : { width: '280px' }"
            :status="hasAvailableEngine(group.extensions) ? 'default' : 'warning'"
            :placeholder="$t('kbSettings.parser.noEngine')"
          >
@@ -123,10 +123,16 @@ interface EngineOption {

 interface Props {
  parserEngineRules?: ParserEngineRule[]
+  /** Compact layout for upload-confirm dialog */
+  embedded?: boolean
+  /** When set, only show file-type groups matching these extensions */
+  relevantExtensions?: string[]
 }

 const props = withDefaults(defineProps<Props>(), {
-  parserEngineRules: () => []
+  parserEngineRules: () => [],
+  embedded: false,
+  relevantExtensions: () => [],
 })

 const emit = defineEmits<{
@@ -182,7 +188,11 @@ const fileTypeGroups = computed(() => {
    })
  }

-  return groups
+  const rel = props.relevantExtensions
+  if (!rel?.length) return groups
+  const relSet = new Set(rel)
+  const filtered = groups.filter(g => g.extensions.some(e => relSet.has(e)))
+  return filtered.length > 0 ? filtered : groups
 })

 function getEngineOptions(extensions: string[]): EngineOption[] {
@@ -468,6 +478,46 @@ watch(() => props.parserEngineRules, (v) => {
    }
  }
 }
+
+.kb-parser-settings--embedded {
+  .setting-row {
+    flex-direction: column;
+    align-items: stretch;
+    gap: 10px;
+    padding: 12px 0;
+  }
+
+  .setting-info {
+    flex: none;
+    max-width: none;
+    padding-right: 0;
+    display: flex;
+    flex-wrap: wrap;
+    align-items: center;
+    gap: 6px 10px;
+  }
+
+  .setting-control {
+    flex: none;
+    max-width: none;
+    align-items: stretch;
+  }
+
+  .group-label {
+    font-size: 14px;
+    margin-bottom: 0;
+  }
+
+  .ext-tags {
+    margin-top: 0;
+    gap: 4px;
+  }
+
+  .ext-tag {
+    font-size: 11px;
+    padding: 2px 6px;
+  }
+}
 </style>

 <style lang="less">
@@ -0,0 +1,70 @@
+import { kbFileTypeVerification } from '@/utils'
+
+export const UPLOAD_VIDEO_EXTENSIONS = ['mp4', 'mov', 'avi', 'mkv', 'webm', 'wmv', 'flv']
+
+export function getUploadFileExt(file: File): string {
+  const dot = file.name.lastIndexOf('.')
+  if (dot < 0) return ''
+  return file.name.substring(dot + 1).toLowerCase()
+}
+
+export function getUploadFileKey(file: File): string {
+  const path = (file as File & { webkitRelativePath?: string }).webkitRelativePath || ''
+  return `${path || file.name}\0${file.size}`
+}
+
+export interface FilterUploadFilesOptions {
+  supportedFileTypes?: Set<string> | string[]
+  fromFolder?: boolean
+  multiFile?: boolean
+}
+
+export interface FilterUploadFilesResult {
+  validFiles: File[]
+  skippedCount: number
+  videoFilteredCount: number
+  hiddenFileCount: number
+}
+
+export function filterUploadFiles(
+  files: FileList | File[],
+  options: FilterUploadFilesOptions = {},
+): FilterUploadFilesResult {
+  const list = Array.from(files)
+  const dynamicTypes = options.supportedFileTypes
+    ? options.supportedFileTypes instanceof Set
+      ? options.supportedFileTypes
+      : new Set(options.supportedFileTypes)
+    : undefined
+
+  const validFiles: File[] = []
+  let skippedCount = 0
+  let videoFilteredCount = 0
+  let hiddenFileCount = 0
+  const multiFile = options.multiFile ?? list.length > 1
+
+  for (const file of list) {
+    if (options.fromFolder) {
+      const relativePath = (file as File & { webkitRelativePath?: string }).webkitRelativePath || file.name
+      if (relativePath.split('/').some(part => part.startsWith('.'))) {
+        hiddenFileCount++
+        continue
+      }
+    }
+
+    const fileExt = getUploadFileExt(file)
+    if (UPLOAD_VIDEO_EXTENSIONS.includes(fileExt)) {
+      videoFilteredCount++
+      continue
+    }
+
+    if (kbFileTypeVerification(file, multiFile, dynamicTypes)) {
+      skippedCount++
+      continue
+    }
+
+    validFiles.push(file)
+  }
+
+  return { validFiles, skippedCount, videoFilteredCount, hiddenFileCount }
+}
@@ -839,6 +839,7 @@ func (s *DataSourceService) ingestItem(ctx context.Context, ds *types.DataSource
 			item.FileName, // customFileName — must include extension for file-type validation
 			tagID,         // auto-tag from data source
 			channel,
+			nil,
 		)
 		return isUpdate, err
 	}
@@ -855,6 +856,7 @@ func (s *DataSourceService) ingestItem(ctx context.Context, ds *types.DataSource
 			item.Title,
 			tagID, // auto-tag from data source
 			channel,
+			nil,
 		)
 		return isUpdate, err
 	}
@@ -290,10 +290,22 @@ func (s *ChunkExtractService) Handle(ctx context.Context, t *asynq.Task) error {
 		handleErr = err
 		return err
 	}
-	if kb.ExtractConfig == nil {
-		logger.Warnf(ctx, "failed to get extract config")
-		graphOut["skipped"] = "no_extract_config"
-		return err
+
+	var processOverrides *types.KnowledgeProcessOverrides
+	knowledgeID := p.KnowledgeID
+	if knowledgeID == "" {
+		knowledgeID = chunk.KnowledgeID
+	}
+	if knowledgeID != "" && s.knowledgeRepo != nil {
+		if k, kerr := s.knowledgeRepo.GetKnowledgeByIDOnly(ctx, knowledgeID); kerr == nil && k != nil {
+			processOverrides, _ = k.ProcessOverrides()
+		}
+	}
+	extractCfg := ResolveProcessConfig(kb, processOverrides).ExtractConfig
+	if !extractCfg.Enabled {
+		logger.Warnf(ctx, "extract config not enabled")
+		graphOut["skipped"] = "extract_disabled"
+		return nil
 	}

 	chatModel, err := s.modelService.GetChatModel(ctx, p.ModelID)
@@ -305,12 +317,12 @@ func (s *ChunkExtractService) Handle(ctx context.Context, t *asynq.Task) error {

 	template := &types.PromptTemplateStructured{
 		Description: s.template.Description,
-		Tags:        kb.ExtractConfig.Tags,
+		Tags:        extractCfg.Tags,
 		Examples: []types.GraphData{
 			{
-				Text:     kb.ExtractConfig.Text,
-				Node:     kb.ExtractConfig.Nodes,
-				Relation: kb.ExtractConfig.Relations,
+				Text:     extractCfg.Text,
+				Node:     extractCfg.Nodes,
+				Relation: extractCfg.Relations,
 			},
 		},
 	}
@@ -201,7 +201,7 @@ func (s *ImageMultimodalService) Handle(ctx context.Context, task *asynq.Task) e
 		}
 	}()

-	vlmModel, err := s.resolveVLM(ctx, payload.KnowledgeBaseID)
+	vlmModel, vlmCfg, err := s.resolveVLM(ctx, payload.KnowledgeBaseID, payload.KnowledgeID)
 	if err != nil {
 		handleErr = fmt.Errorf("resolve VLM: %w", err)
 		return handleErr
@@ -210,12 +210,10 @@ func (s *ImageMultimodalService) Handle(ctx context.Context, task *asynq.Task) e
 	// legacy inline-config path) so the trace shows WHICH model handled
 	// this image. Without this, debugging "VLM is slow" requires a
 	// separate hop to the KB config.
-	if kb, kbErr := s.kbService.GetKnowledgeBaseByIDOnly(ctx, payload.KnowledgeBaseID); kbErr == nil && kb != nil {
-		if id := strings.TrimSpace(kb.VLMConfig.ModelID); id != "" {
-			imgOut["vlm_model_id"] = id
-		} else {
-			imgOut["vlm_model_id"] = "legacy_inline"
-		}
+	if id := strings.TrimSpace(vlmCfg.ModelID); id != "" {
+		imgOut["vlm_model_id"] = id
+	} else {
+		imgOut["vlm_model_id"] = "legacy_inline"
 	}

 	// Read image bytes. A provider:// URL must be resolved via FileService —
@@ -456,27 +454,36 @@ func (s *ImageMultimodalService) indexChunks(ctx context.Context, payload types.

