diff --git a/apps/api/src/services/knowledge_rag_service.py b/apps/api/src/services/knowledge_rag_service.py index ecc0dc5b..ddda5380 100644 --- a/apps/api/src/services/knowledge_rag_service.py +++ b/apps/api/src/services/knowledge_rag_service.py @@ -29,11 +29,12 @@ from src.core.config import get_settings logger = structlog.get_logger(__name__) settings = get_settings() -# nomic-embed-text 在 188 (記憶體充足) -_EMBED_URL = "http://192.168.0.188:11434" +# nomic-embed-text — 使用 settings.OLLAMA_URL (111, K3s NetworkPolicy 可達) +# 188 雖然記憶體更多但 K3s NetworkPolicy 擋住 11434,改用 111 +# 2026-04-10 Claude Sonnet 4.6 _EMBED_MODEL = "nomic-embed-text" -# 生成回答用 111 (快) +# 生成回答用同一個 OLLAMA_URL (111) _GEN_MODEL = "qwen2.5:7b-instruct" _TOP_K = 5 # 取最相似 5 筆 @@ -105,7 +106,7 @@ class KnowledgeRAGService: try: http = await self._get_http() resp = await http.post( - f"{_EMBED_URL}/api/embeddings", + f"{settings.OLLAMA_URL}/api/embeddings", json={"model": _EMBED_MODEL, "prompt": text}, ) if resp.status_code == 200: