← All gamesWatch Jev decide

Jev Chess — watch Jev play chess, one decision at a time

Jev, TypeSafe AI's decision model, plays White against a baseline bot — and every single move is a real, typed API call over the legal moves, with the latency, cost and confidence live above the board. Not a chess engine; a live look at Jev deciding. No key, just press start.

Press start — watch Jev play White0 plies· —ms· $0.00000· conf —
Jev plays White — one API call picks each move (SAN)
—
waiting for Jev’s first move
Move list
press start — watch Jev decide White’s moves live
Build this with Jev

The same demo is one Jev call

  1. Read the board (ASCII) + whose turn
  2. List the legal moves in SAN — captures, checks, quiets
  3. Send them to Jev as one choice question
  4. Jev returns the move it judges best
  5. Play it, the opponent replies, repeat
one decision per move · ~150–300ms · a fraction of a ¢
const res = await fetch("https://jevtypesafeai.com/api/v1/decide", {
  method: "POST",
  headers: { Authorization: `Bearer ${process.env.JEV_API_KEY}` },
  body: JSON.stringify({
    state: game.ascii(),  // the board, as text
    questions: {
      move: { type: "choice", instructions: "Best move for White here?",
             criteria: { "Nf3": "develop knight", "e4": "push center pawn", "Bb5+": "check" } },
    },
  }),
});
const { answers } = await res.json();
game.move(answers.move.choice);   // → "e4" (0.64)
See a real decision E4 · 232ms
State
White to move (opening). Legal: e4, d4, Nf3, c4, g3, …
Question choice · move
e464%
d421%
Nf310%
latency 232mscost $0.000014

How Jev plays chess

Jev plays White; a tiny baseline bot plays Black. Every White move is a single request to the Jev API. We hand Jev the position — the board as ASCII text, whose turn it is, and a quick material tally — together with one typed choice question whose options are the currently-legal moves in Standard Algebraic Notation (SAN). Each option carries a short human label: "Nf3 — develop knight", "exd5 — pawn takes pawn", "Qh5+ — queen gives check". Jev answers with a SAN choice and a calibrated confidence, and we play it on the board.

When the position has more than a couple of dozen legal moves, we pre-filter the choice list down to roughly two dozen — every capture and every check, plus a few quiet developing moves — so the decision stays crisp and cheap. If Jev ever returns a move that isn't in the legal set, we fall back to the first legal move so the game never stalls. Because each move is an ordinary API request, you watch exactly what a developer wiring Jev into a game or an agent would see: the latency, the running cost — a fraction of a cent — and the confidence behind every decision.

Jev vs an LLM at chess

This is the honest part: Jev is not a chess engine, and it is not a large language model narrating its way through a game. It is a System One decision model — it reads a bounded situation and returns a typed, calibrated pick in tens to hundreds of milliseconds. Give it the legal moves and it chooses one, fast, with no prose and no hallucinated illegal moves, because the answer is constrained to the options you defined.

An LLM asked to "play chess" will happily invent illegal moves, drift off into commentary, and cost far more per turn. Jev vs an LLM at chess is really a contrast in shape: Jev returns a guaranteed-valid choice from your candidate set at a fraction of the tokens, which is exactly why it fits real-time game AI, agent tool-call routing, moderation and scoring in production — the same decisions Pac-Man ghosts, a trading bot and a Gomoku opponent make elsewhere on this site.

What to watch for

  • Confidence bars: when Jev's pick is high-confidence it saw a clear best move; a low-confidence pick means the position was murky.
  • Captures and checks: the choice list always includes every capture and check, so watch whether Jev grabs material or presses the attack.
  • Latency and cost: every move is one live call, typically 100–300ms and a fraction of a cent — the cost meter shows the whole game's spend.
  • Fallbacks: a rare "(fallback)" tag means Jev's answer wasn't a legal move that turn, so the first legal move was played instead.

FAQ

Is Jev Chess free to watch?

Yes. It runs in your browser with no API key and no signup. Jev plays White against a baseline bot, and each White move is a free, rate-limited call to the live Jev decision API.

Does Jev really decide each move, or is it scripted?

Jev really decides. On each White move the position and the list of legal moves (in SAN) are sent to the Jev API as one typed choice question, and Jev returns which move to play with a calibrated confidence. Black is a simple built-in bot that grabs the best capture or plays a random legal move.

Is Jev a chess engine?

No. Jev is a general decision model, not a chess engine — it won't outplay Stockfish. The point is to watch a typed, calibrated decision every move: guaranteed-legal picks chosen from the candidate set, fast and cheap, the same way it routes tickets or guards agent tool calls in real products.

How fast and how expensive is each move?

Each move is a single API call that typically returns in about 100–300ms and costs a fraction of a cent in input tokens. The page shows the live latency, the running cost across the whole game, and the confidence behind each chosen move.

More: Jev Gomoku — play vs the AI · Jev Pac-Man — outrun the AI ghosts · Jev Trading Bot — watch it trade · All decision games · Jev API docs

Jev Chess — watch Jev play chess, one API decision per move · Jev by TypeSafe AI