reflexui
TypeScript framework for adaptive screens where Jev ranks, gates and picks existing UI items and code enforces thresholds and stability.
Overview
reflexui is a TypeScript monorepo for adaptive UI surfaces such as dashboards, alerts and suggestions. Every card, alert and suggestion already exists in your code with computed values; a System One model like Jev only ranks, gates and picks among them, returning typed answers with confidence in roughly 70-500 ms. A policy layer in code applies thresholds, hysteresis, pins, a cap on swaps per visit and fallbacks, so screens adapt to the moment, never show invented content, and never block rendering on the model.
Packages include the core (compose, decide, buildRequests), a TypeSafe/Jev decider and a deterministic mock decider, a React hook with a debug panel, Redis stores for a shared cache, and offline replay for evaluation. Typical calls are a score per card for ranking and a Noul per alert for gating. The README compares it with json-render and jev-ui and says its focus is the policy and stability around the model call. Status is pre-alpha: core paths are covered by CI tests, but the live Jev API has only been exercised by hand.
What the decision model decides
| Type | Question | Answers |
|---|---|---|
| Score | How relevant is this card to the user right now?(inferred) | — |
| Yes / no | Should this alert be shown? | — |
| Choice | Which suggestion should be featured?(inferred) | — |
How it works
- Define a surface with existing items and computed values
- Build typed requests (score per card, noul per alert)
- Jev or the mock decider answers with confidence
- Compose applies thresholds, hysteresis, max swaps and fallback before rendering
Strengths
- Never shows model-invented content
- Stability rules prevent layout thrash
- Offline replay for evaluation
- Rendering never waits on the model
Caveats
- Pre-alpha; API may change
- No automated tests against the live Jev API
Getting started
pnpm add @reflexui-jev/core @reflexui-jev/providers, then use typesafeDecider() with TYPESAFE_API_KEY (or mockDecider() offline).At a glance
- Type
- Project
- Status
- prototype
- Language
- TypeScript
- Runtime
- Node 20+, React
- Has code
- yes
- Has tests
- yes
- License
- MIT
- Stars
- 0
Use cases and integrations
How curated lists describe it
Adaptive, explainable screens driven by fast typed-judgment models. A System One model (Jev) decides what's relevant; your code decides what's allowed.
wh000wh000/awesome-jev-live
Frequently asked questions
What does Jev decide in reflexui?
Which existing items to rank higher, which alerts to show, and which suggestion to pick.
Can the model add new UI content?
No; it only ranks, gates and picks among items your code already provides.
Can I try it without a key?
Yes, the mockDecider returns deterministic hash-based answers.
Related projects
Interactive page where HR tool logos rise out of a physics pile as Jev judges which fit a typed HR problem.
DemoprototypeChrome extension that badges X and LinkedIn posts READ, MAYBE or SKIP against your plain-English interest profile as you scroll.
Browser extensionbetaJavaScriptReact components that use Jev judgments to choose which component renders, how lists are ordered and whether affordances show.
ProjectbetaTypeScriptPrompt-first music studio where Jev makes every compositional choice, from plan to each bar pattern, rendered as editable MIDI.
ProjectbetaTypeScriptVercel Labs' generative UI framework with an experimental compose path where Jev picks which catalog components and actions to render.
ServicebetaTypeScriptCommunity shadcn-style React components and interactive lab for building TypeSafe AI interfaces.
ProjectbetaTypeScript