 // resolveVLM creates a vlm.VLM instance for the given knowledge base,
 // supporting both new-style (ModelID) and legacy (inline BaseURL) configs.
-func (s *ImageMultimodalService) resolveVLM(ctx context.Context, kbID string) (vlm.VLM, error) {
+// Per-upload process_overrides on the knowledge entry take precedence over KB defaults.
+func (s *ImageMultimodalService) resolveVLM(ctx context.Context, kbID, knowledgeID string) (vlm.VLM, types.VLMConfig, error) {
 	kb, err := s.kbService.GetKnowledgeBaseByIDOnly(ctx, kbID)
 	if err != nil {
-		return nil, fmt.Errorf("get knowledge base %s: %w", kbID, err)
+		return nil, types.VLMConfig{}, fmt.Errorf("get knowledge base %s: %w", kbID, err)
 	}
 	if kb == nil {
-		return nil, fmt.Errorf("knowledge base %s not found", kbID)
+		return nil, types.VLMConfig{}, fmt.Errorf("knowledge base %s not found", kbID)
 	}

-	vlmCfg := kb.VLMConfig
+	var processOverrides *types.KnowledgeProcessOverrides
+	if knowledgeID != "" && s.knowledgeRepo != nil {
+		if k, kerr := s.knowledgeRepo.GetKnowledgeByIDOnly(ctx, knowledgeID); kerr == nil && k != nil {
+			processOverrides, _ = k.ProcessOverrides()
+		}
+	}
+	vlmCfg := ResolveProcessConfig(kb, processOverrides).VLMConfig
 	if !vlmCfg.IsEnabled() {
-		return nil, fmt.Errorf("VLM is not enabled for knowledge base %s", kbID)
+		return nil, types.VLMConfig{}, fmt.Errorf("VLM is not enabled for knowledge base %s", kbID)
 	}

 	// New-style: resolve model through ModelService
 	if vlmCfg.ModelID != "" {
-		return s.modelService.GetVLMModel(ctx, vlmCfg.ModelID)
+		model, err := s.modelService.GetVLMModel(ctx, vlmCfg.ModelID)
+		return model, vlmCfg, err
 	}

 	// Legacy: create VLM from inline config
-	return vlm.NewVLMFromLegacyConfig(vlmCfg, s.ollamaService)
+	model, err := vlm.NewVLMFromLegacyConfig(vlmCfg, s.ollamaService)
+	return model, vlmCfg, err
 }

 // resolveFileServiceForPayload resolves tenant/KB scoped file service for reading provider:// URLs.
@@ -26,6 +26,7 @@ import (
 // CreateKnowledgeFromFile creates a knowledge entry from an uploaded file
 func (s *knowledgeService) CreateKnowledgeFromFile(ctx context.Context,
 	kbID string, file *multipart.FileHeader, metadata map[string]string, enableMultimodel *bool, customFileName string, tagID string, channel string,
+	processOverrides *types.KnowledgeProcessOverrides,
 ) (*types.Knowledge, error) {
 	logger.Info(ctx, "Start creating knowledge from file")

@@ -60,62 +61,6 @@ func (s *knowledgeService) CreateKnowledgeFromFile(ctx context.Context,
 		return nil, err
 	}

-	// 检查多模态配置完整性 - 只在图片文件时校验
-	if !IsImageType(getFileType(fileName)) {
-		logger.Info(ctx, "Non-image file with multimodal enabled, skipping COS/VLM validation")
-	} else {
-		// 解析有效 provider：优先 KB 级别（新字段 > 旧字段），其次租户默认
-		provider := kb.GetStorageProvider()
-		tenant, _ := ctx.Value(types.TenantInfoContextKey).(*types.Tenant)
-		if provider == "" && tenant != nil && tenant.StorageEngineConfig != nil {
-			provider = strings.ToLower(strings.TrimSpace(tenant.StorageEngineConfig.DefaultProvider))
-		}
-
-		// 根据 provider 校验租户级存储引擎配置
-		switch provider {
-		case "cos":
-			if tenant == nil || tenant.StorageEngineConfig == nil || tenant.StorageEngineConfig.COS == nil ||
-				tenant.StorageEngineConfig.COS.SecretID == "" || tenant.StorageEngineConfig.COS.SecretKey == "" ||
-				tenant.StorageEngineConfig.COS.Region == "" || tenant.StorageEngineConfig.COS.BucketName == "" {
-				logger.Error(ctx, "COS configuration incomplete for image multimodal processing")
-				return nil, werrors.NewBadRequestError("上传图片文件需要完整的对象存储配置信息, 请前往知识库存储设置或系统设置页面进行补全")
-			}
-		case "minio":
-			ok := false
-			if tenant != nil && tenant.StorageEngineConfig != nil && tenant.StorageEngineConfig.MinIO != nil {
-				m := tenant.StorageEngineConfig.MinIO
-				if m.Mode == "remote" {
-					ok = m.Endpoint != "" && m.AccessKeyID != "" && m.SecretAccessKey != "" && m.BucketName != ""
-				} else {
-					ok = os.Getenv("MINIO_ENDPOINT") != "" && os.Getenv("MINIO_ACCESS_KEY_ID") != "" &&
-						os.Getenv("MINIO_SECRET_ACCESS_KEY") != "" &&
-						(m.BucketName != "" || os.Getenv("MINIO_BUCKET_NAME") != "")
-				}
-			}
-			if !ok {
-				logger.Error(ctx, "MinIO configuration incomplete for image multimodal processing")
-				return nil, werrors.NewBadRequestError("上传图片文件需要完整的对象存储配置信息, 请前往知识库存储设置或系统设置页面进行补全")
-			}
-		}
-
-		// 检查VLM配置
-		if !kb.VLMConfig.Enabled || kb.VLMConfig.ModelID == "" {
-			logger.Error(ctx, "VLM model is not configured")
-			return nil, werrors.NewBadRequestError("上传图片文件需要设置VLM模型")
-		}
-
-		logger.Info(ctx, "Image multimodal configuration validation passed")
-	}
-
-	// 检查音频ASR配置完整性 - 只在音频文件时校验
-	if IsAudioType(getFileType(fileName)) {
-		if !kb.ASRConfig.IsASREnabled() {
-			logger.Error(ctx, "ASR model is not configured")
-			return nil, werrors.NewBadRequestError("上传音频文件需要设置ASR语音识别模型")
-		}
-		logger.Info(ctx, "Audio ASR configuration validation passed")
-	}
-
 	// Validate file type
 	logger.Infof(ctx, "Checking file type: %s", fileName)
 	if !isValidFileType(fileName) {
@@ -179,6 +124,64 @@ func (s *knowledgeService) CreateKnowledgeFromFile(ctx context.Context,
 		return nil, werrors.NewValidationError("文件名包含非法字符")
 	}

+	eff := ResolveProcessConfig(kb, processOverrides)
+	if enableMultimodel != nil && (processOverrides == nil || processOverrides.EnableMultimodel == nil) {
+		eff.EnableMultimodel = *enableMultimodel
+	}
+
+	if processOverrides != nil {
+		if err := ValidateProcessOverrides(ctx, kb, processOverrides, []string{getFileType(safeFilename)}); err != nil {
+			return nil, err
+		}
+	} else {
+		// 检查多模态配置完整性 - 只在图片文件时校验
+		if IsImageType(getFileType(safeFilename)) {
+			provider := kb.GetStorageProvider()
+			tenant, _ := ctx.Value(types.TenantInfoContextKey).(*types.Tenant)
+			if provider == "" && tenant != nil && tenant.StorageEngineConfig != nil {
+				provider = strings.ToLower(strings.TrimSpace(tenant.StorageEngineConfig.DefaultProvider))
+			}
+
+			switch provider {
+			case "cos":
+				if tenant == nil || tenant.StorageEngineConfig == nil || tenant.StorageEngineConfig.COS == nil ||
+					tenant.StorageEngineConfig.COS.SecretID == "" || tenant.StorageEngineConfig.COS.SecretKey == "" ||
+					tenant.StorageEngineConfig.COS.Region == "" || tenant.StorageEngineConfig.COS.BucketName == "" {
+					logger.Error(ctx, "COS configuration incomplete for image multimodal processing")
+					return nil, werrors.NewBadRequestError("上传图片文件需要完整的对象存储配置信息, 请前往知识库存储设置或系统设置页面进行补全")
+				}
+			case "minio":
+				ok := false
+				if tenant != nil && tenant.StorageEngineConfig != nil && tenant.StorageEngineConfig.MinIO != nil {
+					m := tenant.StorageEngineConfig.MinIO
+					if m.Mode == "remote" {
+						ok = m.Endpoint != "" && m.AccessKeyID != "" && m.SecretAccessKey != "" && m.BucketName != ""
+					} else {
+						ok = os.Getenv("MINIO_ENDPOINT") != "" && os.Getenv("MINIO_ACCESS_KEY_ID") != "" &&
+							os.Getenv("MINIO_SECRET_ACCESS_KEY") != "" &&
+							(m.BucketName != "" || os.Getenv("MINIO_BUCKET_NAME") != "")
+					}
+				}
+				if !ok {
+					logger.Error(ctx, "MinIO configuration incomplete for image multimodal processing")
+					return nil, werrors.NewBadRequestError("上传图片文件需要完整的对象存储配置信息, 请前往知识库存储设置或系统设置页面进行补全")
+				}
+			}
+
+			if !kb.VLMConfig.Enabled || kb.VLMConfig.ModelID == "" {
+				logger.Error(ctx, "VLM model is not configured")
+				return nil, werrors.NewBadRequestError("上传图片文件需要设置VLM模型")
+			}
+		}
+
+		if IsAudioType(getFileType(safeFilename)) {
+			if !kb.ASRConfig.IsASREnabled() {
+				logger.Error(ctx, "ASR model is not configured")
+				return nil, werrors.NewBadRequestError("上传音频文件需要设置ASR语音识别模型")
+			}
+		}
+	}
+
 	// Prepare knowledge record
 	logger.Info(ctx, "Preparing knowledge record")
 	knowledge := &types.Knowledge{
@@ -201,6 +204,13 @@ func (s *knowledgeService) CreateKnowledgeFromFile(ctx context.Context,
 		Metadata:         metadataJSON,
 	}

+	if processOverrides != nil {
+		if err := knowledge.SetProcessOverrides(processOverrides); err != nil {
+			logger.Errorf(ctx, "Failed to set process overrides: %v", err)
+			return nil, err
+		}
+	}
+
 	// Save the file to storage (use KB-level storage engine if configured)
 	logger.Infof(ctx, "Saving file, knowledge ID: %s", knowledge.ID)
 	fileSvc := s.resolveFileService(ctx, kb)
@@ -223,21 +233,12 @@ func (s *knowledgeService) CreateKnowledgeFromFile(ctx context.Context,

