Merge pull request #14787 from vaclcer/vaclavs-custom-docling

feat: Customize Docling's "Describe Pictures" feature

backend/open_webui/config.py

@@ -1968,6 +1968,54 @@ DOCLING_DO_PICTURE_DESCRIPTION = PersistentConfig(
     os.getenv("DOCLING_DO_PICTURE_DESCRIPTION", "False").lower() == "true",
 )
 
+DOCLING_PICTURE_DESCRIPTION_MODE = PersistentConfig(
+    "DOCLING_PICTURE_DESCRIPTION_MODE",
+    "rag.docling_picture_description_mode",
+    os.getenv("DOCLING_PICTURE_DESCRIPTION_MODE", ""),
+)
+
+DOCLING_PICTURE_DESCRIPTION_LOCAL_REPO_ID = PersistentConfig(
+    "DOCLING_PICTURE_DESCRIPTION_LOCAL_REPO_ID",
+    "rag.docling_picture_description_local_repo_id",
+    os.getenv("DOCLING_PICTURE_DESCRIPTION_LOCAL_REPO_ID", "HuggingFaceTB/SmolVLM-256M-Instruct"),
+)
+
+DOCLING_PICTURE_DESCRIPTION_LOCAL_MAX_TOKENS = PersistentConfig(
+    "DOCLING_PICTURE_DESCRIPTION_LOCAL_MAX_TOKENS",
+    "rag.docling_picture_description_local_max_tokens",
+    int(os.getenv("DOCLING_PICTURE_DESCRIPTION_LOCAL_MAX_TOKENS", 200)),
+)
+
+DOCLING_PICTURE_DESCRIPTION_LOCAL_PROMPT = PersistentConfig(
+    "DOCLING_PICTURE_DESCRIPTION_LOCAL_PROMPT",
+    "rag.docling_picture_description_local_prompt",
+    os.getenv(
+        "DOCLING_PICTURE_DESCRIPTION_LOCAL_PROMPT",
+        "Describe this image in a few sentences.",
+    )
+)
+
+DOCLING_PICTURE_DESCRIPTION_API_URL = PersistentConfig(
+    "DOCLING_PICTURE_DESCRIPTION_API_URL",
+    "rag.docling_picture_description_api_url",
+    os.getenv("DOCLING_PICTURE_DESCRIPTION_API_URL", ""),
+)
+
+DOCLING_PICTURE_DESCRIPTION_API_MODEL = PersistentConfig(   
+    "DOCLING_PICTURE_DESCRIPTION_API_MODEL",
+    "rag.docling_picture_description_api_model",
+    os.getenv("DOCLING_PICTURE_DESCRIPTION_API_MODEL", ""),
+)
+
+DOCLING_PICTURE_DESCRIPTION_API_PROMPT = PersistentConfig(
+    "DOCLING_PICTURE_DESCRIPTION_API_PROMPT",
+    "rag.docling_picture_description_api_prompt",
+    os.getenv(
+        "DOCLING_PICTURE_DESCRIPTION_API_PROMPT",
+        "Describe this image in a few sentences.",
+    )
+)
+
 DOCUMENT_INTELLIGENCE_ENDPOINT = PersistentConfig(
     "DOCUMENT_INTELLIGENCE_ENDPOINT",
     "rag.document_intelligence_endpoint",

backend/open_webui/main.py

@@ -231,6 +231,13 @@ from open_webui.config import (
     DOCLING_OCR_ENGINE,
     DOCLING_OCR_LANG,
     DOCLING_DO_PICTURE_DESCRIPTION,
+    DOCLING_PICTURE_DESCRIPTION_MODE,
+    DOCLING_PICTURE_DESCRIPTION_LOCAL_REPO_ID,
+    DOCLING_PICTURE_DESCRIPTION_LOCAL_MAX_TOKENS,
+    DOCLING_PICTURE_DESCRIPTION_LOCAL_PROMPT,
+    DOCLING_PICTURE_DESCRIPTION_API_URL,
+    DOCLING_PICTURE_DESCRIPTION_API_MODEL,
+    DOCLING_PICTURE_DESCRIPTION_API_PROMPT,
     DOCUMENT_INTELLIGENCE_ENDPOINT,
     DOCUMENT_INTELLIGENCE_KEY,
     MISTRAL_OCR_API_KEY,
@@ -701,6 +708,13 @@ app.state.config.DOCLING_SERVER_URL = DOCLING_SERVER_URL
 app.state.config.DOCLING_OCR_ENGINE = DOCLING_OCR_ENGINE
 app.state.config.DOCLING_OCR_LANG = DOCLING_OCR_LANG
 app.state.config.DOCLING_DO_PICTURE_DESCRIPTION = DOCLING_DO_PICTURE_DESCRIPTION
+app.state.config.DOCLING_PICTURE_DESCRIPTION_MODE = DOCLING_PICTURE_DESCRIPTION_MODE
+app.state.config.DOCLING_PICTURE_DESCRIPTION_LOCAL_REPO_ID = DOCLING_PICTURE_DESCRIPTION_LOCAL_REPO_ID
+app.state.config.DOCLING_PICTURE_DESCRIPTION_LOCAL_MAX_TOKENS = DOCLING_PICTURE_DESCRIPTION_LOCAL_MAX_TOKENS
+app.state.config.DOCLING_PICTURE_DESCRIPTION_LOCAL_PROMPT = DOCLING_PICTURE_DESCRIPTION_LOCAL_PROMPT
+app.state.config.DOCLING_PICTURE_DESCRIPTION_API_URL = DOCLING_PICTURE_DESCRIPTION_API_URL
+app.state.config.DOCLING_PICTURE_DESCRIPTION_API_MODEL = DOCLING_PICTURE_DESCRIPTION_API_MODEL
+app.state.config.DOCLING_PICTURE_DESCRIPTION_API_PROMPT = DOCLING_PICTURE_DESCRIPTION_API_PROMPT
 app.state.config.DOCUMENT_INTELLIGENCE_ENDPOINT = DOCUMENT_INTELLIGENCE_ENDPOINT
 app.state.config.DOCUMENT_INTELLIGENCE_KEY = DOCUMENT_INTELLIGENCE_KEY
 app.state.config.MISTRAL_OCR_API_KEY = MISTRAL_OCR_API_KEY

backend/open_webui/retrieval/loaders/main.py

@@ -2,6 +2,7 @@ import requests
 import logging
 import ftfy
 import sys
+import json
 
 from langchain_community.document_loaders import (
     AzureAIDocumentIntelligenceLoader,
@@ -154,6 +155,41 @@ class DoclingLoader:
                         "do_picture_description"
                     )
 
