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

app/utils/chat_logger.py

@@ -175,21 +175,59 @@ class ChatLogger:
             # Get file size
             file_size = self.log_file.stat().st_size
 
-            # Count by character
+            # Count stats
             character_counts = {}
+            total_tokens = 0
+            tokens_by_model = {}
+            tokens_by_role = {}
+            context_found_count = 0
+            
+            # Simple pricing model (Blended average for OpenRouter)
+            # Input: ~$5/M, Output: ~$15/M -> Avg ~$10/M = $0.00001 per token
+            PRICE_PER_TOKEN = 0.00001 
+
             for line in lines:
                 try:
                     entry = json.loads(line)
+                    
+                    # Character stats
                     char_id = entry.get('character', {}).get('id', 'unknown')
                     character_counts[char_id] = character_counts.get(char_id, 0) + 1
+                    
+                    # Token stats
+                    tokens = entry.get('tokens_estimated', 0)
+                    total_tokens += tokens
+                    
+                    # Model stats
+                    model = entry.get('ai', {}).get('model', 'unknown')
+                    tokens_by_model[model] = tokens_by_model.get(model, 0) + tokens
+                    
+                    # Role stats
+                    tokens_by_role[char_id] = tokens_by_role.get(char_id, 0) + tokens
+                    
+                    # RAG stats
+                    if entry.get('rag', {}).get('context_found'):
+                        context_found_count += 1
+                    
                 except json.JSONDecodeError:
                     continue
 
+            total_interactions = len(lines)
+            context_hit_rate = round((context_found_count / total_interactions * 100), 1) if total_interactions > 0 else 0
+            estimated_cost = round(total_tokens * PRICE_PER_TOKEN, 4)
+
             return {
-                "total_interactions": len(lines),
+                "total_interactions": total_interactions,
+                "total_tokens_estimated": total_tokens,
+                "estimated_cost_usd": estimated_cost,
+                "context_found_count": context_found_count,
+                "context_hit_rate_percent": context_hit_rate,
                 "file_size_bytes": file_size,
                 "file_size_mb": round(file_size / (1024 * 1024), 2),
-                "characters": character_counts
+                "characters": character_counts,
+                "tokens_by_model": tokens_by_model,
+                "tokens_by_role": tokens_by_role,
+                "last_updated": datetime.utcnow().isoformat() + "Z"
             }
 
         except Exception as e:

app/utils/rag_chat.py

@@ -106,7 +106,8 @@ class RAGChatService:
         character_name: str,
         character_prompt: str,
         context_limit: int = 3,
-        lang: str = "de"
+        lang: str = "de",
+        model_override: Optional[str] = None
     ) -> Dict[str, Any]:
         """
         Answer a question using RAG with character persona.
@@ -117,6 +118,7 @@ class RAGChatService:
             character_prompt: Character system prompt
             context_limit: Number of context chunks to retrieve
             lang: Language code
+            model_override: Optional model name to override default
 
         Returns:
             Dictionary with answer and sources
@@ -156,7 +158,8 @@ class RAGChatService:
         # Generate answer with character persona
         answer = self._get_completion_with_system(
             system_message=system_message,
-            user_message=question
+            user_message=question,
+            model_override=model_override
         )
 
         # Format sources
@@ -175,13 +178,14 @@ class RAGChatService:
             'sources': sources,
             'context_found': len(search_results) > 0,
             'provider': self.completion_provider.provider_name,
-            'model': self.completion_provider.model_name
+            'model': model_override or self.completion_provider.model_name
         }
 
     def _get_completion_with_system(
         self,
         system_message: str,
-        user_message: str
+        user_message: str,
+        model_override: Optional[str] = None
     ) -> str:
         """
         Get completion with explicit system message.
@@ -190,6 +194,7 @@ class RAGChatService:
         Args:
             system_message: System prompt with character + context
             user_message: User's question
+            model_override: Optional model name to use
 
         Returns:
             Generated response
@@ -200,7 +205,11 @@ class RAGChatService:
 
         # Call completion provider
         # Note: We pass the system message as context
+        # Assuming provider.get_completion supports model argument?
+        # If not, we might need to check BaseProvider. 
+        # But get_completion usually takes keyword args passed to API.
         return self.completion_provider.get_completion(
             prompt=user_message,
-            context=system_message
+            context=system_message,
+            model=model_override
         )