User repository on GitHub
Play Chinese Chess (Xiangqi) against Jev - TypeSafe System One decision model as the AI. Score fan-out over legal moves.
- Forks
- 0
- Last pushed
- Updated Sep 2026
Repo stats from the GitHub API, cached Oct 2026.
Jev directory
CommunityJev Demos & Experiments
Jev Xiangqi
Play Chinese Chess against Jev: every AI move fans out a Score 0-4 question per legal move in a single system_one() call, then plays the highest-scoring move. No minimax and no eval tables, just Jev's read of the position, with the top 10 shown as score bars.
- Stars
- 1
- Language
- HTML
- License
- BSD-2-CLAUSE
- Reading time
- 3 min
- Published by
- Community
- Author
- zafer-liu
- Last verified
- Oct 2026
Used inJev for Gaming
- Languages
- JavaScript
- Platforms
- Browser
- Capabilities
- Open sourceRequires API key
Highlights
- On each AI turn the frontend enumerates legal moves and the backend builds one Jev call with a Score 0-4 question per legal move — typically 30-45 in mid-game.
- State includes an ASCII board, FEN, side to move, a check flag, the last 12 moves and a numbered legal-move list such as [m7] h2e2.
- The v2 prompt injects a rules primer — piece movement, piece values, motifs like 马后炮, a 3-step simulation instruction — once per call, not per question.
- Measured on jev-latest on 2026-09-21: opening 44 moves cost $0.000322 and 6.6 s; mid-game 36 moves cost $0.000278. Input is $0.042/MTok and output is free.
- The README stresses Jev is not a chess engine: it can miss a 3-move tactic and top-2 margins often sit under 0.1 points.
Watch out
Grounded in the truncated README and GitHub metadata reviewed on 2026-10-15; the v1-vs-v2 table is cut off mid-row, the latency and cost figures are the author's own measurements from 2026-09-21 on one prompt version, and no install or key-setup steps were captured. BSD-2-Clause per the metadata.
Questions about this entry
Short answers drawn from the linked project or page.
It scores every legal move at once: the frontend enumerates the moves, the backend asks one Score 0-4 question per move inside a single system_one() call, and the highest-scoring move is played. No minimax, no eval tables — just Jev's read of the position.
An ASCII board, the FEN, the side to move, a check flag, the last 12 moves and a numbered list of legal moves such as [m7] h2e2 Red Cannon from h2 moves to e2. Each question shares the same five anchors: Blunder, Weak, Neutral, Good, Excellent.
A shared rules primer is injected into the state — piece movement, piece values from Rook 9 down to Pawn 1, tactical motifs and positional principles with an explicit 3-step simulation instruction — so Jev can evaluate trades numerically. The README reports roughly zero net token change versus v1.
The README's own measurements on jev-latest from 2026-09-21: an opening position with 44 legal moves cost $0.000322 and 6.6 s wall latency; a mid-game position with 36 moves cost $0.000278. Input is priced at $0.042 per million tokens and output is free.
Not engine-strength: the README states Jev is not a chess engine, will miss a 3-move tactic, and often has a top-2 margin under 0.1 points, making the pick near-random among the top few. The interesting property is the visible distribution, not playing strength.
The UI renders the top 10 moves with score bars so you can see the shape of Jev's evaluation rather than just the pick. The README notes you can gate on the margin or log the entropy of the distribution without extra API calls.
The captured README does not include install or run steps; it documents the xiangqi.js frontend, the POST /api/move endpoint and the backend Jev call. Check the repository for setup, and note that live play needs a TypeSafe key to call Jev.
More Jev resources like this
A 25-second IT-buzzword typing game built to feel how fast and cheap Jev is: each word gets five typed judgments in one request, and the screen shows the running cost next to the real HTTP request and response.
Jev Demos & Experiments#community#catalog-import#logicrw
Community
pochang6/jev-buzzword-rushJavaScript
pochang6/jev-buzzword-rushJavaScriptJev and GPT-6 Luna race through the same Flappy Bird course, with every flap a live model decision. Both birds request flap or wait every 50 ms, and the browser shows each answer applied, superseded or skipped.
Jev Demos & Experiments#community#catalog-import#logicrw
Community
jaibhasin/jev-flappy-birdJavaScript
jaibhasin/jev-flappy-birdJavaScriptA 12,000-neuron fly-brain subgraph of the maleCNS connectome fights Jev in mk.js: the fly simulates spikes in the browser at 10 decisions per second while Jev reads the fight as JSON and answers one typed choice among seven moves.
Jev Demos & Experiments#community#catalog-import#logicrw
Community
lavallee/mk-jev-fly-brainJavaScript
lavallee/mk-jev-fly-brainJavaScriptFighters in this arena are paragraphs of plain English, and Jev pilots them in real time — typed move, action and reflex decisions with live probabilities. Join the ladder with a pull request and swap the brain: Jev, Laya, kev or any LLM.
Jev Demos & Experiments#community#catalog-import#logicrw
Community
eliot5566/jev-arenaJavaScript
eliot5566/jev-arenaJavaScriptA table-tennis game where every paddle movement on the right is a live Jev decision: the relay renders the state as prose, asks one multiple-choice question, and the option label IS the pixel value. No local prediction, no fallback.
Jev Demos & Experiments#community#catalog-import#logicrw
Community
liuhao-1443/jev-table-tennisPython
liuhao-1443/jev-table-tennisPythonA small web app where TypeSafe's Jev plays Tetris by itself — the only control is Reset. Code enumerates every reachable placement and its board facts, and one typed Choice call picks the move with probabilities, latency and cost.
Jev Demos & Experiments#community#catalog-import#logicrw
Community
pinebit/jev-tetrisJavaScript
pinebit/jev-tetrisJavaScriptFrom the Jev guides
Original write-ups on what this entry is for.
From the Jev AI model community
Posts from builders shipping with Jev right now.