feat: Update CitationsModal to use source_url for YouTube documents and file_id for others

backend/open_webui/retrieval/loaders/youtube.py

@@ -78,6 +78,7 @@ class YoutubeLoader:
             # First try using YouTube Data API v3 if available
             try:
                 from open_webui.config import YOUTUBE_API_KEY
+
                 if YOUTUBE_API_KEY:
                     url = f"https://www.googleapis.com/youtube/v3/videos?id={self.video_id}&key={YOUTUBE_API_KEY}&part=snippet"
                     response = requests.get(url)
@@ -93,7 +94,8 @@ class YoutubeLoader:
             response = requests.get(url)
             if response.status_code == 200:
                 import re
-                title_match = re.search(r'<title>(.+?)</title>', response.text)
+
+                title_match = re.search(r"<title>(.+?)</title>", response.text)
                 if title_match:
                     title = title_match.group(1)
                     return title
@@ -139,10 +141,14 @@ class YoutubeLoader:
             transcript = transcript_list.find_transcript(self.language)
         except NoTranscriptFound:
             # Fallback: try to get any available transcript
-            available_transcripts = list(transcript_list._generated_transcripts.values())
+            available_transcripts = list(
+                transcript_list._generated_transcripts.values()
+            )
             if available_transcripts:
                 transcript = available_transcripts[0]
-                log.info(f"Using first available transcript in language: {transcript.language_code}")
+                log.info(
+                    f"Using first available transcript in language: {transcript.language_code}"
+                )
             else:
                 log.error("No transcripts found for video")
                 return []
@@ -161,27 +167,29 @@ class YoutubeLoader:
         # Combine pieces into a single text while tracking timestamp positions
         full_text = ""
         timestamp_map = []
-        
+
         for piece in transcript_pieces:
             start_char = len(full_text)
             text = piece["text"].strip()
             full_text += text + " "
             end_char = len(full_text)
-            
+
-            timestamp_map.append({
+            timestamp_map.append(
-                "start": start_char,
+                {
-                "end": end_char,
+                    "start": start_char,
-                "time": piece["start"],
+                    "end": end_char,
-                "duration": piece["duration"]
+                    "time": piece["start"],
-            })
+                    "duration": piece["duration"],
+                }
+            )
 
         # Create a single document that will be split by Langchain's text splitter
         doc = Document(
             page_content=full_text.strip(),
             metadata={
                 **self._metadata,
-                "timestamp_map": timestamp_map  # Store timestamp mapping in metadata
+                "timestamp_map": timestamp_map,  # Store timestamp mapping in metadata
-            }
+            },
         )
 
         return [doc]

backend/open_webui/routers/retrieval.py

@@ -152,15 +152,15 @@ def get_rf(
                 raise Exception(ERROR_MESSAGES.DEFAULT("CrossEncoder error"))
     return rf
 
+
 def add_timestamp_to_youtube_url(url: str, timestamp: int) -> str:
     parsed = urlparse(url)
     query_dict = parse_qs(parsed.query)
-    query_dict['t'] = [str(timestamp)]
+    query_dict["t"] = [str(timestamp)]
     new_query = urlencode(query_dict, doseq=True)
     return urlunparse(parsed._replace(query=new_query))
 
 
-
 ##########################################
 #
 # API routes
@@ -662,7 +662,9 @@ async def update_query_settings(
 ####################################
 
 
-def interpolate_timestamp(chunk_start: int, chunk_end: int, timestamp_map: List[dict]) -> Tuple[float, float]:
+def interpolate_timestamp(
+    chunk_start: int, chunk_end: int, timestamp_map: List[dict]
+) -> Tuple[float, float]:
     """
     Find the appropriate timestamp for a chunk based on its character position
     Returns (start_time, end_time) as floats in seconds
@@ -675,7 +677,8 @@ def interpolate_timestamp(chunk_start: int, chunk_end: int, timestamp_map: List[
     else:
         # If not found, use the closest previous timestamp
         start_time = min(
-            [e["time"] for e in timestamp_map if e["start"] <= chunk_start], default=0)
+            [e["time"] for e in timestamp_map if e["start"] <= chunk_start], default=0
+        )
 
     # Find the timestamp entry that contains the end of our chunk
     for entry in reversed(timestamp_map):
@@ -684,11 +687,14 @@ def interpolate_timestamp(chunk_start: int, chunk_end: int, timestamp_map: List[
             break
     else:
         # If not found, use the closest next timestamp
-        end_time = max([e["time"] + e["duration"]
+        end_time = max(
-                       for e in timestamp_map if e["end"] >= chunk_end], default=start_time)
+            [e["time"] + e["duration"] for e in timestamp_map if e["end"] >= chunk_end],
+            default=start_time,
+        )
 
     return start_time, end_time
 
+
 def save_docs_to_vector_db(
     request: Request,
     docs,
@@ -733,12 +739,12 @@ def save_docs_to_vector_db(
 
     if split:
         # Check if this is a YouTube document by looking at the first doc's metadata
-        is_youtube = (len(docs) == 1 and
+        is_youtube = len(docs) == 1 and docs[0].metadata.get("type") == "youtube"
-                      docs[0].metadata.get("type") == "youtube")
 
         # Store timestamp_map before splitting if it's a YouTube document
-        original_timestamp_map = docs[0].metadata.get(
+        original_timestamp_map = (
-            "timestamp_map") if is_youtube else None
+            docs[0].metadata.get("timestamp_map") if is_youtube else None
+        )
 
         if request.app.state.config.TEXT_SPLITTER in ["", "character"]:
             text_splitter = RecursiveCharacterTextSplitter(
@@ -770,15 +776,17 @@ def save_docs_to_vector_db(
                 end_index = start_index + len(doc.page_content)
 
                 start_time, end_time = interpolate_timestamp(
-                    start_index,
+                    start_index, end_index, original_timestamp_map
-                    end_index,
-                    original_timestamp_map
                 )
 
