feat: Fix vector indexing stability, add Gitea linking, enhance admin dashboard

app/lib/embedding_providers/openrouter_provider.py

@@ -96,13 +96,14 @@ class OpenRouterProvider(BaseProvider):
         except KeyError as e:
             raise RuntimeError(f"Unexpected OpenRouter API response format: {str(e)}")
 
-    def get_completion(self, prompt: str, context: str = "") -> str:
+    def get_completion(self, prompt: str, context: str = "", model: str = None) -> str:
         """
         Generate a completion using OpenRouter API.
 
         Args:
             prompt: The user's question
             context: Optional context from retrieved documents
+            model: Optional model to override default
 
         Returns:
             The generated response
@@ -124,11 +125,20 @@ class OpenRouterProvider(BaseProvider):
         messages = []
 
         # Add system message with context if provided
+        # NOTE: If detailed system prompt is passed as context (like in RAGChatService), 
+        # we should treat it as system prompt, not just "Answer based on..." wrapper.
         if context:
-            messages.append({
-                "role": "system",
-                "content": f"Answer the user's question based on the following context:\n\n{context}"
-            })
+            # Check if context looks like a full system prompt or just data
+            if "You are" in context or "Du bist" in context or "Tu es" in context:
+                 messages.append({
+                    "role": "system",
+                    "content": context
+                })
+            else:
+                messages.append({
+                    "role": "system",
+                    "content": f"Answer the user's question based on the following context:\n\n{context}"
+                })
 
         # Add user message
         messages.append({
@@ -137,7 +147,7 @@ class OpenRouterProvider(BaseProvider):
         })
 
         data = {
-            "model": self.completion_model,
+            "model": model or self.completion_model,
             "messages": messages,
             "max_tokens": 2000,
             "temperature": 0.7