Agent tooling
MLX Direct Decider reads existing MLX models' output scores to choose among supplied answers, including no-match and abstain outcomes, without additional training.
Only you can see your notes.
Screenshot unavailable. Open the experiment ↗
The MIT project exposes Python, command-line and web-service interfaces, plus a game example. A decision can return no match or abstain rather than selecting an ill-fitting answer. This early toolkit is distinct from AnyJev and JevMLX; it reads a base model's scores rather than becoming TypeSafe's Jev.
On supported Apple hardware, select a documented MLX model and define the candidate answers through the Python or command-line interface. Use the no-match or abstain result when the available choices do not fit, and assess the base model on your task.