+                    picture_description_mode = self.params.get("picture_description_mode", "").lower()
+
+                    if picture_description_mode == "local":
+
+                        params["picture_description_local"] = json.dumps({
+                            "repo_id": self.params.get(
+                                "picture_description_local_repo_id", "HuggingFaceTB/SmolVLM-256M-Instruct"
+                            ),
+                            "generation_config": {
+                                "max_new_tokens": self.params.get(
+                                    "picture_description_local_max_tokens", 200
+                                )
+                            },
+                            "prompt": self.params.get(
+                                "picture_description_local_prompt", "Describe this image in a few sentences."
+                            )
+                        })
+
+                    elif picture_description_mode == "api":
+
+                        params["picture_description_api"] = json.dumps({
+                            "url": self.params.get(
+                                "picture_description_api_url", ""
+                            ),
+                            "params": {
+                                "model": self.params.get(
+                                    "picture_description_api_model", ""
+                                )
+                            },
+                            "timeout": 30,
+                            "prompt": self.params.get(
+                                "picture_description_api_prompt", "Describe this image in a few sentences."
+                            )
+                        })
+
                 if self.params.get("ocr_engine") and self.params.get("ocr_lang"):
                     params["ocr_engine"] = self.params.get("ocr_engine")
                     params["ocr_lang"] = [
@@ -281,17 +317,25 @@ class Loader:
             if self._is_text_file(file_ext, file_content_type):
                 loader = TextLoader(file_path, autodetect_encoding=True)
             else:
+                # Build params for DoclingLoader
+                params = {
+                    "ocr_engine": self.kwargs.get("DOCLING_OCR_ENGINE"),
+                    "ocr_lang": self.kwargs.get("DOCLING_OCR_LANG"),
+                    "do_picture_description": self.kwargs.get("DOCLING_DO_PICTURE_DESCRIPTION"),
+                    "picture_description_mode": self.kwargs.get("DOCLING_PICTURE_DESCRIPTION_MODE"),
+                    "picture_description_local_repo_id": self.kwargs.get("DOCLING_PICTURE_DESCRIPTION_LOCAL_REPO_ID"),
+                    "picture_description_local_max_tokens": self.kwargs.get("DOCLING_PICTURE_DESCRIPTION_LOCAL_MAX_TOKENS"),
+                    "picture_description_local_prompt": self.kwargs.get("DOCLING_PICTURE_DESCRIPTION_LOCAL_PROMPT"),
+                    "picture_description_api_url": self.kwargs.get("DOCLING_PICTURE_DESCRIPTION_API_URL"),
+                    "picture_description_api_model": self.kwargs.get("DOCLING_PICTURE_DESCRIPTION_API_MODEL"),
+                    "picture_description_api_prompt": self.kwargs.get("DOCLING_PICTURE_DESCRIPTION_API_PROMPT")
+                }
+
                 loader = DoclingLoader(
                     url=self.kwargs.get("DOCLING_SERVER_URL"),
                     file_path=file_path,
                     mime_type=file_content_type,
-                    params={
-                        "ocr_engine": self.kwargs.get("DOCLING_OCR_ENGINE"),
-                        "ocr_lang": self.kwargs.get("DOCLING_OCR_LANG"),
-                        "do_picture_description": self.kwargs.get(
-                            "DOCLING_DO_PICTURE_DESCRIPTION"
-                        ),
-                    },
+                    params=params
                 )
         elif (
             self.engine == "document_intelligence"