 	// Enqueue document processing task to Asynq
 	logger.Info(ctx, "Enqueuing document processing task to Asynq")
-	enableMultimodelValue := false
-	if enableMultimodel != nil {
-		enableMultimodelValue = *enableMultimodel
-	} else {
-		enableMultimodelValue = kb.IsMultimodalEnabled()
-	}
+	enableMultimodelValue := eff.EnableMultimodel

-	// Check question generation config
-	enableQuestionGeneration := false
-	questionCount := 3 // default
-	if kb.QuestionGenerationConfig != nil && kb.QuestionGenerationConfig.Enabled {
-		enableQuestionGeneration = true
-		if kb.QuestionGenerationConfig.QuestionCount > 0 {
-			questionCount = kb.QuestionGenerationConfig.QuestionCount
-		}
+	enableQuestionGeneration := eff.QuestionGenerationConfig.Enabled
+	questionCount := eff.QuestionGenerationConfig.QuestionCount
+	if questionCount <= 0 {
+		questionCount = 3
 	}

 	lang, _ := types.LanguageFromContext(ctx)
@@ -307,13 +308,16 @@ func isFileURL(rawURL, fileName, fileType string) bool {

 func (s *knowledgeService) CreateKnowledgeFromURL(ctx context.Context,
 	kbID string, rawURL string, fileName string, fileType string, enableMultimodel *bool, title string, tagID string, channel string,
+	processOverrides *types.KnowledgeProcessOverrides,
 ) (*types.Knowledge, error) {
 	logger.Info(ctx, "Start creating knowledge from URL")
 	logger.Infof(ctx, "Knowledge base ID: %s, URL: %s", kbID, rawURL)

 	// Route to file_url logic when the URL points to a downloadable file
 	if isFileURL(rawURL, fileName, fileType) {
-		return s.createKnowledgeFromFileURL(ctx, kbID, rawURL, fileName, fileType, enableMultimodel, title, tagID, channel)
+		return s.createKnowledgeFromFileURL(
+			ctx, kbID, rawURL, fileName, fileType, enableMultimodel, title, tagID, channel, processOverrides,
+		)
 	}

 	url := rawURL
@@ -397,6 +401,11 @@ func (s *knowledgeService) CreateKnowledgeFromURL(ctx context.Context,

 	// Save knowledge record
 	logger.Infof(ctx, "Saving knowledge record to database, ID: %s", knowledge.ID)
+	eff, err := ApplyKnowledgeProcessOverrides(ctx, kb, knowledge, processOverrides, []string{"html"}, enableMultimodel)
+	if err != nil {
+		return nil, err
+	}
+
 	if err := s.repo.CreateKnowledge(ctx, knowledge); err != nil {
 		logger.Errorf(ctx, "Failed to create knowledge record: %v", err)
 		return nil, err
@@ -404,21 +413,11 @@ func (s *knowledgeService) CreateKnowledgeFromURL(ctx context.Context,

 	// Enqueue URL processing task to Asynq
 	logger.Info(ctx, "Enqueuing URL processing task to Asynq")
-	enableMultimodelValue := false
-	if enableMultimodel != nil {
-		enableMultimodelValue = *enableMultimodel
-	} else {
-		enableMultimodelValue = kb.IsMultimodalEnabled()
-	}
-
-	// Check question generation config
-	enableQuestionGeneration := false
-	questionCount := 3 // default
-	if kb.QuestionGenerationConfig != nil && kb.QuestionGenerationConfig.Enabled {
-		enableQuestionGeneration = true
-		if kb.QuestionGenerationConfig.QuestionCount > 0 {
-			questionCount = kb.QuestionGenerationConfig.QuestionCount
-		}
+	enableMultimodelValue := eff.EnableMultimodel
+	enableQuestionGeneration := eff.QuestionGenerationConfig.Enabled
+	questionCount := eff.QuestionGenerationConfig.QuestionCount
+	if questionCount <= 0 {
+		questionCount = 3
 	}

 	lang, _ := types.LanguageFromContext(ctx)
@@ -515,6 +514,7 @@ func (s *knowledgeService) createKnowledgeFromFileURL(
 	title string,
 	tagID string,
 	channel string,
+	processOverrides *types.KnowledgeProcessOverrides,
 ) (*types.Knowledge, error) {
 	logger.Info(ctx, "Start creating knowledge from file URL")
 	logger.Infof(ctx, "Knowledge base ID: %s, file URL: %s", kbID, fileURL)
@@ -624,26 +624,32 @@ func (s *knowledgeService) createKnowledgeFromFileURL(
 		knowledge.Title = displayName
 	}

+	resolvedFileType := fileType
+	if resolvedFileType == "" && fileName != "" {
+		resolvedFileType = getFileType(fileName)
+	}
+	if resolvedFileType == "" {
+		resolvedFileType = getFileType(extractFileNameFromURL(fileURL))
+	}
+
+	eff, err := ApplyKnowledgeProcessOverrides(
+		ctx, kb, knowledge, processOverrides, []string{resolvedFileType}, enableMultimodel,
+	)
+	if err != nil {
+		return nil, err
+	}
+
 	if err := s.repo.CreateKnowledge(ctx, knowledge); err != nil {
 		logger.Errorf(ctx, "Failed to create knowledge record: %v", err)
 		return nil, err
 	}

 	// Build async task payload
-	enableMultimodelValue := false
-	if enableMultimodel != nil {
-		enableMultimodelValue = *enableMultimodel
-	} else {
-		enableMultimodelValue = kb.IsMultimodalEnabled()
-	}
-
-	enableQuestionGeneration := false
-	questionCount := 3
-	if kb.QuestionGenerationConfig != nil && kb.QuestionGenerationConfig.Enabled {
-		enableQuestionGeneration = true
-		if kb.QuestionGenerationConfig.QuestionCount > 0 {
-			questionCount = kb.QuestionGenerationConfig.QuestionCount
-		}
+	enableMultimodelValue := eff.EnableMultimodel
+	enableQuestionGeneration := eff.QuestionGenerationConfig.Enabled
+	questionCount := eff.QuestionGenerationConfig.QuestionCount
+	if questionCount <= 0 {
+		questionCount = 3
 	}

 	lang, _ := types.LanguageFromContext(ctx)
@@ -775,6 +781,12 @@ func (s *knowledgeService) CreateKnowledgeFromManual(ctx context.Context,
 		knowledge.ParseStatus = "pending"
 	}

+	if status == types.ManualKnowledgeStatusPublish {
+		if _, err := ApplyKnowledgeProcessOverrides(ctx, kb, knowledge, payload.ProcessConfig, nil, nil); err != nil {
+			return nil, err
+		}
+	}
+
 	if err := s.repo.CreateKnowledge(ctx, knowledge); err != nil {
 		logger.Errorf(ctx, "Failed to create manual knowledge record: %v", err)
 		return nil, err
@@ -998,6 +1010,10 @@ func (s *knowledgeService) UpdateManualKnowledge(ctx context.Context,
 	existing.Description = ""
 	existing.ProcessedAt = nil

+	if _, err := ApplyKnowledgeProcessOverrides(ctx, kb, existing, payload.ProcessConfig, nil, nil); err != nil {
+		return nil, err
+	}
+
 	if err := s.repo.UpdateKnowledge(ctx, existing); err != nil {
 		logger.Errorf(ctx, "Failed to persist manual knowledge before indexing: %v", err)
 		return nil, err
@@ -1101,26 +1117,27 @@ func (s *knowledgeService) triggerManualProcessing(ctx context.Context,
 		}
 	}

+	processOverrides, _ := knowledge.ProcessOverrides()
+	eff := ResolveProcessConfig(kb, processOverrides)
+
 	// Manual content is markdown - chunk directly with Go chunker
-	chunkCfg := buildSplitterConfig(kb)
+	chunkCfg := buildSplitterConfigFromChunking(eff.ChunkingConfig)

 	var parsed []types.ParsedChunk
 	opts := ProcessChunksOptions{
-		// When the KB has VLM enabled and we resolved remote images, pass them
-		// through so processChunks will enqueue image:multimodal tasks (OCR + caption).
-		EnableMultimodel: kb.IsMultimodalEnabled() && len(resolvedImages) > 0,
+		EnableMultimodel: eff.EnableMultimodel && len(resolvedImages) > 0,
 		StoredImages:     resolvedImages,
 	}
-	if kb.QuestionGenerationConfig != nil && kb.QuestionGenerationConfig.Enabled {
+	if eff.QuestionGenerationConfig.Enabled {
 		opts.EnableQuestionGeneration = true
-		opts.QuestionCount = kb.QuestionGenerationConfig.QuestionCount
+		opts.QuestionCount = eff.QuestionGenerationConfig.QuestionCount
 		if opts.QuestionCount <= 0 {
 			opts.QuestionCount = 3
 		}
 	}

-	if kb.ChunkingConfig.EnableParentChild {
-		parentCfg, childCfg := buildParentChildConfigs(kb.ChunkingConfig, chunkCfg)
+	if eff.ChunkingConfig.EnableParentChild {
+		parentCfg, childCfg := buildParentChildConfigs(eff.ChunkingConfig, chunkCfg)
 		pcResult := chunker.SplitParentChild(clean, parentCfg, childCfg)
 		parsed = make([]types.ParsedChunk, len(pcResult.Children))
 		for i, c := range pcResult.Children {
@@ -138,6 +138,7 @@ func TestCreateKnowledgeFromFileDoesNotPersistWhenStorageSaveFails(t *testing.T)
 		"",
 		"",
 		"",
+		nil,
 	)

 	require.Error(t, err)
@@ -168,6 +169,7 @@ func TestCreateKnowledgeFromFilePersistsStoredFilePathOnCreate(t *testing.T) {
 		"",
 		"",
 		"",
+		nil,
 	)

 	require.NoError(t, err)
@@ -201,6 +203,7 @@ func TestCreateKnowledgeFromFileDeletesStoredFileWhenCreateFails(t *testing.T) {
 		"",
 		"",
 		"",
+		nil,
 	)

 	require.EqualError(t, err, "database unavailable")
@@ -211,6 +214,52 @@ func TestCreateKnowledgeFromFileDeletesStoredFileWhenCreateFails(t *testing.T) {
 	require.Equal(t, "stored/"+fileSvc.savedWithKnowledgeID, fileSvc.deletedPath)
 }

+func TestCreateKnowledgeFromFile_PersistsProcessOverrides(t *testing.T) {
+	t.Parallel()
+
+	repo := &createKnowledgeFileRepoStub{}
+	fileSvc := &createKnowledgeFileServiceStub{}
+	task := &createKnowledgeTaskEnqueuerStub{}
+	svc := &knowledgeService{
+		repo:      repo,
+		kbService: &createKnowledgeFileKBServiceStub{kb: &types.KnowledgeBase{ID: "kb-1"}},
+		fileSvc:   fileSvc,
+		task:      task,
+	}
+
+	chunkSize := 512
+	overrides := &types.KnowledgeProcessOverrides{
+		ChunkingConfig: &types.ChunkingConfig{ChunkSize: chunkSize},
+	}
+
+	knowledge, err := svc.CreateKnowledgeFromFile(
+		newCreateKnowledgeFileContext(),
+		"kb-1",
+		newMultipartFileHeader(t, "doc.txt", "hello"),
+		map[string]string{"source": "test"},
+		nil,
+		"",
+		"",
+		"",
+		overrides,
+	)
+
+	require.NoError(t, err)
+	require.NotNil(t, knowledge)
+	require.Equal(t, 1, repo.createCalls)
+	require.NotNil(t, repo.createdKnowledge)
+
+	parsed, err := repo.createdKnowledge.ProcessOverrides()
+	require.NoError(t, err)
+	require.NotNil(t, parsed)
+	require.NotNil(t, parsed.ChunkingConfig)
+	require.Equal(t, chunkSize, parsed.ChunkingConfig.ChunkSize)
+
+	metadataMap, err := repo.createdKnowledge.Metadata.Map()
+	require.NoError(t, err)
+	require.Equal(t, "test", metadataMap["source"])
+}
+
 func newCreateKnowledgeFileContext() context.Context {
 	ctx := context.WithValue(context.Background(), types.TenantIDContextKey, uint64(1))
 	ctx = context.WithValue(ctx, types.TenantInfoContextKey, &types.Tenant{})
@@ -124,6 +124,9 @@ func (s *KnowledgePostProcessService) Handle(ctx context.Context, task *asynq.Ta
 		return fmt.Errorf("get knowledge base %s: %w", payload.KnowledgeBaseID, err)
 	}

+	processOverrides, _ := knowledge.ProcessOverrides()
+	eff := ResolveProcessConfig(kb, processOverrides)
+
 	// 2. Fetch all chunks
 	chunks, err := s.chunkService.ListChunksByKnowledgeID(ctx, payload.KnowledgeID)
 	if err != nil {
@@ -155,7 +158,7 @@ func (s *KnowledgePostProcessService) Handle(ctx context.Context, task *asynq.Ta
 	//    drains is bounded by the housekeeping finalizing sweep.
 	willSpawnSummary := len(textChunks) > 0
 	willSpawnQuestion := willSpawnSummary && kb.NeedsEmbeddingModel() &&
-		kb.QuestionGenerationConfig != nil && kb.QuestionGenerationConfig.Enabled
+		eff.QuestionGenerationConfig.Enabled
 	willSpawnWiki := kb.IndexingStrategy.WikiEnabled && len(textChunks) > 0

 	// Question generation now fans out one subtask per plain text chunk
@@ -184,7 +187,7 @@ func (s *KnowledgePostProcessService) Handle(ctx context.Context, task *asynq.Ta
 	questionBatchCount := (len(questionChunks) + questionGenChunkBatchSize - 1) / questionGenChunkBatchSize

 	graphChunkCount := 0
-	if kb.IsGraphEnabled() {
+	if eff.GraphEnabled {
 		graphChunkCount = len(textChunks)
 	}
 	expectedSubtasks := 0
@@ -302,7 +305,7 @@ func (s *KnowledgePostProcessService) Handle(ctx context.Context, task *asynq.Ta
 					"chunk_count": len(questionChunks),
 				})
 			}
-			enqueuedQuestionCount = s.enqueueQuestionGenerationTasks(ctx, payload, kb, attempt, questionChunks)
+			enqueuedQuestionCount = s.enqueueQuestionGenerationTasks(ctx, payload, eff.QuestionGenerationConfig, attempt, questionChunks)
 		}
 	}

@@ -466,18 +469,18 @@ const postprocessQuestionGroupSpanName = "postprocess.question"
 func (s *KnowledgePostProcessService) enqueueQuestionGenerationTasks(
 	ctx context.Context,
 	payload types.KnowledgePostProcessPayload,
-	kb *types.KnowledgeBase,
+	qg types.QuestionGenerationConfig,
 	attempt int,
 	questionChunks []*types.Chunk,
 ) int {
 	if s.taskEnqueuer == nil || len(questionChunks) == 0 {
 		return 0
 	}
-	if kb.QuestionGenerationConfig == nil || !kb.QuestionGenerationConfig.Enabled {
+	if !qg.Enabled {
 		return 0
 	}

-	questionCount := kb.QuestionGenerationConfig.QuestionCount
+	questionCount := qg.QuestionCount
 	if questionCount <= 0 {
 		questionCount = 3
 	}
@@ -201,13 +201,17 @@ func finalizeIndexedKnowledgeState(
 // identical whether callers come through this path or invoke the chunker
 // directly with a zero-value config.
 func buildSplitterConfig(kb *types.KnowledgeBase) chunker.SplitterConfig {
+	return buildSplitterConfigFromChunking(kb.ChunkingConfig)
+}
+
+func buildSplitterConfigFromChunking(cc types.ChunkingConfig) chunker.SplitterConfig {
 	chunkCfg := chunker.SplitterConfig{
-		ChunkSize:    kb.ChunkingConfig.ChunkSize,
-		ChunkOverlap: kb.ChunkingConfig.ChunkOverlap,
-		Separators:   kb.ChunkingConfig.Separators,
-		Strategy:     kb.ChunkingConfig.Strategy,
-		TokenLimit:   kb.ChunkingConfig.TokenLimit,
-		Languages:    kb.ChunkingConfig.Languages,
+		ChunkSize:    cc.ChunkSize,
+		ChunkOverlap: cc.ChunkOverlap,
+		Separators:   cc.Separators,
+		Strategy:     cc.Strategy,
+		TokenLimit:   cc.TokenLimit,
+		Languages:    cc.Languages,
 	}
 	if chunkCfg.ChunkSize <= 0 {
 		chunkCfg.ChunkSize = chunker.DefaultChunkSize
@@ -1973,6 +1977,9 @@ func (s *knowledgeService) ReparseKnowledge(ctx context.Context, knowledgeID str
 		return nil, err
 	}

+	processOverrides, _ := existing.ProcessOverrides()
+	reparseEff := ResolveProcessConfig(kb, processOverrides)
+
 	// Keep wiki's pending queue consistent across both manual and non-manual
 	// paths. The destructive work (swapping old wiki contributions for new)
 	// happens asynchronously inside mapOneDocument — see its oldPageSlugs
@@ -2060,17 +2067,11 @@ func (s *knowledgeService) ReparseKnowledge(ctx context.Context, knowledgeID str
 	if existing.FilePath != "" {
 		tenantID := ctx.Value(types.TenantIDContextKey).(uint64)

-		// Determine multimodal setting
-		enableMultimodel := kb.IsMultimodalEnabled()
-
-		// Check question generation config
-		enableQuestionGeneration := false
-		questionCount := 3 // default
-		if kb.QuestionGenerationConfig != nil && kb.QuestionGenerationConfig.Enabled {
-			enableQuestionGeneration = true
-			if kb.QuestionGenerationConfig.QuestionCount > 0 {
-				questionCount = kb.QuestionGenerationConfig.QuestionCount
-			}
+		enableMultimodel := reparseEff.EnableMultimodel
+		enableQuestionGeneration := reparseEff.QuestionGenerationConfig.Enabled
+		questionCount := reparseEff.QuestionGenerationConfig.QuestionCount
+		if questionCount <= 0 {
+			questionCount = 3
 		}

 		lang, _ := types.LanguageFromContext(ctx)
@@ -2119,16 +2120,11 @@ func (s *knowledgeService) ReparseKnowledge(ctx context.Context, knowledgeID str
 	if existing.Type == "file_url" && existing.Source != "" {
 		tenantID := ctx.Value(types.TenantIDContextKey).(uint64)

-		enableMultimodel := kb.IsMultimodalEnabled()
-
-		// Check question generation config
-		enableQuestionGeneration := false
-		questionCount := 3
-		if kb.QuestionGenerationConfig != nil && kb.QuestionGenerationConfig.Enabled {
-			enableQuestionGeneration = true
-			if kb.QuestionGenerationConfig.QuestionCount > 0 {
-				questionCount = kb.QuestionGenerationConfig.QuestionCount
-			}
+		enableMultimodel := reparseEff.EnableMultimodel
+		enableQuestionGeneration := reparseEff.QuestionGenerationConfig.Enabled
+		questionCount := reparseEff.QuestionGenerationConfig.QuestionCount
+		if questionCount <= 0 {
+			questionCount = 3
 		}

 		lang, _ := types.LanguageFromContext(ctx)
@@ -2172,16 +2168,11 @@ func (s *knowledgeService) ReparseKnowledge(ctx context.Context, knowledgeID str
 	if existing.Type == "url" && existing.Source != "" {
 		tenantID := ctx.Value(types.TenantIDContextKey).(uint64)

