問題文
A security policy forbids sending document text to any endpoint outside the database, but the team still needs embeddings for a similarity search over those documents. Which approach fits the policy?
選択肢
- Store the raw document text in a table and let the similarity search read the text columns.
- Call a third party embedding endpoint through a database credential and cache each response in a table.
- Send each document to a hosted embedding service and store the returned vector next to the business columns.
- Import an embedding model in ONNX format and call it from SQL keeping the text inside the database.