MotherDuck ships prompt_jev(), a 50x-faster SQL classifier function

TypeSafe AI makes Jev, a model built to answer a specific yes/no or multiple-choice question about a piece of text and return a typed decision, not open-ended prose. MotherDuck, the DuckDB-based cloud data warehouse, added a SQL function, prompt_jev(), that calls Jev directly from a database query: a user passes a text column and a question, and gets back a probability, a chosen category, or a position on a scale for every row, without leaving SQL or running a separate classification pipeline.

MotherDuck reports the function classifying 100,000 rows in about 40 seconds for roughly $0.50, which it frames as around 50 times faster and about 1% of the cost of doing the same job by calling a general-purpose LLM row by row. TypeSafe AI's tweet promotes the integration but does not itself state those MotherDuck-specific figures; they come from MotherDuck's own published account of its measured conditions, not from independent testing, so they describe that integration rather than a universal speed claim for other databases or workloads.