User repository on GitHub
Challenge the Jev's intelligence in Rubik Cube puzzles
- Forks
- 0
- Last pushed
- Updated Sep 2026
Repo stats from the GitHub API, cached Oct 2026.Homepage
CommunityJev Demos & Experiments
Rubikjev
A single-page Rubik's cube playground where jev rates your scramble, then solves it one move at a time: each turn is judged from the live cube state, streamed to a 3D cube over SSE, and verified against a reference model before it animates.
- Stars
- 7
- Language
- TypeScript
- Published by
- Community
- Author
- 0xtrou
- Last verified
- Oct 2026
About this project
You scramble like a menace. Jev — the in-house AI solver — judges the chaos, roasts you, and cooks the solve live. An AI-gamified twisty puzzle — scramble it, gamble it, watch Jev cook. 🎰. A one-page, no-scroll Rubik's cube playground with an AI attitude: 🌪️ You scramble — pick a preset (Chill → Spicy → GIGACHAD → 🤯 UNHINGED), mash the manual turn buttons, or unleash a custom number of turns (up to 4900).
Quoted from the project's README. Read the full README.
Works forJev for Classification
Used inJev for Software & DevTools
- Languages
- TypeScript
Highlights
- Each turn is judged from live facelets and progress only — the move history is deliberately withheld from the model.
- The one non-jev mover is the Kociemba two-phase solver, and it only runs when jev itself invokes it; those moves are tagged in the stream.
- Every solve is checked against an independent reference model before anything animates, with an audit tab logging who turned what.
- Raw inference — questions, criteria, probabilities, model ids, latency — stays server-side; the client sees five typed events and a token count.
Watch out
Read this as an entertainment demo, because the README says it is one: fan-made, provided as is, not affiliated with Rubik's Brand or Spin Master, with AI output that is probabilistic and may be inaccurate. The judgements — five meme tiers, a difficulty rating, one roast — are a novelty layer, and the actual cube work is done by the Kociemba solver whenever jev invokes it, so nothing here should be read as a claim about decision quality. The project also makes no accuracy or benchmark claims by design: a compliance document maps each obligation of the engine's master customer agreement to a specific file, including a deliberate no-benchmarks rule. To run it you supply your own engine endpoint and key as server-only secrets. Copy reviewed against the repository README on 2026-10-04, not from running it.
Questions about this entry
Short answers drawn from the linked project or page.
Both, and the stream distinguishes them. The model judges every single move: the 18-move alphabet is declared once, then each turn it sees only the live cube state — facelets and progress — and plays one move, with the move history withheld from it on purpose. The one non-jev mover is the Kociemba two-phase solver, and it only runs when the model invokes it as a tool. Those moves are tagged separately in the stream and in the audit tab, alongside your own manual turns. It is a loop where a decision model chooses when to reach for the superhuman tool.
There are checks at two levels, and you can run them. Every solution is verified against an independent reference model before anything streams, so a wrong move sequence should not reach the 3D cube. Separately, there are harnesses: the check script cross-checks the cube model, facelets and randomized solver output against that reference, a cube-math script runs 200 animation-math trials, and a stress script runs a 500-turn scramble. The project claims every cubie returns home — you can verify that rather than take it on trust.
Server-side only, and the split is deliberate. The engine key and base URL are read from server environment variables, never exposed to the client bundle, and the endpoint is not hardcoded anywhere. Raw inference — the questions, criteria, probabilities, model ids and latency — stays inside the API route and is never logged or streamed. The browser receives exactly five typed events plus a generic token count, and the project states a zero-provider-strings check that is verified by grepping the built bundle.
No, and that is deliberate. The compliance document in the repository maps each obligation of the engine's master customer agreement to the specific file that satisfies it, and that mapping explicitly includes a rule against publishing benchmarks. The README makes no accuracy claim for the judgements either, describing them as probabilistic and for fun, never a fact. If you want to know how good the decision quality is, this project will not tell you — the interesting engineering here is the streaming boundary and the cube verification.
A one-page, no-scroll playground. Pick a scramble preset from Chill through Spicy and GIGACHAD to UNHINGED, mash the manual turn buttons, or unleash up to 4,900 custom turns. Jev then rates the chaos as one of five meme tiers, a one-to-five star difficulty and exactly one roast. Afterwards you bank the XP or gamble it on Jev's Gambit, a double-or-nothing slot at x2, x3, x5, x10 or a bust. XP, ranks from NPC to Sigma Orchestrator, badges and run stats persist in localStorage.
Next.js 16 App Router, React Three Fiber, Tailwind v4, shadcn/ui, zustand for local storage and canvas-confetti, deployed on Vercel. Run pnpm install, put JEV_ENGINE_API_KEY and JEV_ENGINE_BASE_URL in .env.local, then pnpm dev. The cube machinery is adapted from buing/Rubiks-Cube, the most-starred Three.js cube, rebuilt for React Three Fiber, and the fonts are Amatic SC and Nunito.
Five curated event types and nothing else: meta, waypoint, move, done and error, signed engine: "jev-stream/1", plus a generic token count. Questions, criteria, probabilities, model ids and latency stay inside the API route, never logged and never streamed, and a bundle grep checks the client for zero provider strings. The Audit tab separately logs every turn that rotated the cube, tagged as yours, Jev's, or its tool's.
More Jev resources like this
98GitHub stars
Self-hosted Docker service and web UI that packages Laya structured-decision inference behind a compatible System One endpoint.
Jev Demos & Experiments#community#jevbest#catalog-import
Community
1Panel-dev/laya-serverPython
1Panel-dev/laya-serverPython23GitHub stars
Browser drone-fleet lab where TypeSafe Jev steers each drone as fast reflexes and an optional planner offers one-use strategy advice when confidence drops, without pausing control.
Jev Demos & Experiments#community#jevbest#catalog-import
23GitHub stars
Chrome extension that sends each X post to Jev with six typed questions in one call near the viewport, then renders a behavior chip under the post plus a session cost and latency panel.
Jev Demos & Experiments#community#jevbest#catalog-import
Community
oso95/x-scannerTypeScript
oso95/x-scannerTypeScript14GitHub stars
Manifest V3 Chrome extension in TypeScript that sends each LinkedIn post to Jev and collapses bait, self-promo, off-topic, and sponsored posts into a one-line bar with reasons and scores.
Jev Demos & Experiments#community#jevbest#catalog-import
Community
stefw/lkcleanTypeScript
stefw/lkcleanTypeScript7GitHub stars
Demo that turns free-text UI descriptions into real shadcn components using only Jev typed Choice, Noul, and Score answers, with no generated code on the decision path.
Jev Demos & Experiments#community#jevbest#catalog-import
8GitHub stars
Claude-driven browser automation where plain-English chat drives Brave or Chrome, Jev makes on-page decisions, and a guard freezes irreversible clicks.
Jev Demos & Experiments#community#jevbest#catalog-import
Community
ranjan2829/AskJevTypeScript
ranjan2829/AskJevTypeScriptFrom the Jev guides
Original write-ups on what this entry is for.
A game tick is a state plus a legal action list. How Clash Royale, Minecraft, and the earlier Mario, Pokémon, StarCraft, and drone loops ask Jev one step at a time.
Jev Demos & Experiments#gaming#agents#control
Read article
From the Jev community
Posts from builders shipping with Jev right now.