backend/open_webui/routers/retrieval.py

@@ -414,6 +414,13 @@ async def get_rag_config(request: Request, user=Depends(get_admin_user)):
         "DOCLING_OCR_ENGINE": request.app.state.config.DOCLING_OCR_ENGINE,
         "DOCLING_OCR_LANG": request.app.state.config.DOCLING_OCR_LANG,
         "DOCLING_DO_PICTURE_DESCRIPTION": request.app.state.config.DOCLING_DO_PICTURE_DESCRIPTION,
+        "DOCLING_PICTURE_DESCRIPTION_MODE": request.app.state.config.DOCLING_PICTURE_DESCRIPTION_MODE,
+        "DOCLING_PICTURE_DESCRIPTION_LOCAL_REPO_ID": request.app.state.config.DOCLING_PICTURE_DESCRIPTION_LOCAL_REPO_ID,
+        "DOCLING_PICTURE_DESCRIPTION_LOCAL_MAX_TOKENS": request.app.state.config.DOCLING_PICTURE_DESCRIPTION_LOCAL_MAX_TOKENS,
+        "DOCLING_PICTURE_DESCRIPTION_LOCAL_PROMPT": request.app.state.config.DOCLING_PICTURE_DESCRIPTION_LOCAL_PROMPT,
+        "DOCLING_PICTURE_DESCRIPTION_API_URL": request.app.state.config.DOCLING_PICTURE_DESCRIPTION_API_URL,
+        "DOCLING_PICTURE_DESCRIPTION_API_MODEL": request.app.state.config.DOCLING_PICTURE_DESCRIPTION_API_MODEL,
+        "DOCLING_PICTURE_DESCRIPTION_API_PROMPT": request.app.state.config.DOCLING_PICTURE_DESCRIPTION_API_PROMPT,
         "DOCUMENT_INTELLIGENCE_ENDPOINT": request.app.state.config.DOCUMENT_INTELLIGENCE_ENDPOINT,
         "DOCUMENT_INTELLIGENCE_KEY": request.app.state.config.DOCUMENT_INTELLIGENCE_KEY,
         "MISTRAL_OCR_API_KEY": request.app.state.config.MISTRAL_OCR_API_KEY,
@@ -575,6 +582,13 @@ class ConfigForm(BaseModel):
     DOCLING_OCR_ENGINE: Optional[str] = None
     DOCLING_OCR_LANG: Optional[str] = None
     DOCLING_DO_PICTURE_DESCRIPTION: Optional[bool] = None
+    DOCLING_PICTURE_DESCRIPTION_MODE: Optional[str] = None
+    DOCLING_PICTURE_DESCRIPTION_LOCAL_REPO_ID: Optional[str] = None
+    DOCLING_PICTURE_DESCRIPTION_LOCAL_MAX_TOKENS: Optional[int] = None
+    DOCLING_PICTURE_DESCRIPTION_LOCAL_PROMPT: Optional[str] = None
+    DOCLING_PICTURE_DESCRIPTION_API_URL: Optional[str] = None
+    DOCLING_PICTURE_DESCRIPTION_API_MODEL: Optional[str] = None
+    DOCLING_PICTURE_DESCRIPTION_API_PROMPT: Optional[str] = None
     DOCUMENT_INTELLIGENCE_ENDPOINT: Optional[str] = None
     DOCUMENT_INTELLIGENCE_KEY: Optional[str] = None
     MISTRAL_OCR_API_KEY: Optional[str] = None
@@ -748,6 +762,42 @@ async def update_rag_config(
         else request.app.state.config.DOCLING_DO_PICTURE_DESCRIPTION
     )
 
