AI Runtime UI
Surfaces for the on-device intelligence layer.
Whispers, intent tiles, confidence indicators, and inline reasoning shells. Designed for local-first inference where latency is sub-100ms and tokens stream into the UI.
State Logic
Reactive surfaces this module exposes to the runtime.
idle | thinking | streaming | done | refused.
Drives indicator ring and color.
Live throughput meter on streaming surfaces.
Local model identifier. Always shown, never hidden.
Visual Variants
Allowed rendering modes. No ad-hoc styles outside this set.
One-line assistance under inputs.
Floating suggestion, dismissible.
Navigation prediction in sidebar.
Expandable chain-of-thought block.
Scoped denial with policy reason.
Motion Tokens
Deterministic timing curves. No custom easing per component.
| token | value | use |
|---|---|---|
| ai.stream | 60ms ease-out per token | Token-by-token reveal. |
| ai.think | 1200ms ease-in-out loop | Pulse while inferring. |
| ai.settle | 240ms cubic-bezier(.2,.8,.2,1) | Final layout on done. |
Isolation Rules
Hard boundaries enforced at the module edge.
Model identity is non-removable from any AI surface.
Confidence below 0.4 must render the indicator in warn tone.
Whispers never block input. Intent tiles never auto-execute.
No AI surface may write to wallet or agent state directly.