feat(D1): models.json 集中化 — ADR-067 五大 Ollama 應用 hardcode 消除

- models.json v1.3.0: providers.ollama.models 新增 9 個 purpose keys
  (drift_summary/drift_intent/log_anomaly/nemoclaw/playbook_draft/
   code_review/embedding/rag_generate/image_analysis)
- drift_narrator_service: NARRATOR_MODEL → get_model("ollama","drift_summary")
- drift_interpreter: MODEL → get_model("ollama","drift_intent")
- log_summary_service: SUMMARY_MODEL → get_model("ollama","log_anomaly")
- local_code_review_service: _MODEL_OLLAMA → get_model("ollama","code_review")
- image_analysis_service: _MODEL → get_model("ollama","image_analysis")
- decision_manager: nemoclaw + playbook_draft 兩處 → get_model()
- embedding_service: get_embedding_service() factory → get_model("ollama","embedding")
- knowledge_service: OllamaEmbeddingService(model=...) → get_model()

所有模型名稱現在統一由 models.json 管理，修改模型只需改一個檔案。
LOGBOOK 更新：D1 完成 + B2 已完成確認

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

apps/api/src/services/decision_manager.py

@@ -332,8 +332,10 @@ async def _nemoclaw_second_opinion(incident: "Incident", primary_result: dict) -
         from src.core.config import settings
         import httpx as _httpx
 
+        from src.services.model_registry import get_model as _get_model
         ollama_url = getattr(settings, "OLLAMA_URL", "http://192.168.0.188:11434")
-        model = "deepseek-r1:14b"
+        # D1 集中化 2026-04-11: 從 models.json providers.ollama.models.nemoclaw 讀取
+        model = _get_model("ollama", "nemoclaw")
 
         signals_summary = ""
         if incident.signals:
@@ -423,11 +425,14 @@ async def _generate_playbook_draft_if_new(incident: "Incident") -> None:
             f"## 驗收條件\n（如何確認修復成功）"
         )
 
+        from src.services.model_registry import get_model as _get_model
         ollama_url = getattr(settings, "OLLAMA_URL", "http://192.168.0.188:11434")
+        # D1 集中化 2026-04-11: 從 models.json providers.ollama.models.playbook_draft 讀取
+        _pb_model = _get_model("ollama", "playbook_draft")
         async with _httpx.AsyncClient(timeout=45.0) as client:
             resp = await client.post(
                 f"{ollama_url}/api/generate",
-                json={"model": "qwen2.5:7b-instruct", "prompt": prompt, "stream": False},
+                json={"model": _pb_model, "prompt": prompt, "stream": False},
             )
             resp.raise_for_status()
             content = resp.json().get("response", "").strip()

apps/api/src/services/drift_interpreter.py

@@ -107,12 +107,14 @@ class NemotronDriftInterpreter:
         """
         import httpx
         from src.core.config import get_settings
+        from src.services.model_registry import get_model as _get_model
 
         # C1 修復 2026-04-11: 禁止寫死內網 IP（feedback_frontend_internal_ip_ban 鐵律）
         # 改從 settings.OLLAMA_URL 讀取（已有此設定，default=http://192.168.0.111:11434）
         _settings = get_settings()
         OLLAMA_URL = getattr(_settings, "OLLAMA_URL", "http://192.168.0.111:11434")
-        MODEL = "qwen2.5:7b-instruct"
+        # D1 集中化 2026-04-11: 從 models.json providers.ollama.models.drift_intent 讀取
+        MODEL = _get_model("ollama", "drift_intent")
         TIMEOUT = 45.0
 
         try:

apps/api/src/services/drift_narrator_service.py

@@ -24,6 +24,7 @@ import httpx
 import structlog
 
 from src.core.redis_client import get_redis
+from src.services.model_registry import get_model
 
 if TYPE_CHECKING:
     from src.models.drift import DriftInterpretation, DriftReport
@@ -34,7 +35,8 @@ logger = structlog.get_logger(__name__)
 # 設定
 # ============================================================
 OLLAMA_URL = "http://192.168.0.111:11434"
-NARRATOR_MODEL = "qwen2.5:7b-instruct"
+# D1 集中化 2026-04-11: 從 models.json providers.ollama.models.drift_summary 讀取
+NARRATOR_MODEL = get_model("ollama", "drift_summary")
 NARRATOR_TIMEOUT = 90.0  # seconds
 CACHE_TTL = 3600  # 1 小時
 CACHE_PREFIX = "drift_narrative:"

apps/api/src/services/embedding_service.py

@@ -21,6 +21,7 @@ import httpx
 import structlog
 
 from src.core.config import settings
+from src.services.model_registry import get_model as _get_model
 
 logger = structlog.get_logger(__name__)
 
