jev-experiments
Jev plays Chrome Dino and a local shooter arena from structured state, with speculative jump/duck questions and code-owned execution.
Overview
jev-experiments contains System One game loops where Jev, or a labelled heuristic fallback, sees structured state and returns typed actions while Python executes them. Dino-Jev drives the real chrome://dino game in Chromium, and Krunker-Jev drives a local Krunker-style arena; both have browser inspectors. The canvas is only a view and never goes to the model.
In Dino-Jev, code reads the game's internal Runner state (pose, speed, obstacles), sends one TypeSafe request per tick with an action Choice (run, jump or duck), speculative jump-now and duck-now Nouls, and an urgency Score, following the speculative fan-out pattern. Code gates execution: high pterodactyls cannot be jumped into and cacti cannot be ducked, and actions below 0.28 confidence keep the last action. Urgency is displayed only.
What the decision model decides
| Type | Question | Answers |
|---|---|---|
| Choice | Should the dino run, jump or duck? | run · jump · duck |
| Yes / no | Should it jump now? Should it duck now? | — |
| Score | How urgent is the situation? | — |
How it works
- Code snapshots game state from the engine
- One request per tick asks action, speculative jump/duck and urgency
- Policy gates execution and keeps the last action below 0.28 confidence
- Inspector shows decisions; the canvas stays local
Strengths
- Speculative fan-out keeps one request per tick
- Policy gates prevent impossible moves
- Heuristic baseline for comparison
Caveats
- Game experiments only
- Requires a key for Jev policy
Getting started
`pip install -e '.[dev]'`, set JEV_API_KEY, then `python -m dino_jev play --policy jev --seconds 20`.At a glance
- Type
- Project
- Status
- prototype
- Language
- Python
- Runtime
- Chromium automation with local inspector
- Has code
- yes
Use cases and integrations
How curated lists describe it
Uses Jev to play Chrome Dino and a local shooter arena while Python executes structured decisions.
logicrw/awesome-jev-projects
Frequently asked questions
What does Jev decide?
Whether to run, jump or duck, plus speculative jump-now and duck-now answers and urgency.
Does Jev see the game screen?
No, it receives structured state read from the game engine.
Is there a baseline?
Yes, a heuristic policy can play for comparison.
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.
ProjectbetaTypeScriptBrowser chess lab where Jev, LLMs and people play, each Jev turn being one Choice over the legal moves, with Stockfish analysis.
ProjectbetaPythonNiceGUI 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.
ProjectprototypeJavaScript