Update 5 files

- /rag/config_models.py
- /rag/utils/config_loader.py
- /rag/rag_orchestrator.py
- /rag/rag_server.py
- /rag/rag_api.py

rag/rag_orchestrator.py

@@ -8,19 +8,19 @@
 import logging
 from typing import Optional, Dict, List, Any
 
-from rag.services.postgres_service import PostgresService
-from rag.services.qdrant_service import QdrantService
-from rag.services.embedding_service import EmbeddingService
-from rag.services.kb_service import KBService
-from rag.services.giga_client import GigaClient
-from rag.services.file_service import FileService
-from rag.history_manager import HistoryManager
-from rag.intent_router import IntentRouter
-from rag.query_processor import QueryProcessor
-from rag.indexing_manager import IndexingManager
-from rag.functions.intent_classify import classify_intent
-from rag.utils.text_utils import count_tokens
-from rag.config_models import AppConfig
+from .services.postgres_service import PostgresService
+from .services.qdrant_service import QdrantService
+from .services.embedding_service import EmbeddingService
+from .services.kb_service import KBService
+from .services.giga_client import GigaClient
+from .services.file_service import FileService
+from .history_manager import HistoryManager
+from .intent_router import IntentRouter
+from .query_processor import QueryProcessor
+from .indexing_manager import IndexingManager
+from .functions.intent_classify import classify_intent
+from .utils.text_utils import count_tokens
+from .config_models import AppConfig
 
 logger = logging.getLogger(__name__)
 
@@ -183,9 +183,9 @@ class RAGOrchestrator:
         system_prompt = prompts.get('system', None)
 
         # 2. Расчёт лимитов токенов
-        max_model_tokens = self.config.rag.max_model_tokens
-        reserved_for_answer = self.config.rag.reserved_for_answer_tokens
-        reserved_for_overhead = self.config.rag.reserved_for_overhead_tokens
+        max_model_tokens = self.config.max_model_tokens
+        reserved_for_answer = self.config.reserved_for_answer_tokens
+        reserved_for_overhead = self.config.reserved_for_overhead_tokens
 
         token_info = self._prepare_prompt_parts(
             synthesis_template=synthesis_template,
@@ -211,7 +211,7 @@ class RAGOrchestrator:
 
         # 5. Классификация намерений
         intent = intent_override
-        if intent is None and self.config.features.enable_intent_classification:
+        if intent is None and self.config.enable_intent_classification:
             intent_prompt = prompts.get('intent', '')
             if intent_prompt:
                 intent = await classify_intent(
@@ -226,7 +226,7 @@ class RAGOrchestrator:
             intent = intent or "GENERAL"
 
         # 6. Принудительная установка SURGICAL по ключевым словам
-        keywords = self.config.features.surgical_keywords
+        keywords = self.config.surgical_keywords
         if any(kw in query.lower() for kw in keywords) and last_file_path:
             intent = "SURGICAL"
             logger.info(f"Принудительный Intent: SURGICAL (есть файл и ключевые слова)")
@@ -255,7 +255,6 @@ class RAGOrchestrator:
             sources = router_result.get("sources", [])
         else:
             # Обычный RAG-пайплайн (GENERAL, FACT, PROCEDURE, COMPARISON, CALCULATION)
-            # Вычисляем, сколько токенов осталось для контекста после истории
             history_tokens = sum(count_tokens(msg['content']) for msg in formatted_history)
             available_for_context = available_for_history_and_context - history_tokens
             available_for_context = max(available_for_context, 0)