fix: Normalize MIME type for PDF uploads and update collection management endpoint to use vector store API

services/aiknowledge_sync_utils.py

@@ -12,6 +12,7 @@ import hashlib
 import json
 from typing import Dict, Any, Optional, List, Tuple
 from datetime import datetime
+from urllib.parse import unquote
 
 from services.sync_utils_base import BaseSyncUtils
 from services.models import (

services/xai_service.py

@@ -63,14 +63,31 @@ class XAIService:
         Raises:
             RuntimeError: bei HTTP-Fehler oder fehlendem file_id in der Antwort
         """
-        self._log(f"📤 Uploading {len(file_content)} bytes to xAI: {filename}")
+        # Normalize MIME type: xAI needs correct Content-Type for proper processing
+        # If generic octet-stream but file is clearly a PDF, fix it
+        if mime_type == 'application/octet-stream' and filename.lower().endswith('.pdf'):
+            mime_type = 'application/pdf'
+            self._log(f"⚠️  Corrected MIME type to application/pdf for {filename}")
+        
+        self._log(f"📤 Uploading {len(file_content)} bytes to xAI: {filename} ({mime_type})")
 
         session = await self._get_session()
         url = f"{XAI_FILES_URL}/v1/files"
         headers = {"Authorization": f"Bearer {self.api_key}"}
 
-        form = aiohttp.FormData()
-        form.add_field('file', file_content, filename=filename, content_type=mime_type)
+        # Create multipart form with explicit UTF-8 filename encoding
+        # aiohttp automatically URL-encodes filenames with special chars,
+        # but xAI expects raw UTF-8 in the filename parameter
+        form = aiohttp.FormData(quote_fields=False)
+        form.add_field(
+            'file',
+            file_content,
+            filename=filename,
+            content_type=mime_type
+        )
+        # CRITICAL: purpose="file_search" enables proper PDF processing
+        # Without this, xAI throws "internal error" on complex PDFs
+        form.add_field('purpose', 'file_search')
 
         async with session.post(url, data=form, headers=headers) as response:
             try:
@@ -95,9 +112,12 @@ class XAIService:
 
     async def add_to_collection(self, collection_id: str, file_id: str) -> None:
         """
-        Fügt eine Datei einer xAI-Collection hinzu.
+        Fügt eine Datei einer xAI-Collection (Vector Store) hinzu.
 
-        POST https://management-api.x.ai/v1/collections/{collection_id}/documents/{file_id}
+        POST https://api.x.ai/v1/vector_stores/{vector_store_id}/files
+
+        Uses the OpenAI-compatible API pattern for adding files to vector stores.
+        This triggers proper indexing and processing.
 
         Raises:
             RuntimeError: bei HTTP-Fehler
@@ -105,13 +125,16 @@ class XAIService:
         self._log(f"📚 Adding file {file_id} to collection {collection_id}")
 
         session = await self._get_session()
-        url = f"{XAI_MANAGEMENT_URL}/v1/collections/{collection_id}/documents/{file_id}"
+        # Use the OpenAI-compatible endpoint (not management API)
+        url = f"{XAI_FILES_URL}/v1/vector_stores/{collection_id}/files"
         headers = {
-            "Authorization": f"Bearer {self.management_key}",
+            "Authorization": f"Bearer {self.api_key}",
             "Content-Type": "application/json",
         }
 
-        async with session.post(url, headers=headers) as response:
+        payload = {"file_id": file_id}
+
+        async with session.post(url, json=payload, headers=headers) as response:
             if response.status not in (200, 201):
                 raw = await response.text()
                 raise RuntimeError(