core/embeddings.py

# embeddings.py
# Environment-aware embeddings:
#   HF Spaces → HuggingFaceEmbeddings (CPU)
#   Local      → OllamaEmbeddings

import os
from dotenv import load_dotenv

load_dotenv()

IS_HF_SPACES: bool = bool(os.getenv("SPACE_ID", ""))


class EmbeddingStore:
    def __init__(self, model_name=None):
        if IS_HF_SPACES:
            from langchain_huggingface import HuggingFaceEmbeddings

            hf_model = model_name or os.getenv(
                "EMBED_MODEL", "sentence-transformers/all-MiniLM-L6-v2"
            )

            if "/" not in hf_model:
                hf_model = f"sentence-transformers/{hf_model}"

            self.embeddings = HuggingFaceEmbeddings(
                model_name=hf_model,
                model_kwargs={"device": "cpu"},
                encode_kwargs={"normalize_embeddings": True},
            )
        else:
            from langchain_ollama import OllamaEmbeddings

            ollama_model = model_name or os.getenv("EMBED_MODEL", "nomic-embed-text")
            self.embeddings = OllamaEmbeddings(model=ollama_model)

    # ✅ REQUIRED by Chroma / LangChain
    def embed_documents(self, texts: list) -> list:
        return self.embeddings.embed_documents(texts)

    def embed_query(self, text: str) -> list:
        return self.embeddings.embed_query(text)

    # Optional helper
    def embed_chunks(self, texts: list) -> list:
        return self.embed_documents(texts)