Jev Dino — the AI plays the Chrome dino game
Watch Jev, TypeSafe AI's decision model, run the Chrome dinosaur game live. For every cactus and bird it fires one real API call — jump, duck, or run — as the runner speeds up. No key, just press play and watch the AI dodge.
The same demo is one Jev call
- Read the runner — next obstacle type + distance, speed
- The choices are jump / duck / run
- Send it to Jev as one
choicequestion - Jev returns the action (with confidence)
- Act, then decide again at the next obstacle
const res = await fetch("https://jevtypesafeai.com/api/v1/decide", {
method: "POST",
headers: { Authorization: `Bearer ${process.env.JEV_API_KEY}` },
body: JSON.stringify({
state: `Cactus ahead in 3 tiles, speed fast. Dino on the ground.`, // the runner state, as text
questions: {
action: { type: "choice", instructions: "What should the dino do to clear the next obstacle?",
criteria: { jump: "clear a low cactus", duck: "dodge a high bird", run: "nothing in range" } },
},
}),
});
const { answers } = await res.json();
act(answers.action.choice); // → "jump" (0.88)See a real decision JUMP · 205ms
Cactus (low) ahead in 3 tiles, speed fast. Dino on the ground.
How Jev plays the dino game
The dinosaur runs on its own while obstacles — ground cacti, head-height birds, and high-flying birds — scroll in from the right, faster and faster. Each time an obstacle enters the decision window, the game sends the Jev API a single typed request: one choice question over three options — jump, duck, or run — with the state describing the incoming hazard (its type, its height, and how many units and ticks away it is). Jev answers in roughly 100–300ms with a calibrated probability, and the dino jumps the cactus, ducks the low bird, or runs level under the high one.
There is exactly one Jev call per obstacle, and calls are throttled so two never fire closer than about 600ms apart — enough to keep a shared free key comfortably under the rate limit while the runner accelerates. Because every dodge is an ordinary API request, you see precisely what a developer wiring an LLM into real game logic would see: the live latency, the running cost — a fraction of a cent — and the confidence behind each jump, duck, or run, above the track.
What is Jev?
Jev is TypeSafe AI's first System One model. Instead of writing prose like a chatbot, it reads a situation and returns a typed, calibrated decision — a choice between options, a score on a scale, or a yes/no probability — in tens to hundreds of milliseconds, with no hallucinations. Picking jump, duck, or run for the hazard bearing down on the dino is exactly the kind of small, fast, bounded decision it is built for, which is why it can react in real time as the runner speeds up.
The same model powers ticket routing, content moderation, lead scoring, agent tool-call guardrails and model routing in real products. The Chrome dino game is just a visible, fun way to watch it decide — here every jump and duck is a genuine, typed API call, and the confidence reading shows how sure Jev was about each one.
Why the dino runner is a good decision demo
- It is a pure reflex loop: one hazard, three moves, a hard right answer — the shape of decision Jev is fastest at.
- The stakes are visible. A wrong jump or a missed duck ends the run instantly, so you can watch calibration matter frame by frame.
- It shows cost honestly. The running total is real input-token cost at Jev's own rate, so you can see how little a live decision costs.
- It mirrors real integrations. Swap 'jump / duck / run' for 'approve / review / reject' and this is the exact loop behind moderation or routing.
FAQ
Is Jev Dino free to watch?
Yes. It runs in your browser with no API key and no signup. The dino plays itself — every jump, duck, or run 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. When an obstacle enters the decision window the game sends one /api/jev choice question — jump, duck, or run — describing the incoming hazard, and Jev's answer becomes the dino's move. There is one call per obstacle; if a call is rate-limited or times out, a deterministic fallback keeps the dino alive and is marked as a fallback.
How fast and how expensive is each decision?
Each obstacle 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, and the confidence behind each jump, duck, or run.
Can I control the dino myself?
No — this is an autoplay demo of Jev deciding, not a game you steer. The dino runs automatically and Jev makes every jump/duck/run choice. If you want to play against Jev, try Jev Pac-Man or Jev Gomoku instead.
More: Jev 2048 — watch the AI play · Jev Tetris — AI drops the pieces · Jev Pac-Man — outrun the AI · All decision games · Jev API docs