Adaptive interfaces within defined rules.

Weft is a runtime for product teams in which a model arranges each screen from a closed catalogue of the team's own components and design tokens. It is for teams that want interfaces to adapt to context without the model inventing layout, styling or behaviour. It is in early access and not yet generally available.

By design, the model can only choose from a catalogue of components with typed properties. Each composed screen is checked against a schema and semantic rules before it is shown, every screen served is recorded so it can be reproduced, and a surface can be pinned back to a hand-authored fallback. The checks apply to a screen's structure. They do not judge whether it is the best screen for a user, and they do not replace testing the components themselves.