-		enableMultimodel := kb.IsMultimodalEnabled()
-
-		// Check question generation config
-		enableQuestionGeneration := false
-		questionCount := 3
-		if kb.QuestionGenerationConfig != nil && kb.QuestionGenerationConfig.Enabled {
-			enableQuestionGeneration = true
-			if kb.QuestionGenerationConfig.QuestionCount > 0 {
-				questionCount = kb.QuestionGenerationConfig.QuestionCount
-			}
+		enableMultimodel := reparseEff.EnableMultimodel
+		enableQuestionGeneration := reparseEff.QuestionGenerationConfig.Enabled
+		questionCount := reparseEff.QuestionGenerationConfig.QuestionCount
+		if questionCount <= 0 {
+			questionCount = 3
 		}

 		lang, _ := types.LanguageFromContext(ctx)
@@ -2734,6 +2725,9 @@ func (s *knowledgeService) ProcessDocument(ctx context.Context, t *asynq.Task) e
 		return nil
 	}

+	processOverrides, _ := knowledge.ProcessOverrides()
+	eff := ResolveProcessConfig(kb, processOverrides)
+
 	// Re-check abort status right before flipping to "processing" — closes
 	// the race where the user cancels between the entry guard above and
 	// this write (otherwise the worker would overwrite cancelled→processing
@@ -2775,7 +2769,7 @@ func (s *knowledgeService) ProcessDocument(ctx context.Context, t *asynq.Task) e
 	}

 	// 检查音频ASR配置（仅对文件导入）
-	if payload.FilePath != "" && IsAudioType(payload.FileType) && !kb.ASRConfig.IsASREnabled() {
+	if payload.FilePath != "" && IsAudioType(payload.FileType) && !eff.ASRConfig.IsASREnabled() {
 		logger.GetLogger(ctx).WithField("knowledge_id", knowledge.ID).
 			Errorf("processDocument audio without ASR model configured")
 		knowledge.ParseStatus = "failed"
@@ -2857,7 +2851,7 @@ func (s *knowledgeService) ProcessDocument(ctx context.Context, t *asynq.Task) e
 		payload.FilePath = filePath
 		payload.FileName = resolvedFileName
 		payload.FileType = resolvedFileType
-		convertResult, err = s.convert(ctx, payload, kb, knowledge, isLastRetry)
+		convertResult, err = s.convert(ctx, payload, kb, knowledge, eff, isLastRetry)
 		if err != nil {
 			return err
 		}
@@ -2866,7 +2860,7 @@ func (s *knowledgeService) ProcessDocument(ctx context.Context, t *asynq.Task) e
 		}
 	} else if payload.URL != "" {
 		// URL import
-		convertResult, err = s.convert(ctx, payload, kb, knowledge, isLastRetry)
+		convertResult, err = s.convert(ctx, payload, kb, knowledge, eff, isLastRetry)
 		if err != nil {
 			return err
 		}
@@ -2910,7 +2904,7 @@ func (s *knowledgeService) ProcessDocument(ctx context.Context, t *asynq.Task) e
 		return nil
 	} else {
 		// File import
-		convertResult, err = s.convert(ctx, payload, kb, knowledge, isLastRetry)
+		convertResult, err = s.convert(ctx, payload, kb, knowledge, eff, isLastRetry)
 		if err != nil {
 			return err
 		}
@@ -2921,7 +2915,7 @@ func (s *knowledgeService) ProcessDocument(ctx context.Context, t *asynq.Task) e

 	// Step 1.5: ASR transcription for audio files
 	if convertResult != nil && convertResult.IsAudio && len(convertResult.AudioData) > 0 {
-		if !kb.ASRConfig.IsASREnabled() {
+		if !eff.ASRConfig.IsASREnabled() {
 			logger.Error(ctx, "Audio file detected but ASR is not configured")
 			knowledge.ParseStatus = "failed"
 			knowledge.ErrorMessage = "ASR model is not configured for audio transcription"
@@ -2933,7 +2927,7 @@ func (s *knowledgeService) ProcessDocument(ctx context.Context, t *asynq.Task) e
 		logger.Infof(ctx, "[ASR] Starting audio transcription for knowledge %s, audio size=%d bytes",
 			knowledge.ID, len(convertResult.AudioData))

-		asrModel, err := s.modelService.GetASRModel(ctx, kb.ASRConfig.ModelID)
+		asrModel, err := s.modelService.GetASRModel(ctx, eff.ASRConfig.ModelID)
 		if err != nil {
 			logger.Errorf(ctx, "[ASR] Failed to get ASR model: %v", err)
 			knowledge.ParseStatus = "failed"
@@ -3003,7 +2997,7 @@ func (s *knowledgeService) ProcessDocument(ctx context.Context, t *asynq.Task) e
 	}

 	// Step 3: Split into chunks using Go chunker
-	chunkCfg := buildSplitterConfig(kb)
+	chunkCfg := buildSplitterConfigFromChunking(eff.ChunkingConfig)

 	processOpts := ProcessChunksOptions{
 		EnableQuestionGeneration: payload.EnableQuestionGeneration,
@@ -3016,8 +3010,8 @@ func (s *knowledgeService) ProcessDocument(ctx context.Context, t *asynq.Task) e
 		processOpts.Metadata = convertResult.Metadata
 	}

-	if kb.ChunkingConfig.EnableParentChild {
-		parentCfg, childCfg := buildParentChildConfigs(kb.ChunkingConfig, chunkCfg)
+	if eff.ChunkingConfig.EnableParentChild {
+		parentCfg, childCfg := buildParentChildConfigs(eff.ChunkingConfig, chunkCfg)
 		pcResult := chunker.SplitParentChild(convertResult.MarkdownContent, parentCfg, childCfg)
 		chunks = make([]types.ParsedChunk, len(pcResult.Children))
 		for i, c := range pcResult.Children {
@@ -3064,6 +3058,7 @@ func (s *knowledgeService) convert(
 	payload types.DocumentProcessPayload,
 	kb *types.KnowledgeBase,
 	knowledge *types.Knowledge,
+	eff types.EffectiveProcessConfig,
 	isLastRetry bool,
 ) (*types.ReadResult, error) {
 	// Stage tracking: docreader. Mark the stage as running here so the
@@ -3097,13 +3092,13 @@ func (s *knowledgeService) convert(
 		}
 	}

-	parserEngine := kb.ChunkingConfig.ResolveParserEngine(fileType)
+	parserEngine := eff.ChunkingConfig.ResolveParserEngine(fileType)
 	if isURL {
-		parserEngine = kb.ChunkingConfig.ResolveParserEngine("url")
+		parserEngine = eff.ChunkingConfig.ResolveParserEngine("url")
 	}

 	logger.Infof(ctx, "[convert] kb=%s fileType=%s isURL=%v engine=%q rules=%+v",
-		kb.ID, fileType, isURL, parserEngine, kb.ChunkingConfig.ParserEngineRules)
+		kb.ID, fileType, isURL, parserEngine, eff.ChunkingConfig.ParserEngineRules)

 	var reader interfaces.DocReader = s.resolveDocReader(ctx, parserEngine, fileType, isURL, overrides)
 	if reader == nil {
@@ -0,0 +1,231 @@
+package service
+
+import (
+	"context"
+	"os"
+	"strings"
+
+	werrors "github.com/Tencent/WeKnora/internal/errors"
+	"github.com/Tencent/WeKnora/internal/types"
+)
+
+// ResolveProcessConfig merges KB defaults with per-upload overrides for the parse pipeline.
+func ResolveProcessConfig(kb *types.KnowledgeBase, overrides *types.KnowledgeProcessOverrides) types.EffectiveProcessConfig {
+	eff := types.EffectiveProcessConfig{
+		ChunkingConfig:           kb.ChunkingConfig,
+		EnableMultimodel:         kb.IsMultimodalEnabled(),
+		VLMConfig:                kb.VLMConfig,
+		ASRConfig:                kb.ASRConfig,
+		QuestionGenerationConfig: defaultQuestionGenerationConfig(kb),
+		GraphEnabled:             kb.IsGraphEnabled(),
+		ExtractConfig:            derefExtractConfig(kb.ExtractConfig),
+	}
+	if overrides == nil {
+		return eff
+	}
+
+	if overrides.ChunkingConfig != nil {
+		eff.ChunkingConfig = mergeChunkingConfig(eff.ChunkingConfig, overrides.ChunkingConfig)
+	}
+	if len(overrides.ParserEngineRules) > 0 {
+		eff.ChunkingConfig.ParserEngineRules = overrides.ParserEngineRules
+	}
+	if overrides.EnableMultimodel != nil {
+		eff.EnableMultimodel = *overrides.EnableMultimodel
+	}
+	if overrides.VLMConfig != nil {
+		eff.VLMConfig = *overrides.VLMConfig
+	}
+	if overrides.ASRConfig != nil {
+		eff.ASRConfig = *overrides.ASRConfig
+	}
+	if overrides.QuestionGenerationConfig != nil {
+		eff.QuestionGenerationConfig = *overrides.QuestionGenerationConfig
+	}
+	if overrides.GraphEnabled != nil {
+		eff.GraphEnabled = *overrides.GraphEnabled
+	}
+	if overrides.ExtractConfig != nil {
+		eff.ExtractConfig = mergeExtractConfig(eff.ExtractConfig, overrides.ExtractConfig)
+	}
+
+	return eff
+}
+
+// ValidateProcessOverrides validates batch overrides against file types in the upload.
+func ValidateProcessOverrides(
+	ctx context.Context,
+	kb *types.KnowledgeBase,
+	overrides *types.KnowledgeProcessOverrides,
+	fileTypes []string,
+) error {
+	if overrides == nil {
+		return nil
+	}
+
+	hasImage := false
+	hasAudio := false
+	for _, ft := range fileTypes {
+		if IsImageType(ft) {
+			hasImage = true
+		}
+		if IsAudioType(ft) {
+			hasAudio = true
+		}
+	}
+
+	eff := ResolveProcessConfig(kb, overrides)
+
+	if hasImage {
+		if err := validateImageMultimodalConfig(ctx, kb); err != nil {
+			return err
+		}
+		if !eff.VLMConfig.IsEnabled() {
+			return werrors.NewBadRequestError("上传图片文件需要设置VLM模型")
+		}
+	}
+
+	if hasAudio && !eff.ASRConfig.IsASREnabled() {
+		return werrors.NewBadRequestError("上传音频文件需要设置ASR语音识别模型")
+	}
+
+	return nil
+}
+
+// ApplyKnowledgeProcessOverrides validates optional overrides, persists them on the
+// knowledge record, and returns the effective config for task enqueue.
+func ApplyKnowledgeProcessOverrides(
+	ctx context.Context,
+	kb *types.KnowledgeBase,
+	knowledge *types.Knowledge,
+	processOverrides *types.KnowledgeProcessOverrides,
+	fileTypes []string,
+	enableMultimodel *bool,
+) (types.EffectiveProcessConfig, error) {
+	eff := ResolveProcessConfig(kb, processOverrides)
+	if enableMultimodel != nil && (processOverrides == nil || processOverrides.EnableMultimodel == nil) {
+		eff.EnableMultimodel = *enableMultimodel
+	}
+	if processOverrides == nil {
+		return eff, nil
+	}
+	if err := ValidateProcessOverrides(ctx, kb, processOverrides, fileTypes); err != nil {
+		return eff, err
+	}
+	if err := knowledge.SetProcessOverrides(processOverrides); err != nil {
+		return eff, err
+	}
+	return eff, nil
+}
+
+func defaultQuestionGenerationConfig(kb *types.KnowledgeBase) types.QuestionGenerationConfig {
+	if kb == nil || kb.QuestionGenerationConfig == nil {
+		return types.QuestionGenerationConfig{}
+	}
+	return *kb.QuestionGenerationConfig
+}
+
+func derefExtractConfig(cfg *types.ExtractConfig) types.ExtractConfig {
+	if cfg == nil {
+		return types.ExtractConfig{}
+	}
+	return *cfg
+}
+
+func mergeChunkingConfig(base types.ChunkingConfig, override *types.ChunkingConfig) types.ChunkingConfig {
+	if override == nil {
+		return base
+	}
+	result := base
+	if override.ChunkSize != 0 {
+		result.ChunkSize = override.ChunkSize
+	}
+	if override.ChunkOverlap != 0 {
+		result.ChunkOverlap = override.ChunkOverlap
+	}
+	if len(override.Separators) > 0 {
+		result.Separators = override.Separators
+	}
+	if override.EnableMultimodal {
+		result.EnableMultimodal = override.EnableMultimodal
+	}
+	if len(override.ParserEngineRules) > 0 {
+		result.ParserEngineRules = override.ParserEngineRules
+	}
+	if override.EnableParentChild {
+		result.EnableParentChild = override.EnableParentChild
+	}
+	if override.ParentChunkSize != 0 {
+		result.ParentChunkSize = override.ParentChunkSize
+	}
+	if override.ChildChunkSize != 0 {
+		result.ChildChunkSize = override.ChildChunkSize
+	}
+	if override.Strategy != "" {
+		result.Strategy = override.Strategy
+	}
+	if override.TokenLimit != 0 {
+		result.TokenLimit = override.TokenLimit
+	}
+	if len(override.Languages) > 0 {
+		result.Languages = override.Languages
+	}
+	return result
+}
+
+func mergeExtractConfig(base types.ExtractConfig, override *types.ExtractConfig) types.ExtractConfig {
+	if override == nil {
+		return base
+	}
+	result := base
+	if override.Enabled {
+		result.Enabled = true
+	}
+	if override.Text != "" {
+		result.Text = override.Text
+	}
+	if len(override.Tags) > 0 {
+		result.Tags = override.Tags
+	}
+	if len(override.Nodes) > 0 {
+		result.Nodes = override.Nodes
+	}
+	if len(override.Relations) > 0 {
+		result.Relations = override.Relations
+	}
+	return result
+}
+
+func validateImageMultimodalConfig(ctx context.Context, kb *types.KnowledgeBase) error {
+	provider := kb.GetStorageProvider()
+	tenant, _ := ctx.Value(types.TenantInfoContextKey).(*types.Tenant)
+	if provider == "" && tenant != nil && tenant.StorageEngineConfig != nil {
+		provider = strings.ToLower(strings.TrimSpace(tenant.StorageEngineConfig.DefaultProvider))
+	}
+
+	switch provider {
+	case "cos":
+		if tenant == nil || tenant.StorageEngineConfig == nil || tenant.StorageEngineConfig.COS == nil ||
+			tenant.StorageEngineConfig.COS.SecretID == "" || tenant.StorageEngineConfig.COS.SecretKey == "" ||
+			tenant.StorageEngineConfig.COS.Region == "" || tenant.StorageEngineConfig.COS.BucketName == "" {
+			return werrors.NewBadRequestError("上传图片文件需要完整的对象存储配置信息, 请前往知识库存储设置或系统设置页面进行补全")
+		}
+	case "minio":
+		ok := false
+		if tenant != nil && tenant.StorageEngineConfig != nil && tenant.StorageEngineConfig.MinIO != nil {
+			m := tenant.StorageEngineConfig.MinIO
+			if m.Mode == "remote" {
+				ok = m.Endpoint != "" && m.AccessKeyID != "" && m.SecretAccessKey != "" && m.BucketName != ""
+			} else {
+				ok = os.Getenv("MINIO_ENDPOINT") != "" && os.Getenv("MINIO_ACCESS_KEY_ID") != "" &&
+					os.Getenv("MINIO_SECRET_ACCESS_KEY") != "" &&
+					(m.BucketName != "" || os.Getenv("MINIO_BUCKET_NAME") != "")
+			}
+		}
+		if !ok {
+			return werrors.NewBadRequestError("上传图片文件需要完整的对象存储配置信息, 请前往知识库存储设置或系统设置页面进行补全")
+		}
+	}
+
+	return nil
+}
@@ -0,0 +1,245 @@
+package service
+
+import (
+	"context"
+	"testing"
+
+	werrors "github.com/Tencent/WeKnora/internal/errors"
+	"github.com/Tencent/WeKnora/internal/types"
+	"github.com/stretchr/testify/require"
+)
+
+func processConfigBoolPtr(v bool) *bool {
+	return &v
+}
+
+func testKBWithGraphEnabled(enabled bool) *types.KnowledgeBase {
+	return &types.KnowledgeBase{
+		IndexingStrategy: types.IndexingStrategy{GraphEnabled: enabled},
+		ExtractConfig:    &types.ExtractConfig{Enabled: enabled},
+	}
+}
+
+func TestResolveProcessConfig_OverridesChunkSize(t *testing.T) {
+	t.Parallel()
+
+	kb := &types.KnowledgeBase{
+		ChunkingConfig: types.ChunkingConfig{ChunkSize: 512, ChunkOverlap: 50},
+	}
+	overrides := &types.KnowledgeProcessOverrides{
