守住 AI provider Ollama-first

services/ai_provider.py

@@ -2,7 +2,7 @@
 # -*- coding: utf-8 -*-
 """
 AI 提供者抽象層
-統一 Ollama 和 Gemini 的介面，支援動態切換 AI 提供者
+統一 Ollama 和 Gemini 的介面；通用生成一律 Ollama-first，Gemini 僅作備援
 """
 
 import os
@@ -13,8 +13,8 @@ from datetime import date, datetime
 
 logger = logging.getLogger(__name__)
 
-# AI 提供者設定
-AI_PROVIDER = os.getenv('AI_PROVIDER', 'ollama')  # 預設使用 Ollama
+# AI 提供者設定：Gemini 不可作為預設，只能在 Ollama 失敗後備援
+AI_PROVIDER = os.getenv('AI_PROVIDER', 'ollama')
 
 # 引入服務
 from .ollama_service import OllamaService, OllamaResponse
@@ -60,7 +60,8 @@ class AIProviderService:
     """
     AI 提供者服務 - 統一的 AI 介面
 
-    支援動態切換 Ollama 和 Gemini，並提供統一的生成介面
+    通用文案 / 關鍵字 / 洞察生成一律先走 Ollama 三主機級聯。
+    Gemini 只在 Ollama 主路徑失敗時作為 fallback，不允許被設為預設主路徑。
     """
 
     def __init__(self, default_provider: str = None):
@@ -68,9 +69,9 @@ class AIProviderService:
         初始化 AI 提供者服務
 
         Args:
-            default_provider: 預設提供者 ('ollama' 或 'gemini')
+            default_provider: 預設提供者（Gemini 會被降回 Ollama）
         """
-        self._default_provider = default_provider or AI_PROVIDER
+        self._default_provider = self._sanitize_default_provider(default_provider or AI_PROVIDER)
         self._ollama = OllamaService()
         self._gemini = GeminiService()
         self._elephant = ElephantService()
@@ -87,11 +88,20 @@ class AIProviderService:
     @default_provider.setter
     def default_provider(self, value: str):
         """設定預設提供者"""
-        if value not in ('ollama', 'gemini', 'elephant'):
-            raise ValueError("Provider must be 'ollama', 'gemini' or 'elephant'")
-        self._default_provider = value
+        self._default_provider = self._sanitize_default_provider(value)
         logger.info(f"AI 預設提供者已切換至: {value}")
 
+    @staticmethod
+    def _sanitize_default_provider(value: str) -> str:
+        """Gemini 是 fallback-only；任何預設 Gemini 設定都強制回 Ollama。"""
+        normalized = (value or 'ollama').strip().lower()
+        if normalized == 'gemini':
+            logger.warning("AI_PROVIDER=gemini 已被拒絕；Gemini 僅作 Ollama 失敗備援")
+            return 'ollama'
+        if normalized not in ('ollama', 'elephant'):
+            raise ValueError("Provider must be 'ollama' or 'elephant'; gemini is fallback-only")
+        return normalized
+
     def get_status(self, force_refresh: bool = False) -> Dict[str, Any]:
         """
         取得所有 AI 服務的狀態
@@ -148,18 +158,15 @@ class AIProviderService:
 
     def _get_recommended_provider(self, ollama_ok: bool, gemini_ok: bool, elephant_ok: bool) -> str:
         """根據可用性推薦提供者"""
+        if ollama_ok:
+            return 'ollama'
         if self._default_provider == 'elephant' and elephant_ok:
             return 'elephant'
-        if ollama_ok and gemini_ok:
-            return self._default_provider
-        elif elephant_ok:
-            return 'elephant'
-        elif ollama_ok:
-            return 'ollama'
-        elif gemini_ok:
+        if gemini_ok:
             return 'gemini'
-        else:
-            return 'none'
+        if elephant_ok:
+            return 'elephant'
+        return 'none'
 
     def _convert_response(self, response: Union[OllamaResponse, GeminiResponse],
                           provider: str) -> AIResponse:
@@ -218,6 +225,21 @@ class AIProviderService:
                 error='Unknown response type'
             )
 
+    def _gemini_fallback(self, ollama_result: AIResponse, fallback_call) -> AIResponse:
+        """Ollama 失敗時才呼叫 Gemini，讓通用 AI 入口符合 Ollama-first。"""
+        if ollama_result.success:
+            return ollama_result
+        logger.warning("Ollama 主路徑失敗，啟用 Gemini 備援：%s", ollama_result.error)
+        try:
+            gemini_response = fallback_call()
+            gemini_result = self._convert_response(gemini_response, 'gemini')
+            if not gemini_result.success and ollama_result.error:
+                gemini_result.error = f"Ollama 失敗：{ollama_result.error}；Gemini 備援失敗：{gemini_result.error}"
+            return gemini_result
+        except Exception as exc:
+            ollama_result.error = f"{ollama_result.error}; Gemini 備援例外：{exc}"
+            return ollama_result
+
     def generate(self, prompt: str, provider: str = None, model: str = None,
                  system_prompt: str = None, temperature: float = 0.7,
                  timeout: int = None) -> AIResponse:
@@ -226,7 +248,7 @@ class AIProviderService:
 
