Generative UI runtime
Early accessScreens that compose themselves. From your components, inside your rules.
A runtime where app interfaces are assembled by a model per user, per moment, from a closed catalogue of components you define. Every screen passes a schema gate before it reaches a device. Every screen ever served is reproducible from an audit trail. One API call pins any surface back to a hand-authored fallback. Deterministic, PII-free by construction, and near-zero AI cost at steady state.
Compose pipeline
- 08:02Contextrole + workload bucketed
- 08:02Composescreen assembled from catalogue
- 08:02Gateschema + lint passed
- 08:02Serve12ms from fingerprint cache
Same context, same screen. Every time.
FAQ
What is Weft and who is it for?
Weft is a generative server-driven UI runtime for product teams. Instead of hand-building every screen and shipping the same layout to every user, Weft composes each screen at request time from the team's own components and design tokens. The model works inside a closed, typed catalogue, so it cannot invent UI, styling or behaviour. It is built for teams who want adaptive interfaces with production guarantees, not demos.
How does Weft keep AI-composed screens safe?
Four mechanisms. A closed component catalogue with typed props and enumerated design tokens, so invalid or off-brand screens are structurally impossible. A validation gate that checks every composed screen against the schema and semantic rules before it reaches a device. A deterministic fingerprint model, so identical context always produces the identical screen, served from cache. And an append-only audit trail plus a pin API that can roll any surface back to a hand-authored fallback in one call. User data never enters a prompt; values bind after validation.