+		ChunkingConfig: &types.ChunkingConfig{ChunkSize: 2048},
+	}
+	eff := ResolveProcessConfig(kb, overrides)
+	require.Equal(t, 2048, eff.ChunkingConfig.ChunkSize)
+	require.Equal(t, 50, eff.ChunkingConfig.ChunkOverlap)
+}
+
+func TestResolveProcessConfig_GraphDisabled(t *testing.T) {
+	t.Parallel()
+
+	kb := testKBWithGraphEnabled(true)
+	overrides := &types.KnowledgeProcessOverrides{GraphEnabled: processConfigBoolPtr(false)}
+	eff := ResolveProcessConfig(kb, overrides)
+	require.False(t, eff.GraphEnabled)
+}
+
+func TestResolveProcessConfig_NilOverridesUsesKBDefaults(t *testing.T) {
+	t.Parallel()
+
+	kb := &types.KnowledgeBase{
+		ChunkingConfig: types.ChunkingConfig{ChunkSize: 512, ChunkOverlap: 50},
+		VLMConfig:      types.VLMConfig{Enabled: true, ModelID: "vlm-1"},
+		ASRConfig:      types.ASRConfig{Enabled: true, ModelID: "asr-1"},
+		QuestionGenerationConfig: &types.QuestionGenerationConfig{
+			Enabled:       true,
+			QuestionCount: 3,
+		},
+		IndexingStrategy: types.IndexingStrategy{GraphEnabled: true},
+		ExtractConfig:    &types.ExtractConfig{Enabled: true, Tags: []string{"tag-a"}},
+	}
+	kb.ChunkingConfig.EnableMultimodal = true
+
+	eff := ResolveProcessConfig(kb, nil)
+
+	require.Equal(t, 512, eff.ChunkingConfig.ChunkSize)
+	require.Equal(t, 50, eff.ChunkingConfig.ChunkOverlap)
+	require.True(t, eff.EnableMultimodel)
+	require.Equal(t, "vlm-1", eff.VLMConfig.ModelID)
+	require.Equal(t, "asr-1", eff.ASRConfig.ModelID)
+	require.True(t, eff.QuestionGenerationConfig.Enabled)
+	require.Equal(t, 3, eff.QuestionGenerationConfig.QuestionCount)
+	require.True(t, eff.GraphEnabled)
+	require.True(t, eff.ExtractConfig.Enabled)
+	require.Equal(t, []string{"tag-a"}, eff.ExtractConfig.Tags)
+}
+
+func TestBuildSplitterConfigFromChunking_UsesEffectiveChunkingConfig(t *testing.T) {
+	t.Parallel()
+
+	kb := &types.KnowledgeBase{
+		ChunkingConfig: types.ChunkingConfig{ChunkSize: 512, ChunkOverlap: 50, Strategy: "token"},
+	}
+	overrides := &types.KnowledgeProcessOverrides{
+		ChunkingConfig: &types.ChunkingConfig{ChunkSize: 1500, ChunkOverlap: 120, Strategy: "character"},
+	}
+	eff := ResolveProcessConfig(kb, overrides)
+	cfg := buildSplitterConfigFromChunking(eff.ChunkingConfig)
+
+	require.Equal(t, 1500, cfg.ChunkSize)
+	require.Equal(t, 120, cfg.ChunkOverlap)
+	require.Equal(t, "character", cfg.Strategy)
+}
+
+func TestEffectiveChunkingConfig_ResolveParserEngineFromOverrides(t *testing.T) {
+	t.Parallel()
+
+	kb := &types.KnowledgeBase{
+		ChunkingConfig: types.ChunkingConfig{
+			ParserEngineRules: []types.ParserEngineRule{
+				{FileTypes: []string{"pdf"}, Engine: "builtin"},
+			},
+		},
+	}
+	overrides := &types.KnowledgeProcessOverrides{
+		ParserEngineRules: []types.ParserEngineRule{
+			{FileTypes: []string{"pdf"}, Engine: "mineru"},
+		},
+	}
+	eff := ResolveProcessConfig(kb, overrides)
+	require.Equal(t, "mineru", eff.ChunkingConfig.ResolveParserEngine("pdf"))
+}
+
+func TestResolveProcessConfig_ParserEngineRulesReplaced(t *testing.T) {
+	t.Parallel()
+
+	kb := &types.KnowledgeBase{
+		ChunkingConfig: types.ChunkingConfig{
+			ParserEngineRules: []types.ParserEngineRule{
+				{FileTypes: []string{"pdf"}, Engine: "builtin"},
+			},
+		},
+	}
+	overrides := &types.KnowledgeProcessOverrides{
+		ParserEngineRules: []types.ParserEngineRule{
+			{FileTypes: []string{"docx"}, Engine: "custom"},
+		},
+	}
+	eff := ResolveProcessConfig(kb, overrides)
+	require.Len(t, eff.ChunkingConfig.ParserEngineRules, 1)
+	require.Equal(t, []string{"docx"}, eff.ChunkingConfig.ParserEngineRules[0].FileTypes)
+	require.Equal(t, "custom", eff.ChunkingConfig.ParserEngineRules[0].Engine)
+}
+
+func TestResolveProcessConfig_EnableMultimodelOverride(t *testing.T) {
+	t.Parallel()
+
+	kb := &types.KnowledgeBase{
+		VLMConfig: types.VLMConfig{Enabled: true, ModelID: "vlm-1"},
+	}
+	overrides := &types.KnowledgeProcessOverrides{
+		EnableMultimodel: processConfigBoolPtr(false),
+	}
+	eff := ResolveProcessConfig(kb, overrides)
+	require.False(t, eff.EnableMultimodel)
+}
+
+func TestResolveProcessConfig_ExtractConfigFieldMerge(t *testing.T) {
+	t.Parallel()
+
+	kb := &types.KnowledgeBase{
+		ExtractConfig: &types.ExtractConfig{
+			Enabled: true,
+			Text:    "base text",
+			Tags:    []string{"base-tag"},
+		},
+	}
+	overrides := &types.KnowledgeProcessOverrides{
+		ExtractConfig: &types.ExtractConfig{
+			Tags: []string{"override-tag"},
+		},
+	}
+	eff := ResolveProcessConfig(kb, overrides)
+	require.True(t, eff.ExtractConfig.Enabled)
+	require.Equal(t, "base text", eff.ExtractConfig.Text)
+	require.Equal(t, []string{"override-tag"}, eff.ExtractConfig.Tags)
+}
+
+func TestValidateProcessOverrides_NilOverrides(t *testing.T) {
+	t.Parallel()
+
+	err := ValidateProcessOverrides(context.Background(), &types.KnowledgeBase{}, nil, []string{"png"})
+	require.NoError(t, err)
+}
+
+func TestValidateProcessOverrides_ImageRequiresVLM(t *testing.T) {
+	t.Parallel()
+
+	kb := &types.KnowledgeBase{
+		VLMConfig: types.VLMConfig{Enabled: false},
+	}
+	err := ValidateProcessOverrides(context.Background(), kb, &types.KnowledgeProcessOverrides{}, []string{"png"})
+	require.Error(t, err)
+	var badReq *werrors.AppError
+	require.ErrorAs(t, err, &badReq)
+}
+
+func TestValidateProcessOverrides_ImageWithEffectiveVLM(t *testing.T) {
+	t.Parallel()
+
+	kb := &types.KnowledgeBase{
+		VLMConfig: types.VLMConfig{Enabled: false},
+	}
+	overrides := &types.KnowledgeProcessOverrides{
+		VLMConfig: &types.VLMConfig{Enabled: true, ModelID: "vlm-1"},
+	}
+	err := ValidateProcessOverrides(context.Background(), kb, overrides, []string{"jpg"})
+	require.NoError(t, err)
+}
+
+func TestValidateProcessOverrides_AudioRequiresASR(t *testing.T) {
+	t.Parallel()
+
+	kb := &types.KnowledgeBase{
+		ASRConfig: types.ASRConfig{Enabled: false},
+	}
+	err := ValidateProcessOverrides(context.Background(), kb, &types.KnowledgeProcessOverrides{}, []string{"mp3"})
+	require.Error(t, err)
+}
+
+func TestValidateProcessOverrides_AudioWithEffectiveASR(t *testing.T) {
+	t.Parallel()
+
+	kb := &types.KnowledgeBase{
+		ASRConfig: types.ASRConfig{Enabled: false},
+	}
+	overrides := &types.KnowledgeProcessOverrides{
+		ASRConfig: &types.ASRConfig{Enabled: true, ModelID: "asr-1"},
+	}
+	err := ValidateProcessOverrides(context.Background(), kb, overrides, []string{"wav"})
+	require.NoError(t, err)
+}
+
+func TestValidateProcessOverrides_NonMediaFileTypes(t *testing.T) {
+	t.Parallel()
+
+	kb := &types.KnowledgeBase{}
+	err := ValidateProcessOverrides(context.Background(), kb, &types.KnowledgeProcessOverrides{}, []string{"pdf", "txt"})
+	require.NoError(t, err)
+}
+
+func TestValidateProcessOverrides_COSIncompleteForImage(t *testing.T) {
+	t.Parallel()
+
+	ctx := context.WithValue(context.Background(), types.TenantInfoContextKey, &types.Tenant{
+		StorageEngineConfig: &types.StorageEngineConfig{
+			COS: &types.COSEngineConfig{SecretID: "id"},
+		},
+	})
+	kb := &types.KnowledgeBase{
+		VLMConfig: types.VLMConfig{Enabled: true, ModelID: "vlm-1"},
+	}
+	kb.SetStorageProvider("cos")
+
+	err := ValidateProcessOverrides(ctx, kb, &types.KnowledgeProcessOverrides{}, []string{"png"})
+	require.Error(t, err)
+}
@@ -237,6 +237,7 @@ func (h *KnowledgeHandler) enqueueKnowledgeListDelete(
 // @Param        fileName          formData  string  false  "自定义文件名"
 // @Param        metadata          formData  string  false  "元数据JSON"
 // @Param        enable_multimodel formData  bool    false  "启用多模态处理"
+// @Param        process_config    formData  string  false  "处理配置JSON（KnowledgeProcessOverrides）"
 // @Success      200               {object}  map[string]interface{}  "创建的知识"
 // @Failure      400               {object}  errors.AppError         "请求参数错误"
 // @Failure      409               {object}  map[string]interface{}  "文件重复"
@@ -318,6 +319,23 @@ func (h *KnowledgeHandler) CreateKnowledgeFromFile(c *gin.Context) {
 		enableMultimodel = &parseBool
 	}