@@ -239,10 +240,11 @@ def get_embedding_service() -> OllamaEmbeddingService:
     """
     取得 Embedding Service 單例
 
+    D1 集中化 2026-04-11: 預設模型從 models.json providers.ollama.models.embedding 讀取
     Returns:
         OllamaEmbeddingService: 共用實例
     """
     global _embedding_service
     if _embedding_service is None:
-        _embedding_service = OllamaEmbeddingService()
+        _embedding_service = OllamaEmbeddingService(model=_get_model("ollama", "embedding"))
     return _embedding_service

apps/api/src/services/image_analysis_service.py

@@ -30,6 +30,7 @@ import httpx
 import structlog
 
 from src.core.config import get_settings
+from src.services.model_registry import get_model
 
 if TYPE_CHECKING:
     pass
@@ -37,7 +38,8 @@ if TYPE_CHECKING:
 logger = structlog.get_logger(__name__)
 settings = get_settings()
 
-_MODEL = "llava:latest"
+# D1 集中化 2026-04-11: 從 models.json providers.ollama.models.image_analysis 讀取
+_MODEL = get_model("ollama", "image_analysis")
 _TIMEOUT_S = 120.0
 _MAX_SIZE_BYTES = 5 * 1024 * 1024  # 5MB
 _ALLOWED_MIME = {"image/jpeg", "image/png", "image/webp", "image/gif"}

apps/api/src/services/knowledge_service.py

@@ -29,6 +29,7 @@ from src.models.knowledge import (
 from src.repositories.interfaces import IKnowledgeRepository
 from src.repositories.knowledge_repository import KnowledgeDBRepository
 from src.services.embedding_service import OllamaEmbeddingService
+from src.services.model_registry import get_model as _get_model
 
 logger = structlog.get_logger(__name__)
 
@@ -52,7 +53,8 @@ class KnowledgeService:
 
     def __init__(self) -> None:
         # I2: 注入 embedding service，避免每次呼叫 new 實例
-        self._embed_svc = OllamaEmbeddingService(model="nomic-embed-text", timeout=15.0)
+        # D1 集中化 2026-04-11: 從 models.json providers.ollama.models.embedding 讀取
+        self._embed_svc = OllamaEmbeddingService(model=_get_model("ollama", "embedding"), timeout=15.0)
         # I1: 持有背景 Task 引用，防止 GC 提前回收
         self._pending_tasks: set[asyncio.Task] = set()  # type: ignore[type-arg]
 

apps/api/src/services/local_code_review_service.py

@@ -19,11 +19,13 @@ import httpx
 import structlog
 
 from src.core.config import get_settings
+from src.services.model_registry import get_model
 
 logger = structlog.get_logger(__name__)
 settings = get_settings()
 
-_MODEL_OLLAMA = "qwen2.5-coder:7b"
+# D1 集中化 2026-04-11: 從 models.json providers.ollama.models.code_review 讀取
+_MODEL_OLLAMA = get_model("ollama", "code_review")
 _TIMEOUT_OLLAMA = 120.0
 _MAX_DIFF_BYTES = 50 * 1024  # 50KB → fallback to Gemini
 _SEMAPHORE = asyncio.Semaphore(2)  # 最多 2 個同時審查

apps/api/src/services/log_summary_service.py

@@ -26,6 +26,7 @@ import httpx
 import structlog
 
 from src.core.redis_client import get_redis
+from src.services.model_registry import get_model
 
 logger = structlog.get_logger(__name__)
 
@@ -33,7 +34,8 @@ logger = structlog.get_logger(__name__)
 # 設定
 # ============================================================
 OLLAMA_URL = "http://192.168.0.111:11434"
-SUMMARY_MODEL = "deepseek-r1:14b"
+# D1 集中化 2026-04-11: 從 models.json providers.ollama.models.log_anomaly 讀取
+SUMMARY_MODEL = get_model("ollama", "log_anomaly")
 LLM_TIMEOUT = 180.0       # deepseek-r1 硬超時
 SOFT_TIMEOUT = 5.0         # 主流程軟超時（超過回 None）
 LOG_TAIL_LINES = 100