jev-xiangqi
Play Chinese Chess against Jev, which scores every legal move 0-4 in one parallel call and plays the highest, with no search.
Overview
Jev Xiangqi lets you play Chinese Chess against Jev. On the AI's turn the frontend enumerates every legal move and sends the position to a backend, which builds one Jev call whose state includes an ASCII board, FEN, side to move, check flag, last 12 moves and a numbered list of legal moves described in words, plus a shared rules primer. There is one Score question per legal move (typically 30 to 45 mid-game) with five anchors from Blunder to Excellent. Jev answers all in parallel, the highest score is played, and the UI shows the top ten with score bars.
Measured on jev-latest, a call cost about $0.0003 with roughly 6.6 seconds of wall latency. The author is clear that Jev is not a chess engine: it misses short tactics and its top-two margin is often under 0.1, making the pick near-random among the best few. The interesting property is the distribution: you can see the top moves, gate on margin or log entropy without extra calls. A naive prompt played weakly; the second version adds a rules primer for domain grounding. Bilingual documentation; BSD-2-Clause licence.
What the decision model decides
| Type | Question | Answers |
|---|---|---|
| Score | How good is this legal move? | Blunder · Weak · Neutral · Good · Excellent |
How it works
- Enumerate legal moves in the frontend
- One Jev call with a Score per move and a shared rules primer
- Play the argmax; show top ten scores
- Repeat each AI turn
Strengths
- Full distribution over moves from one call
- About $0.0003 per move per the README
- Simple architecture with no search
Caveats
- Weak play: misses short tactics
- About 6.6 s per move
- Top choices often nearly tied
At a glance
- Type
- Project
- Status
- prototype
- Language
- JavaScript
- Runtime
- Web app + backend
- Has code
- yes
- License
- BSD-2-Clause
Use cases and integrations
How curated lists describe it
Play Chinese Chess (Xiangqi) against Jev - TypeSafe System One decision model as the AI. Score fan-out over legal moves.
logicrw/awesome-jev-projects
Frequently asked questions
Is Jev a strong Xiangqi player?
No. The author says it misses short tactics and top moves are often nearly tied.
How much does a move cost?
About $0.0003 with roughly 6.6 seconds latency, per the README.
What improved its play?
Adding a shared rules primer to the state in the v2 prompt.
Related projects
Source for browser chess and Chess960 where code computes each legal move's facts and Jev picks one per turn as a typed Choice.
ProjectproductionTypeScriptBrowser 1v1 sniper arena where Jev reads match state about 9 times a second and chooses movement, aim, ADS, firing and jumping.
ServicebetaTypeScriptInstrumented 2048 web lab where every move is a Jev Choice over four directions, with probabilities, latency and cost shown live.
ProjectbetaTypeScriptFifteen live arcade games where Jev and LLMs play each other, led by WikiRace in which Jev picks only real links.
ProjectbetaTypeScriptReal-time bot battles where you write a fighter's strategy in plain English and a System One model pilots it several times a second.
ProjectbetaTypeScriptBrowser extension that lets Jev, local Laya or an OpenAI-compatible model take over a running ra2web (WannaFire) RTS match.
ProjectbetaTypeScript