-                doc.metadata.update({
+                doc.metadata.update(
-                    "start_time": start_time,
+                    {
-                    "source_url": add_timestamp_to_youtube_url(doc.metadata['source_url'], int(start_time))
+                        "start_time": start_time,
-                })
+                        "source_url": add_timestamp_to_youtube_url(
+                            doc.metadata["source_url"], int(start_time)
+                        ),
+                    }
+                )
 
                 # Remove the timestamp_map from individual chunks
                 doc.metadata.pop("timestamp_map", None)
@@ -915,10 +923,7 @@ def process_file(
                 "status": True,
                 "collection_name": form_data.collection_name,
                 "content": content,
-                "file": {
+                "file": {"id": file.id, "meta": metadata},
-                    "id": file.id,
-                    "meta": metadata
-                }
             }
 
         collection_name = form_data.collection_name
@@ -1148,7 +1153,7 @@ def process_youtube_video(
         content = " ".join([doc.page_content for doc in docs])
         log.debug(f"text_content: {content}")
 
-                # Get video title from metadata or fallback to URL
+        # Get video title from metadata or fallback to URL
         video_title = docs[0].metadata.get("title", form_data.url)
 
         # Create a unique file ID for this video
@@ -1168,11 +1173,9 @@ def process_youtube_video(
                         "size": len(content),
                         "source": form_data.url,
                         "source_url": add_timestamp_to_youtube_url(form_data.url, 0),
-                        "type": "youtube"
+                        "type": "youtube",
                     },
-                    "data": {
+                    "data": {"content": content},
-                        "content": content
-                    }
                 }
             ),
         )
@@ -1185,7 +1188,7 @@ def process_youtube_video(
             "type": "youtube",
             "name": video_title,
             "file_id": file_id,
-            "created_by": user.id if user else None
+            "created_by": user.id if user else None,
         }
 
         # Update all docs with the file metadata
@@ -1194,18 +1197,20 @@ def process_youtube_video(
             # Debug log
             log.info(f"Document metadata before saving: {doc.metadata}")
 
-        save_docs_to_vector_db(request, docs, collection_name, overwrite=False, add=True)
+        save_docs_to_vector_db(
+            request, docs, collection_name, overwrite=False, add=True
+        )
 
         return {
             "status": True,
             "collection_name": collection_name,
             "id": file_id,  # Return the file ID directly
-             "filename": video_title,
+            "filename": video_title,
             "file": {
                 "data": {
                     "content": content,
                 },
-                "meta": file_metadata
+                "meta": file_metadata,
             },
         }
     except Exception as e:

package-lock.json

@@ -12589,4 +12589,4 @@
 			}
 		}
 	}
-}
+}

package.json

@@ -110,4 +110,4 @@
 		"node": ">=18.13.0 <=22.x.x",
 		"npm": ">=6.0.0"
 	}
-}
+}

src/lib/components/chat/Messages/CitationsModal.svelte

@@ -91,7 +91,9 @@
 								<div class="text-sm dark:text-gray-400 flex items-center gap-2 w-fit">
 									<a
 										class="hover:text-gray-500 hover:dark:text-gray-100 underline flex-grow"
-										href={document?.metadata?.file_id
+										href={document?.metadata?.type === 'youtube'
+										? document?.metadata?.source_url
+										: document?.metadata?.file_id
 											? `${WEBUI_API_BASE_URL}/files/${document?.metadata?.file_id}/content${document?.metadata?.page !== undefined ? `#page=${document.metadata.page + 1}` : ''}`
 											: document.source?.url?.includes('http')
 												? document.source.url

src/lib/i18n/locales/fr-FR/translation.json

@@ -1062,4 +1062,4 @@
 	"Your entire contribution will go directly to the plugin developer; Open WebUI does not take any percentage. However, the chosen funding platform might have its own fees.": "L'intégralité de votre contribution ira directement au développeur du plugin ; Open WebUI ne prend aucun pourcentage. Cependant, la plateforme de financement choisie peut avoir ses propres frais.",
 	"Youtube": "YouTube",
 	"Youtube Loader Settings": "Paramètres de l'outil de téléchargement YouTube"
-}
+}

src/lib/i18n/locales/ie-GA/translation.json

@@ -1062,4 +1062,4 @@
 	"Your entire contribution will go directly to the plugin developer; Open WebUI does not take any percentage. However, the chosen funding platform might have its own fees.": "Rachaidh do ranníocaíocht iomlán go díreach chuig an bhforbróir breiseán; Ní ghlacann Open WebUI aon chéatadán. Mar sin féin, d'fhéadfadh a tháillí féin a bheith ag an ardán maoinithe roghnaithe.",
 	"Youtube": "Youtube",
 	"Youtube Loader Settings": "Socruithe Luchtaire Youtube"
-}
+}

src/lib/i18n/locales/zh-CN/translation.json

@@ -1062,4 +1062,4 @@
 	"Your entire contribution will go directly to the plugin developer; Open WebUI does not take any percentage. However, the chosen funding platform might have its own fees.": "您的全部捐款将直接给到插件开发者，Open WebUI 不会收取任何比例。但众筹平台可能会有服务费、抽成。",
 	"Youtube": "YouTube",
 	"Youtube Loader Settings": "YouTube 爬取设置"
-}
+}