clef-omni is a multimodal decision model built on a 30b-parameter mixture-of-experts (3b active) backbone. it turns a state and a schema of typed questions into decisions, reads the state as text, json, images, audio, or video, and returns a probability for every allowed option of every question.