+    request.app.state.config.DOCLING_PICTURE_DESCRIPTION_MODE = (
+        form_data.DOCLING_PICTURE_DESCRIPTION_MODE
+        if form_data.DOCLING_PICTURE_DESCRIPTION_MODE is not None
+        else request.app.state.config.DOCLING_PICTURE_DESCRIPTION_MODE
+    )
+    request.app.state.config.DOCLING_PICTURE_DESCRIPTION_LOCAL_REPO_ID = (
+        form_data.DOCLING_PICTURE_DESCRIPTION_LOCAL_REPO_ID
+        if form_data.DOCLING_PICTURE_DESCRIPTION_LOCAL_REPO_ID is not None
+        else request.app.state.config.DOCLING_PICTURE_DESCRIPTION_LOCAL_REPO_ID
+    )
+    request.app.state.config.DOCLING_PICTURE_DESCRIPTION_LOCAL_MAX_TOKENS = (
+        form_data.DOCLING_PICTURE_DESCRIPTION_LOCAL_MAX_TOKENS
+        if form_data.DOCLING_PICTURE_DESCRIPTION_LOCAL_MAX_TOKENS is not None
+        else request.app.state.config.DOCLING_PICTURE_DESCRIPTION_LOCAL_MAX_TOKENS
+    )
+    request.app.state.config.DOCLING_PICTURE_DESCRIPTION_LOCAL_PROMPT = (
+        form_data.DOCLING_PICTURE_DESCRIPTION_LOCAL_PROMPT
+        if form_data.DOCLING_PICTURE_DESCRIPTION_LOCAL_PROMPT is not None
+        else request.app.state.config.DOCLING_PICTURE_DESCRIPTION_LOCAL_PROMPT
+    )
+    request.app.state.config.DOCLING_PICTURE_DESCRIPTION_API_URL = (
+        form_data.DOCLING_PICTURE_DESCRIPTION_API_URL
+        if form_data.DOCLING_PICTURE_DESCRIPTION_API_URL is not None
+        else request.app.state.config.DOCLING_PICTURE_DESCRIPTION_API_URL
+    )
+    request.app.state.config.DOCLING_PICTURE_DESCRIPTION_API_MODEL = (
+        form_data.DOCLING_PICTURE_DESCRIPTION_API_MODEL
+        if form_data.DOCLING_PICTURE_DESCRIPTION_API_MODEL is not None
+        else request.app.state.config.DOCLING_PICTURE_DESCRIPTION_API_MODEL
+    )
+    request.app.state.config.DOCLING_PICTURE_DESCRIPTION_API_PROMPT = (
+        form_data.DOCLING_PICTURE_DESCRIPTION_API_PROMPT
+        if form_data.DOCLING_PICTURE_DESCRIPTION_API_PROMPT is not None
+        else request.app.state.config.DOCLING_PICTURE_DESCRIPTION_API_PROMPT
+    )
+
     request.app.state.config.DOCUMENT_INTELLIGENCE_ENDPOINT = (
         form_data.DOCUMENT_INTELLIGENCE_ENDPOINT
         if form_data.DOCUMENT_INTELLIGENCE_ENDPOINT is not None
@@ -985,6 +1035,13 @@ async def update_rag_config(
         "DOCLING_OCR_ENGINE": request.app.state.config.DOCLING_OCR_ENGINE,
         "DOCLING_OCR_LANG": request.app.state.config.DOCLING_OCR_LANG,
         "DOCLING_DO_PICTURE_DESCRIPTION": request.app.state.config.DOCLING_DO_PICTURE_DESCRIPTION,
