fix: enhance CSV ticket handling and update LLM backend initialization

backend/agents.py

@@ -56,7 +56,6 @@ def _env_int(name: str, default: int) -> int:
 
 # Local CSV service
 from csv_data import get_csv_ticket_service
-
 from langchain_core.tools import StructuredTool
 
 # Third-party - LangGraph
@@ -186,10 +185,11 @@ def __init__(self):
         """
         Initialize the agent service.
 
-        Uses OpenAI when OPENAI_API_KEY is set, otherwise falls back
-        to LiteLLM (supports GitHub Copilot, Ollama, etc.).
+        Defaults to LiteLLM with GitHub Copilot backend.
+        Set AGENT_BACKEND=openai to force OpenAI SDK (requires OPENAI_API_KEY).
         """
-        if OPENAI_API_KEY:
+        force_openai = os.getenv("AGENT_BACKEND", "").lower() == "openai"
+        if force_openai and OPENAI_API_KEY:
             from langchain_openai import ChatOpenAI
             self.llm = ChatOpenAI(
                 model=OPENAI_MODEL,

backend/app.py

@@ -37,12 +37,19 @@
 # CSV ticket service
 from csv_data import Ticket, get_csv_ticket_service
 
+# FastMCP client for direct ticket MCP calls (no AI)
+from fastmcp import Client as MCPClient
+
 # KBA Drafter
 from kba_exceptions import (
     DraftNotFoundError,
     DuplicateKBADraftError,
     InvalidLLMOutputError,
     InvalidStatusError,
+    LLMAuthenticationError,
+    LLMRateLimitError,
+    LLMTimeoutError,
+    LLMUnavailableError,
     PublishFailedError,
     TicketNotFoundError,
 )
@@ -53,15 +60,6 @@
     KBADraftUpdate,
     KBAPublishRequest,
 )
-from kba_exceptions import (
-    LLMTimeoutError,
-    LLMUnavailableError,
-    LLMRateLimitError,
-    LLMAuthenticationError,
-)
-
-# FastMCP client for direct ticket MCP calls (no AI)
-from fastmcp import Client as MCPClient
 from mcp_handler import handle_mcp_request
 from operations import (
     CSV_TICKET_FIELDS,
@@ -111,8 +109,9 @@
 @app.before_serving
 async def startup():
     """Initialize scheduler on application startup"""
-    from scheduler import start_scheduler
     import logging
+
+    from scheduler import start_scheduler
 
     logger = logging.getLogger(__name__)
     logger.info("Starting auto-generation scheduler...")
@@ -127,8 +126,9 @@ async def startup():
 @app.after_serving
 async def shutdown():
     """Cleanup scheduler on application shutdown"""
-    from scheduler import stop_scheduler
     import logging
+
+    from scheduler import stop_scheduler
 
     logger = logging.getLogger(__name__)
     logger.info("Stopping auto-generation scheduler...")
@@ -585,6 +585,12 @@ async def get_qa_tickets():
 if _csv_data_path.exists():
     _csv_loaded = _csv_ticket_service.load_csv(_csv_data_path)
     print(f"📊 Loaded {_csv_loaded} tickets from CSV")
+else:
+    print(
+        f"⚠️  CSV data file not found: {_csv_data_path.resolve()}\n"
+        f"   Ticket features will be unavailable.\n"
+        f"   To fix: place your BMC Remedy/ITSM CSV export at csv/data.csv"
+    )
 
 
 @app.route("/api/csv-tickets/fields", methods=["GET"])
@@ -974,6 +980,7 @@ async def rest_kba_publish_draft(draft_id: str):
 async def rest_kba_list_drafts():
     """REST wrapper: list KBA drafts with filtering."""
     from operations import op_kba_list_drafts
+
     # Parse query parameters
     filters = KBADraftFilter(
         status=request.args.get("status"),
@@ -1039,8 +1046,8 @@ async def rest_kba_get_auto_gen_settings():
 @app.route("/api/kba/auto-gen/settings", methods=["PATCH"])
 async def rest_kba_update_auto_gen_settings():
     """REST wrapper: update auto-generation settings."""
-    from operations import op_kba_update_auto_gen_settings
     from auto_gen_models import AutoGenSettingsUpdate
+    from operations import op_kba_update_auto_gen_settings
 
     data = await request.get_json()
     updates = AutoGenSettingsUpdate(**data)

backend/llm_service.py

@@ -30,14 +30,13 @@
 import os
 from typing import Any, Optional, Type
 
-from pydantic import BaseModel
-
 from kba_exceptions import (
-    LLMUnavailableError,
-    LLMTimeoutError,
+    LLMAuthenticationError,
     LLMRateLimitError,
-    LLMAuthenticationError
+    LLMTimeoutError,
+    LLMUnavailableError,
 )
+from pydantic import BaseModel
 
 logger = logging.getLogger(__name__)
 
@@ -89,7 +88,8 @@ def __init__(
         """
         self.timeout = timeout
 
-        # Determine backend: LiteLLM is the default, OpenAI only when forced
+        # Determine backend: LiteLLM with Copilot is the default.
+        # OpenAI only when explicitly forced via backend="openai".
         resolved_api_key = api_key or OPENAI_API_KEY
         if backend == "openai":
             self._backend = "openai"
@@ -112,9 +112,9 @@ def __init__(
                 client_kwargs["base_url"] = self.base_url
             self._client = AsyncOpenAI(**client_kwargs)
         else:
-            # LiteLLM backend
+            # LiteLLM backend — always default to Copilot model
             self.api_key = resolved_api_key or None
-            self.model = model or (OPENAI_MODEL if resolved_api_key else LITELLM_MODEL)
+            self.model = model or LITELLM_MODEL
             self.base_url = base_url or OPENAI_BASE_URL or None
             self._client = None
             # Build fallback chain: primary model + configured fallbacks (deduplicated)
@@ -295,9 +295,9 @@ def _handle_openai_error(self, error: Exception) -> Exception:
         from openai import (
             APIConnectionError,
             APITimeoutError,
-            RateLimitError,
             AuthenticationError,
-            BadRequestError
+            BadRequestError,
+            RateLimitError,
         )
 
         if isinstance(error, APITimeoutError):

backend/operations.py

@@ -85,8 +85,9 @@ def _get_kba_session() -> Session:
     global _kba_db_engine, _kba_session
     if _kba_db_engine is None:
         from pathlib import Path
-        from kba_models import KBADraftTable, KBAAuditLog
+
         from auto_gen_models import AutoGenSettingsTable
+        from kba_models import KBAAuditLog, KBADraftTable
         from sqlmodel import SQLModel
 
         db_path = Path(__file__).parent / "data" / "kba.db"
@@ -162,8 +163,18 @@ def _ensure_csv_loaded() -> None:
     if default_csv_path.exists():
         try:
             _csv_service.load_csv(default_csv_path)
-        except Exception:
-            pass
+        except Exception as exc:
+            import logging
+            logging.getLogger(__name__).warning(
+                "Failed to load CSV data from %s: %s", default_csv_path, exc
+            )
+    else:
+        import logging
+        logging.getLogger(__name__).warning(
+            "CSV data file not found: %s — ticket operations will return empty results. "
+            "Place your BMC Remedy/ITSM CSV export at csv/data.csv to enable ticket features.",
+            default_csv_path.resolve(),
+        )
     _csv_loaded = True