AnyDecisionModel gives Swift applications typed answers instead of prose to parse
The decision API exposes choices and probabilities. Its session behavior and calibration requirements still need explicit handling.
AnyDecisionModel returns typed decisions: a Boolean answer, a selection, or a rating. Application code can consume those values without first extracting a choice from generated prose. Mattt introduced the library alongside AnySpeechRecognizer in an article published October 29, 2026.
The session contract deserves attention. A DecisionSession accepts text or JSON state, but answers do not automatically become context for subsequent questions. Each question is independent. For a workflow where an earlier choice changes the next decision, I would put that dependency into the supplied state explicitly.
Probabilities introduce a different concern. The article describes overconfidence in raw outputs and provides calibration, which adjusts probabilities using labeled examples. It calls for separate data when fitting and evaluating that adjustment. Entropy describes how concentrated the distribution is; a concentrated distribution can still favor the wrong answer.
I would use the typed interface to simplify the boundary between the model and application code, while evaluating decision errors separately. The remaining engineering limit is semantic: a valid value can still represent an incorrect choice.
Source: Mattt, the library announcement.