+        "DOCLING_PICTURE_DESCRIPTION_MODE": request.app.state.config.DOCLING_PICTURE_DESCRIPTION_MODE,
+        "DOCLING_PICTURE_DESCRIPTION_LOCAL_REPO_ID": request.app.state.config.DOCLING_PICTURE_DESCRIPTION_LOCAL_REPO_ID,
+        "DOCLING_PICTURE_DESCRIPTION_LOCAL_MAX_TOKENS": request.app.state.config.DOCLING_PICTURE_DESCRIPTION_LOCAL_MAX_TOKENS,
+        "DOCLING_PICTURE_DESCRIPTION_LOCAL_PROMPT": request.app.state.config.DOCLING_PICTURE_DESCRIPTION_LOCAL_PROMPT,
+        "DOCLING_PICTURE_DESCRIPTION_API_URL": request.app.state.config.DOCLING_PICTURE_DESCRIPTION_API_URL,
+        "DOCLING_PICTURE_DESCRIPTION_API_MODEL": request.app.state.config.DOCLING_PICTURE_DESCRIPTION_API_MODEL,
+        "DOCLING_PICTURE_DESCRIPTION_API_PROMPT": request.app.state.config.DOCLING_PICTURE_DESCRIPTION_API_PROMPT,
         "DOCUMENT_INTELLIGENCE_ENDPOINT": request.app.state.config.DOCUMENT_INTELLIGENCE_ENDPOINT,
         "DOCUMENT_INTELLIGENCE_KEY": request.app.state.config.DOCUMENT_INTELLIGENCE_KEY,
         "MISTRAL_OCR_API_KEY": request.app.state.config.MISTRAL_OCR_API_KEY,
@@ -1334,6 +1391,13 @@ def process_file(
                     DOCLING_OCR_ENGINE=request.app.state.config.DOCLING_OCR_ENGINE,
                     DOCLING_OCR_LANG=request.app.state.config.DOCLING_OCR_LANG,
                     DOCLING_DO_PICTURE_DESCRIPTION=request.app.state.config.DOCLING_DO_PICTURE_DESCRIPTION,
+                    DOCLING_PICTURE_DESCRIPTION_MODE=request.app.state.config.DOCLING_PICTURE_DESCRIPTION_MODE,
+                    DOCLING_PICTURE_DESCRIPTION_LOCAL_REPO_ID=request.app.state.config.DOCLING_PICTURE_DESCRIPTION_LOCAL_REPO_ID,
+                    DOCLING_PICTURE_DESCRIPTION_LOCAL_MAX_TOKENS=request.app.state.config.DOCLING_PICTURE_DESCRIPTION_LOCAL_MAX_TOKENS,
+                    DOCLING_PICTURE_DESCRIPTION_LOCAL_PROMPT=request.app.state.config.DOCLING_PICTURE_DESCRIPTION_LOCAL_PROMPT,
+                    DOCLING_PICTURE_DESCRIPTION_API_URL=request.app.state.config.DOCLING_PICTURE_DESCRIPTION_API_URL,
+                    DOCLING_PICTURE_DESCRIPTION_API_MODEL=request.app.state.config.DOCLING_PICTURE_DESCRIPTION_API_MODEL,
+                    DOCLING_PICTURE_DESCRIPTION_API_PROMPT=request.app.state.config.DOCLING_PICTURE_DESCRIPTION_API_PROMPT,
                     PDF_EXTRACT_IMAGES=request.app.state.config.PDF_EXTRACT_IMAGES,
                     DOCUMENT_INTELLIGENCE_ENDPOINT=request.app.state.config.DOCUMENT_INTELLIGENCE_ENDPOINT,
                     DOCUMENT_INTELLIGENCE_KEY=request.app.state.config.DOCUMENT_INTELLIGENCE_KEY,