+	var processOverrides *types.KnowledgeProcessOverrides
+	if raw := c.PostForm("process_config"); raw != "" {
+		processOverrides = &types.KnowledgeProcessOverrides{}
+		if err := json.Unmarshal([]byte(raw), processOverrides); err != nil {
+			logger.Error(ctx, "Failed to parse process_config", err)
+			c.Error(errors.NewBadRequestError("Invalid process_config format").WithDetails(err.Error()))
+			return
+		}
+	}
+	if enableMultimodel != nil && (processOverrides == nil || processOverrides.EnableMultimodel == nil) {
+		if processOverrides == nil {
+			processOverrides = &types.KnowledgeProcessOverrides{EnableMultimodel: enableMultimodel}
+		} else {
+			processOverrides.EnableMultimodel = enableMultimodel
+		}
+	}
+
 	// 获取分类ID（如果提供），用于知识分类管理
 	tagID := c.PostForm("tag_id")
 	// 过滤特殊值，空字符串或 "__untagged__" 表示未分类
@@ -328,7 +346,7 @@ func (h *KnowledgeHandler) CreateKnowledgeFromFile(c *gin.Context) {
 	channel := c.PostForm("channel")

 	// Create knowledge entry from the file
-	knowledge, err := h.kgService.CreateKnowledgeFromFile(ctx, kbID, file, metadata, enableMultimodel, customFileName, tagID, channel)
+	knowledge, err := h.kgService.CreateKnowledgeFromFile(ctx, kbID, file, metadata, enableMultimodel, customFileName, tagID, channel, processOverrides)
 	// Check for duplicate knowledge error
 	if err != nil {
 		if h.handleDuplicateKnowledgeError(c, err, knowledge, "file") {
@@ -389,13 +407,14 @@ func (h *KnowledgeHandler) CreateKnowledgeFromURL(c *gin.Context) {

 	// Parse URL from request body
 	var req struct {
-		URL              string `json:"url" binding:"required"`
-		FileName         string `json:"file_name"`
-		FileType         string `json:"file_type"`
-		EnableMultimodel *bool  `json:"enable_multimodel"`
-		Title            string `json:"title"`
-		TagID            string `json:"tag_id"`
-		Channel          string `json:"channel"`
+		URL              string                           `json:"url" binding:"required"`
+		FileName         string                           `json:"file_name"`
+		FileType         string                           `json:"file_type"`
+		EnableMultimodel *bool                            `json:"enable_multimodel"`
+		Title            string                           `json:"title"`
+		TagID            string                           `json:"tag_id"`
+		Channel          string                           `json:"channel"`
+		ProcessConfig    *types.KnowledgeProcessOverrides `json:"process_config"`
 	}
 	if err := c.ShouldBindJSON(&req); err != nil {
 		logger.Error(ctx, "Failed to parse URL request", err)
@@ -423,7 +442,9 @@ func (h *KnowledgeHandler) CreateKnowledgeFromURL(c *gin.Context) {
 	)

 	// Create knowledge entry from the URL
-	knowledge, err := h.kgService.CreateKnowledgeFromURL(ctx, kbID, req.URL, req.FileName, req.FileType, req.EnableMultimodel, req.Title, req.TagID, req.Channel)
+	knowledge, err := h.kgService.CreateKnowledgeFromURL(
+		ctx, kbID, req.URL, req.FileName, req.FileType, req.EnableMultimodel, req.Title, req.TagID, req.Channel, req.ProcessConfig,
+	)
 	// Check for duplicate knowledge error
 	if err != nil {
 		if h.handleDuplicateKnowledgeError(c, err, knowledge, "url") {
@@ -2430,7 +2430,7 @@ func (s *Service) processFileToKnowledgeBase(ctx context.Context, msg *IncomingM
 	fh := newInMemoryFileHeader(fileName, content)

 	// Create knowledge entry via the knowledge service
-	knowledge, err := s.knowledgeService.CreateKnowledgeFromFile(kbCtx, kbID, fh, nil, nil, "", "", imPlatformToChannel(channel.Platform))
+	knowledge, err := s.knowledgeService.CreateKnowledgeFromFile(kbCtx, kbID, fh, nil, nil, "", "", imPlatformToChannel(channel.Platform), nil)
 	if err != nil {
 		errMsg := err.Error()
 		// Check for duplicate file
@@ -22,6 +22,7 @@ type KnowledgeService interface {
 		customFileName string,
 		tagID string,
 		channel string,
+		processOverrides *types.KnowledgeProcessOverrides,
 	) (*types.Knowledge, error)
 	// CreateKnowledgeFromURL creates knowledge from a URL.
 	// When fileName or fileType is provided (or the URL path has a known file extension),
@@ -37,6 +38,7 @@ type KnowledgeService interface {
 		title string,
 		tagID string,
 		channel string,
+		processOverrides *types.KnowledgeProcessOverrides,
 	) (*types.Knowledge, error)
 	// CreateKnowledgeFromPassage creates knowledge from text passages.
 	// channel identifies the ingestion channel; empty defaults to "web".
@@ -204,11 +204,12 @@ type ManualKnowledgeMetadata struct {

 // ManualKnowledgePayload represents the payload for manual knowledge operations.
 type ManualKnowledgePayload struct {
-	Title   string `json:"title"`
-	Content string `json:"content"`
-	Status  string `json:"status"`
-	TagID   string `json:"tag_id"`
-	Channel string `json:"channel"`
+	Title         string                     `json:"title"`
+	Content       string                     `json:"content"`
+	Status        string                     `json:"status"`
+	TagID         string                     `json:"tag_id"`
+	Channel       string                     `json:"channel"`
+	ProcessConfig *KnowledgeProcessOverrides `json:"process_config,omitempty"`
 }

 // KnowledgeSearchScope defines a (tenant_id, knowledge_base_id) scope for knowledge search (e.g. own KBs + shared KBs).
@@ -342,6 +343,62 @@ func (p ManualKnowledgePayload) IsDraft() bool {
 	return p.Status == "" || p.Status == ManualKnowledgeStatusDraft
 }

+const metadataKeyProcessOverrides = "process_overrides"
+
+// ProcessOverrides parses process config overrides from knowledge metadata.
+func (k *Knowledge) ProcessOverrides() (*KnowledgeProcessOverrides, error) {
+	if k == nil || len(k.Metadata) == 0 {
+		return nil, nil
+	}
+	metadataMap, err := k.Metadata.Map()
+	if err != nil {
+		return nil, err
+	}
+	raw, ok := metadataMap[metadataKeyProcessOverrides]
+	if !ok || raw == nil {
+		return nil, nil
+	}
+	bytes, err := json.Marshal(raw)
+	if err != nil {
+		return nil, err
+	}
+	var overrides KnowledgeProcessOverrides
+	if err := json.Unmarshal(bytes, &overrides); err != nil {
+		return nil, err
+	}
+	return &overrides, nil
+}
+
+// SetProcessOverrides merges process config overrides into knowledge metadata.
+func (k *Knowledge) SetProcessOverrides(o *KnowledgeProcessOverrides) error {
+	if k == nil {
+		return nil
+	}
+	metadataMap, err := k.Metadata.Map()
+	if err != nil {
+		return err
+	}
+	if o == nil {
+		delete(metadataMap, metadataKeyProcessOverrides)
+	} else {
+		bytes, err := json.Marshal(o)
+		if err != nil {
+			return err
+		}
+		var value interface{}
+		if err := json.Unmarshal(bytes, &value); err != nil {
+			return err
+		}
+		metadataMap[metadataKeyProcessOverrides] = value
+	}
+	bytes, err := json.Marshal(metadataMap)
+	if err != nil {
+		return err
+	}
+	k.Metadata = JSON(bytes)
+	return nil
+}
+
 // KnowledgeCheckParams defines parameters used to check if knowledge already exists.
 type KnowledgeCheckParams struct {
 	// File parameters
@@ -0,0 +1,24 @@
+package types
+
+// KnowledgeProcessOverrides stores per-upload parse config overrides in knowledge metadata.
+type KnowledgeProcessOverrides struct {
+	ParserEngineRules        []ParserEngineRule        `json:"parser_engine_rules,omitempty"`
+	ChunkingConfig           *ChunkingConfig           `json:"chunking_config,omitempty"`
+	EnableMultimodel         *bool                     `json:"enable_multimodel,omitempty"`
+	VLMConfig                *VLMConfig                `json:"vlm_config,omitempty"`
+	ASRConfig                *ASRConfig                `json:"asr_config,omitempty"`
+	QuestionGenerationConfig *QuestionGenerationConfig `json:"question_generation_config,omitempty"`
+	GraphEnabled             *bool                     `json:"graph_enabled,omitempty"`
+	ExtractConfig            *ExtractConfig            `json:"extract_config,omitempty"`
+}
+
+// EffectiveProcessConfig is the merged view used by the parse pipeline.
+type EffectiveProcessConfig struct {
+	ChunkingConfig           ChunkingConfig
+	EnableMultimodel         bool
+	VLMConfig                VLMConfig
+	ASRConfig                ASRConfig
+	QuestionGenerationConfig QuestionGenerationConfig
+	GraphEnabled             bool
+	ExtractConfig            ExtractConfig
+}
@@ -0,0 +1,48 @@
+package types
+
+import (
+	"testing"
+
+	"github.com/stretchr/testify/require"
+)
+
+func boolPtr(v bool) *bool {
+	return &v
+}
+
+func TestKnowledgeProcessOverridesRoundtrip(t *testing.T) {
+	k := &Knowledge{}
+	overrides := &KnowledgeProcessOverrides{
+		EnableMultimodel: boolPtr(true),
+		ChunkingConfig:   &ChunkingConfig{ChunkSize: 1024},
+	}
+	require.NoError(t, k.SetProcessOverrides(overrides))
+	got, err := k.ProcessOverrides()
+	require.NoError(t, err)
+	require.NotNil(t, got)
+	require.True(t, *got.EnableMultimodel)
+	require.Equal(t, 1024, got.ChunkingConfig.ChunkSize)
+}
+
+func TestSetProcessOverridesPreservesOtherMetadata(t *testing.T) {
+	k := &Knowledge{}
+	manualMeta := NewManualKnowledgeMetadata("# hello", ManualKnowledgeStatusDraft, 1)
+	require.NoError(t, k.SetManualMetadata(manualMeta))
+
+	overrides := &KnowledgeProcessOverrides{
+		EnableMultimodel: boolPtr(false),
+	}
+	require.NoError(t, k.SetProcessOverrides(overrides))
+
+	gotManual, err := k.ManualMetadata()
+	require.NoError(t, err)
+	require.NotNil(t, gotManual)
+	require.Equal(t, "# hello", gotManual.Content)
+	require.Equal(t, ManualKnowledgeFormatMarkdown, gotManual.Format)
+	require.Equal(t, ManualKnowledgeStatusDraft, gotManual.Status)
+
+	gotOverrides, err := k.ProcessOverrides()
+	require.NoError(t, err)
+	require.NotNil(t, gotOverrides)
+	require.False(t, *gotOverrides.EnableMultimodel)
+}