         Args:
             prompt: 使用者提示
-            provider: 指定提供者 ('ollama' 或 'gemini')
+            provider: 指定提供者；Gemini 只會在 Ollama 失敗後備援
             model: 指定模型
             system_prompt: 系統提示
             temperature: 創意度
@@ -235,17 +257,9 @@ class AIProviderService:
         Returns:
             AIResponse
         """
-        provider = provider or self._default_provider
+        provider = (provider or self._default_provider or 'ollama').strip().lower()
 
-        if provider == 'gemini':
-            response = self._gemini.generate(
-                prompt=prompt,
-                model=model,
-                system_prompt=system_prompt,
-                temperature=temperature,
-                timeout=timeout
-            )
-        elif provider == 'elephant':
+        if provider == 'elephant':
             response = self._elephant.generate(
                 prompt=prompt,
                 model=model,
@@ -253,16 +267,27 @@ class AIProviderService:
                 temperature=temperature,
                 timeout=timeout
             )
-        else:  # ollama
-            response = self._ollama.generate(
+
+            return self._convert_response(response, provider)
+
+        ollama_response = self._ollama.generate(
+            prompt=prompt,
+            model=model,
+            system_prompt=system_prompt,
+            temperature=temperature,
+            timeout=timeout
+        )
+        ollama_result = self._convert_response(ollama_response, 'ollama')
+        return self._gemini_fallback(
+            ollama_result,
+            lambda: self._gemini.generate(
                 prompt=prompt,
                 model=model,
                 system_prompt=system_prompt,
                 temperature=temperature,
                 timeout=timeout
-            )
-
-        return self._convert_response(response, provider)
+            ),
+        )
 
     def generate_sales_copy(self, product_name: str, provider: str = None,
                            model: str = None, trend_keywords: List[str] = None,
@@ -283,27 +308,27 @@ class AIProviderService:
         Returns:
             AIResponse
         """
-        provider = provider or self._default_provider
+        provider = (provider or self._default_provider or 'ollama').strip().lower()
 
-        if provider == 'gemini':
-            response = self._gemini.generate_sales_copy(
+        ollama_response = self._ollama.generate_sales_copy(
+            product_name=product_name,
+            trend_keywords=trend_keywords,
+            style=style,
+            upcoming_holidays=upcoming_holidays,
+            bestseller_products=bestseller_products
+        )
+        ollama_result = self._convert_response(ollama_response, 'ollama')
+        return self._gemini_fallback(
+            ollama_result,
+            lambda: self._gemini.generate_sales_copy(
                 product_name=product_name,
                 trend_keywords=trend_keywords,
                 style=style,
                 upcoming_holidays=upcoming_holidays,
                 bestseller_products=bestseller_products,
                 model=model
-            )
-        else:  # ollama
-            response = self._ollama.generate_sales_copy(
-                product_name=product_name,
-                trend_keywords=trend_keywords,
-                style=style,
-                upcoming_holidays=upcoming_holidays,
-                bestseller_products=bestseller_products
-            )
-
-        return self._convert_response(response, provider)
+            ),
+        )
 
     def extract_keywords(self, text: str, provider: str = None,
                         model: str = None, max_keywords: int = 10) -> AIResponse:
@@ -319,14 +344,12 @@ class AIProviderService:
         Returns:
             AIResponse
         """
-        provider = provider or self._default_provider
-
-        if provider == 'gemini':
-            response = self._gemini.extract_keywords(text, max_keywords, model)
-        else:  # ollama
-            response = self._ollama.extract_keywords(text, max_keywords)
-
-        return self._convert_response(response, provider)
+        ollama_response = self._ollama.extract_keywords(text, max_keywords)
+        ollama_result = self._convert_response(ollama_response, 'ollama')
+        return self._gemini_fallback(
+            ollama_result,
+            lambda: self._gemini.extract_keywords(text, max_keywords, model),
+        )
 
     def search_product_insights(self, product_name: str, provider: str = None,
                                 model: str = None, include_competitors: bool = True,
@@ -346,24 +369,22 @@ class AIProviderService:
         Returns:
             AIResponse
         """
-        provider = provider or self._default_provider
-
-        if provider == 'gemini':
-            response = self._gemini.search_product_insights(
+        ollama_response = self._ollama.search_product_insights(
+            product_name=product_name,
+            include_competitors=include_competitors,
+            include_trends=include_trends,
+            web_context=web_context
+        )
+        ollama_result = self._convert_response(ollama_response, 'ollama')
+        return self._gemini_fallback(
+            ollama_result,
+            lambda: self._gemini.search_product_insights(
                 product_name=product_name,
                 include_competitors=include_competitors,
                 include_trends=include_trends,
                 model=model
-            )
-        else:  # ollama
-            response = self._ollama.search_product_insights(
-                product_name=product_name,
-                include_competitors=include_competitors,
-                include_trends=include_trends,
-                web_context=web_context
-            )
-
-        return self._convert_response(response, provider)
+            ),
+        )
 
     def web_search(self, query: str, provider: str = None, model: str = None,
                    num_results: int = 5, search_type: str = "general") -> AIResponse: