system-one-chess
Browser chess lab where Jev, LLMs and people play, each Jev turn being one Choice over the legal moves, with Stockfish analysis.
Overview
ai-chess-lab lets LLMs, System One models and people play chess in the browser, in single games, Swiss tournaments or full leagues, with every move, token, second and dollar recorded. For a System One model like Jev (via Vercel AI Gateway or OpenRouter), every turn is one Choice question whose options are the legal moves; a position has at most 218 legal moves and a Choice accepts up to 255, so one request always fits and an illegal move is impossible. Chat LLMs get the same position and move list through the chat API and must name one.
It uses the lichess chessground board, and Stockfish 19 in WebAssembly analyses games locally: evaluation charts, average centipawn loss, blunders, and an analysis of whether Jev beats chance by ranking its moves among all legal moves. Jev playing both sides costs about $0.00005 per move at 300-400 ms, about $0.0009 for a 20-move game. Local Laya and Kev can also play (Laya receives shorter move labels due to its token budget). The README is clear that Jev is a fast classifier, not a chess engine: expect plausible, not strong, moves. A first battle of 16 players and 72 games is included. Built with the pico framework.
What the decision model decides
| Type | Question | Answers |
|---|---|---|
| Choice | Which legal move should be played in this position? | — |
How it works
- Compute legal moves for the position
- Ask Jev one Choice over them (or ask an LLM via chat)
- Play the move and record tokens, latency and cost
- Analyse games with in-browser Stockfish
Strengths
- Illegal moves impossible for Jev
- Very cheap per move
- Rigorous comparison against random and LLM players
Caveats
- Jev is not a strong chess player
- Laya gets less descriptive move labels
At a glance
- Type
- Project
- Status
- beta
- Language
- Python
- Runtime
- Web app (server + browser)
- Has code
- yes
Use cases and integrations
How curated lists describe it
Chess against Jev, TypeSafe AI's System One model, through OpenRouter. Built with the pico framework.
hellogumbo/awesome-jev
Frequently asked questions
What does Jev decide?
Each move, as one Choice over all legal moves.
How strong is it?
The README says expect plausible moves, not strong ones; it is not a chess engine.
What does a game cost?
About $0.0009 for a 20-move game with Jev on both sides.
Related projects
Fifteen live arcade games where Jev and LLMs play each other, led by WikiRace in which Jev picks only real links.
ProjectbetaTypeScriptOne-page web app where Jev plays chess against OpenRouter LLMs, Stockfish or you, showing the moves it weighed and their probabilities.
AppbetaJavaScriptPac-Man where Jev or other System One models choose each move at junctions, with live odds, fallbacks and a leaderboard.
ProjectbetaTypeScriptNiceGUI desktop chess app for evaluating Jev's move choices against Stockfish, random players or itself, with PGN export.
ProjectprototypePythonBrowser 2D driving sim where Jev answers lane, speed, hazard and pedestrian-yield questions every 200 ms, with an LLM comparison mode.
ProjectprototypeJavaScriptReal-time versus Tetris where Jev and an LLM or open Laya model pick placements on the same seeded pieces and clock.
ProjectprototypeTypeScript