Four families of metacognitive models in active development. Each lands here when it ships — with a full model card describing capabilities, intended uses, and known limits.
Different modalities of intelligence, the same metacognitive spine running underneath each one. The capacity to know your own boundaries doesn't just make a model safer — it is what lets the model scale honestly into genuinely new territory.
Large language models that reason about their own uncertainty and the boundaries of their own knowledge. Calibrated confidence on every output; honest abstention when a question falls outside what the model has actually learned.
Speech and audio understanding whose self-awareness opens the door to scale — to new languages, new voices, and new accents. The model recognises unfamiliar signal instead of fabricating around it.
Image and video understanding whose self-awareness opens the door to scale — to new scenes, new objects, and new visual domains. The model recognises the unfamiliar instead of forcing it into a familiar label.
Models for scientific discovery — drug design, materials, biology, climate — whose self-awareness gives researchers calibrated uncertainty on every prediction, and honest abstention past the boundary of what the data supports. The basis for results that hold up at the bench.
Four modalities. One metacognitive spine. Released when ready.