from qdrant_client import QdrantClient from langchain_qdrant import QdrantVectorStore, RetrievalMode from langchain_openai import OpenAIEmbeddings from app.config import settings _client: QdrantClient | None = None _vector_store: QdrantVectorStore | None = None _embeddings: OpenAIEmbeddings | None = None def get_qdrant_client() -> QdrantClient: global _client if _client is None: _client = QdrantClient( url=settings.qdrant_url, api_key=settings.qdrant_api_key, timeout=120, ) return _client def get_embeddings() -> OpenAIEmbeddings: global _embeddings if _embeddings is None: _embeddings = OpenAIEmbeddings( model=settings.embedding_model, api_key=settings.openai_api_key, openai_api_base=settings.openai_api_base, ) return _embeddings def get_vector_store() -> QdrantVectorStore: """Return the singleton QdrantVectorStore connected to the existing collection.""" global _vector_store if _vector_store is None: client = get_qdrant_client() embeddings = get_embeddings() # Use DENSE retrieval (existing collection was indexed with dense vectors) # HYBRID requires a sparse embedder configured separately _vector_store = QdrantVectorStore( client=client, collection_name=settings.collection_name, embedding=embeddings, retrieval_mode=RetrievalMode.DENSE, vector_name=settings.vector_name, ) return _vector_store