Apparite (jev2ui)

September 20, 2026 · View on GitHub

All projects · Web apps

Local design-mock lab: TypeSafe Jev chooses information architecture and component anatomy; Gemini writes copy; code assembles A2UI-inspired mocks painted by a DESIGN.md.

At a glanceDetails
SourceSource
TagsOpen source · Free source build · BYOK
Product homepageProject homepage — local Vite app (Apparite); no separate commercial product page.
Pricing and accessApache-2.0 source has no app purchase fee, checked 2026-09-20. Live mocks need a TypeSafe/Jev key and a Gemini key (see .env.example). Provider usage is billed separately.
Jev evidenceInspected src/server/models.ts: @typesafe-ai/sdk TypeSafeClient with default model jev-latest, plus design-mix / IA probe paths that issue typed System One questions. Live mock generation was not run.
DisclosureFree source access does not include TypeSafe or Gemini usage. AI-assisted, independently curated listing; no commercial relationship was declared. Inclusion is not endorsement. Implementation reviewed from public source.
Maintainerdglazkov.
FormatTypeScript Vite + Node server (jev2ui / Apparite 1.0.0).
Platform and availabilitySource build; npm run dev serves the UI (default Vite localhost). Experimental research/demo quality—not a hosted SaaS.
Jev's roleChooses archetypes, blocks, anatomy, and instance controls as calibrated Choice/Score/Noul answers; Gemini supplies prose; application code owns tree assembly, validity, and chrome. Optional gev endpoint speaks the same wire format for experiments.
RequirementsNode.js matching the lockfile; JEV_API_KEY (or upstream-documented TypeSafe key) and Gemini credentials for full mock generation. IA probes can run Jev-only without Gemini.
LicenseApache-2.0.

When to use

Use it to study “Jev decides structure, text model writes words, code assembles UI” with inspectable per-turn probabilities. Prefer production design tools or component libraries when you need shipped product UI—not an experimental mock loop. Do not treat probe timing or sample-sweep writeups as catalog-measured quality scores.

How it works

A prompt drives nested Jev questions that fill a mock tree (archetype → blocks → anatomy → instance controls). Gemini streams copy into slots; DESIGN.md themes the result. Compare mode and IA probes (probe:ia, observe:journeys) exercise decision endpoints without always rendering full mocks.

Get started

git clone https://github.com/dglazkov/jev2ui.git
cd jev2ui
git checkout d55836912ff022db5671e5cf0b75011d35607bb5
npm ci --ignore-scripts
cp .env.example .env   # add keys for live mocks
npm run typecheck
npm run build
npm run dev            # http://localhost:5173 — live keys required for full generation

Examples and demos

  • Upstream README walkthrough and closed-IA / sample-sweep docs under docs/.
  • compare.html dual-run UI for Jobs / Sections / Baseline modes.
  • No npm test script at this revision; typecheck and production build were used as offline checks.

Limits and data handling

Prompts and design state go to TypeSafe Jev and Gemini when live. Experimental UX and catalogs may change. IA sample results in docs are author-reported. Failures and uncertain answers should be read from the decision list—do not assume a pretty mock is correct.

Review and maintenance

Reviewed on 2026-09-20 at commit d558369: 1.0.0, Apache-2.0. AI-assisted source review of models.ts, design-mix paths, README, and LICENSE. On Node.js 22.23.2: npm run typecheck and npm run build passed. No live Jev or Gemini calls.

Related: JevSlop, RefGarden.