src/lib/components/admin/Settings/Documents.svelte

@@ -510,6 +510,140 @@
 									</div>
 								</div>
 							</div>
+							{#if RAGConfig.DOCLING_DO_PICTURE_DESCRIPTION}
+								<div class="flex w-full mt-2">
+									<div class="flex-1 flex items-center gap-4">
+										<label class="flex items-center gap-1 text-xs font-medium">
+											<Tooltip
+												content={$i18n.t('Use a model locally executed by Docling for picture description.')}
+												placement="top-start"
+											>
+												<input
+													type="radio"
+													name="picture-description-mode"
+													value="local"
+													bind:group={RAGConfig.DOCLING_PICTURE_DESCRIPTION_MODE}
+													checked={RAGConfig.DOCLING_PICTURE_DESCRIPTION_MODE === 'local'}
+												/>
+												<span style="padding-left: 0.5em">{$i18n.t('Local Description')}</span>
+											</Tooltip>
+										</label>
+										<label class="flex items-center gap-1 text-xs font-medium">
+											<Tooltip
+												content={$i18n.t('Use a remote API for picture description.')}
+												placement="top-start"
+											>
+												<input
+													type="radio"
+													name="picture-description-mode"
+													value="api"
+													bind:group={RAGConfig.DOCLING_PICTURE_DESCRIPTION_MODE}
+													checked={RAGConfig.DOCLING_PICTURE_DESCRIPTION_MODE === 'api'}
+												/>
+												<span style="padding-left: 0.5em">{$i18n.t('Remote Description')}</span>
+											</Tooltip>
+										</label>
+									</div>
+								</div>
+
+								{#if RAGConfig.DOCLING_PICTURE_DESCRIPTION_MODE === 'local'}
+									<div class="flex flex-col gap-2 mt-2 ml-4">
+										<div class="flex items-center gap-2">
+											<div class="min-w-fit text-xs font-medium">
+												<Tooltip
+													content={$i18n.t('The HuggingFace repo ID for the local vision-language model.')}
+													placement="top-start"
+												>
+													{$i18n.t('Repo ID')}
+												</Tooltip>
+											</div>
+											<input
+												class=" w-full rounded-lg py-1.5 px-4 text-sm bg-gray-50 dark:text-gray-300 dark:bg-gray-850 outline-hidden"
+												placeholder={$i18n.t('HuggingFaceTB/SmolVLM-256M-Instruct')}
+												bind:value={RAGConfig.DOCLING_PICTURE_DESCRIPTION_LOCAL_REPO_ID}
+											/>
+										</div>
+										<div class="flex items-center gap-2">
+											<div class="min-w-fit text-xs font-medium">
+												<Tooltip
+													content={$i18n.t('Maximum number of tokens for the generated description.')}
+													placement="top-start"
+												>
+													{$i18n.t('Max Tokens')}
+												</Tooltip>
+											</div>
+											<input
+												class=" w-full rounded-lg py-1.5 px-4 text-sm bg-gray-50 dark:text-gray-300 dark:bg-gray-850 outline-hidden"
+												placeholder={$i18n.t('200')}
+												bind:value={RAGConfig.DOCLING_PICTURE_DESCRIPTION_LOCAL_MAX_TOKENS}
+											/>
+										</div>
+										<div class="flex items-center gap-2">
+											<div class="min-w-fit text-xs font-medium">
+												<Tooltip
+													content={$i18n.t('Prompt to use for describing the image.')}
+													placement="top-start"
+												>
+													{$i18n.t('Prompt')}
+												</Tooltip>
+											</div>
+											<input
+												class=" w-full rounded-lg py-1.5 px-4 text-sm bg-gray-50 dark:text-gray-300 dark:bg-gray-850 outline-hidden"
+												placeholder={$i18n.t('Describe this image in a few sentences.')}
+												bind:value={RAGConfig.DOCLING_PICTURE_DESCRIPTION_LOCAL_PROMPT}
+											/>
+										</div>
+									</div>
+								{:else if RAGConfig.DOCLING_PICTURE_DESCRIPTION_MODE === 'api'}
+									<div class="flex flex-col gap-2 mt-2 ml-4">
+										<div class="flex items-center gap-2">
+											<div class="min-w-fit text-xs font-medium">
+												<Tooltip
+													content={$i18n.t('The remote API endpoint for picture description.')}
+													placement="top-start"
+												>
+													{$i18n.t('URL')}
+												</Tooltip>
+											</div>
+											<input
+												class=" w-full rounded-lg py-1.5 px-4 text-sm bg-gray-50 dark:text-gray-300 dark:bg-gray-850 outline-hidden"
+												placeholder={$i18n.t('Enter Remote API URL')}
+												bind:value={RAGConfig.DOCLING_PICTURE_DESCRIPTION_API_URL}
+											/>
+										</div>
+										<div class="flex items-center gap-2">
+											<div class="min-w-fit text-xs font-medium">
+												<Tooltip
+													content={$i18n.t('The model name to use for remote picture description.')}
+													placement="top-start"
+												>
+													{$i18n.t('Model')}
+												</Tooltip>
+											</div>
+											<input
+												class=" w-full rounded-lg py-1.5 px-4 text-sm bg-gray-50 dark:text-gray-300 dark:bg-gray-850 outline-hidden"
+												placeholder={$i18n.t('Enter Model Name')}
+												bind:value={RAGConfig.DOCLING_PICTURE_DESCRIPTION_API_MODEL}
+											/>
+										</div>
+										<div class="flex items-center gap-2">
+											<div class="min-w-fit text-xs font-medium">
+												<Tooltip
+													content={$i18n.t('Prompt to use for describing the image via remote API.')}
+													placement="top-start"
+												>
+													{$i18n.t('Prompt')}
+												</Tooltip>
+											</div>
+											<input
+												class=" w-full rounded-lg py-1.5 px-4 text-sm bg-gray-50 dark:text-gray-300 dark:bg-gray-850 outline-hidden"
+												placeholder={$i18n.t('Describe this image in a few sentences.')}
+												bind:value={RAGConfig.DOCLING_PICTURE_DESCRIPTION_API_PROMPT}
+											/>
+										</div>
+									</div>
+								{/if}
+							{/if}
 						{:else if RAGConfig.CONTENT_EXTRACTION_ENGINE === 'document_intelligence'}
 							<div class="my-0.5 flex gap-2 pr-2">
 								<input