3.0
SLMs
Specialized small models built on domain-specific data from our collection work. Designed to run on constrained hardware, close to the data source, with no cloud dependency.
Shelf-1
A retail vision model family (a detector, a matcher and a decision model) trained from routine shelf photos taken by store staff, checked by a verifier and people, and recorded as ShelfTrace. In development.
Commerce-1
A planned small model, to be trained on Machine Commerce behavioral traces, to predict and classify agent decision quality in e-commerce flows. Designed to run as a lightweight evaluator inside agent pipelines, flagging likely failures before they surface in production. In design.