User repository on GitHub
Laya vs Jev: local MLX and hosted AI decisions playing T-Rex side by side, with live metrics and replay recording
- Forks
- 11
- Last pushed
- Updated Sep 2026
Repo stats from the GitHub API, cached Oct 2026.
Jev directory
CommunityJev Cookbooks
Laya vs Jev — virajbhartiya
A side-by-side T-Rex game where local Laya runs on Apple Silicon through MLX and hosted Jev uses the TypeSafe API, showing choices, latency, survival streaks, and crash replays on one course.
- Stars
- 112
- Language
- Python
- License
- APACHE-2.0
- Reading time
- 3 min
- Published by
- Community
- Author
- virajbhartiya
- Last verified
- Sep 2026
About this project
Two AI decision models play Chrome's T-Rex game side by side: Laya runs locally on Apple Silicon through MLX, while Jev uses TypeSafe's hosted API. Watch their choices, response times, survival streaks, and crash replays on a shared obstacle course. Requires an Apple Silicon Mac, Python 3.11+, uv, and a TypeSafe API key for Jev. Initial setup downloads model weights; Jev needs network access and uses your API account.
Quoted from the project's README. Read the full README.
Works forJev for Classification
Used inJev for Gaming
Highlights
- A deterministic physics planner labels the moves and the models choose between them; safety checks can override unsafe or stale picks.
- Assisted scores measure the combined system rather than model skill, and --unassisted disables the safety shields for comparison.
- Both players finished two published assisted rounds without deaths; ties favor fewer live interventions when distances are equal.
- Includes staged courses, live latency charts, decision feedback, slow-motion crash replays, and 1080p 30 FPS JSONL replay export.
- Requires an Apple Silicon Mac, Python 3.11+, uv, downloaded model weights, and a TypeSafe key for Jev.
Quickstart
bash
uv sync --extra demo --extra trex --extra dev
uv run --extra demo hf download aac6fef/laya-multilingual-mlx --local-dir models/hub/laya-multilingual-mlx
cp .env.example .env
uv run --extra trex laya-trex --round-seconds 0 --env-file .envWatch out
Apache-2.0 with attribution in NOTICE, and Chromium game assets under BSD-3-Clause. Needs an Apple Silicon Mac with Python 3.11+ and uv; weights download separately and Jev calls use your API account.
Questions about this entry
Short answers drawn from the linked project or page.
No. The 13.4 ms median and the 7.4 ms multilingual figure come from the underlying Laya-MLX runtime's one-question API benchmark, and the README states explicitly that those are not T-Rex gameplay measurements. Gameplay uses a three-question loop per move plus the safety layer, and the benchmark excludes model loading and uses batch_size=64 for the throughput row.
The README publishes no standing winner. A deterministic physics planner labels the moves and each model chooses between them, so an assisted score measures the model plus the safety layer that can override unsafe or stale picks. Use --unassisted to drop the shields. Equal-distance ties in timed matches favor fewer live interventions, which is itself affected by how often each model decides.
Yes, for Laya alone: uv run --extra trex laya-trex --players laya --course random --round-seconds 0 needs no API key. Jev goes through the hosted TypeSafe API, so it needs network access and a key on your account. Either way you need an Apple Silicon Mac, Python 3.11+, uv, and a separate download of the model weights.
Not model skill alone. Crashes restart a player and play continues, and the visible safety layer can override a proposed move, so a streak is a property of the combined system. The README keeps this separate from the best-move statistic, which measures agreement with the planner. Staged courses, latency charts, decision feedback, and slow-motion crash replays are in the demo; full controls are in docs/TREX_DEMO.md.
Chrome's T-Rex game: Laya runs locally on Apple Silicon through MLX and Jev uses TypeSafe's hosted API, with visible response times, survival streaks and crash replays on a shared course.
A deterministic physics planner labels the moves and safety checks can override unsafe or stale choices, so assisted scores measure the combined system; use --unassisted to disable the shields.
Run laya-trex with --record artifacts/trex/run.jsonl --record-seconds 300, then export that file to a 1080p 30 FPS MP4 with the export subcommand.
Reactions & coverage
Posts, threads, and videos about this entry from around the web.
Hacker News: Show HN: OpenDecision, a local decision model with a Doom demo.
dwa35926 pointsShow HN: OpenDecision – a 400M zero-shot model makes local decisions, plays DoomRead the thread on Hacker NewsX: Standard One SH (StandardThinking)
LinkedIn: What multimodal decision models actually change, and what they do not.
LinkedInJevDirectory.org
Hacker News: Decision Index 0.2.1 scores open reproductions of Jev on a frozen suite.
jackbravo4 pointsDecision Index 0.2.1 – 70 open reproductions of TypeSafe Jev's Decision ModelRead the thread on Hacker NewsOther listings of this project
Forks and copies of this project in the directory. This entry is the canonical one.
An independent Laya-versus-Jev comparison on one RTX 5090: phishing detection over 2,000 emails, the typed-decisions suite, latency scaling to 100 questions per pass and eleven probe assertions.
Jev Benchmarks & EvaluationsDocs#community#benchmark#evaluation
Community
More Jev resources like this
A browser Snake game where code computes the legal moves, food distance, and reachable space and Jev picks one move per tick with a single Choice question; a late answer leaves the snake going straight.
Jev Cookbooks#community#typescript#games
Community
sorrycc/typesafe-snakeTypeScript
sorrycc/typesafe-snakeTypeScriptTwo WebAssembly builds of the Chocolate Doom 3.1.1 engine play the same Freedoom map from the same initial state, with a Jev Choice decision picking each tactical macro and a local controller turning it into inputs.
Jev Cookbooks#community#typescript#games
Community
lukaske/jev-doom-agentTypeScript
lukaske/jev-doom-agentTypeScriptAn original Civilization II browser harness for Jev: named Choice vectors cover empire policy, cities, research, diplomacy, units, exploration, and warfare, with live decision probabilities in a Roman-themed spectator display.
Jev Cookbooks#community#python#games
Community
phyous/tsai-civ2Python
phyous/tsai-civ2PythonA desktop Snake experiment where Jev assesses the board each tick and Python combines its answers into a move; one recorded run reached 29 foods in 461 ticks before trapping itself.
Jev Cookbooks#community#python#games
Community
siroccomask/snake-jevPython
siroccomask/snake-jevPythonAn experimental controller that has Jev choose NES inputs to play Super Mario Bros. from structured emulator telemetry instead of screenshots, logging every decision with probabilities and latency.
Jev Cookbooks#community#python#games
Community
fhshaik/typesafe-marioPython
fhshaik/typesafe-marioPythonA MuJoCo and Franka Panda workbench where Jev chooses the next manipulation step for pick-and-place, stacking, and obstacle-carry tasks, with rule baselines and side-by-side model comparison.
Jev Cookbooks#community#python#robotics
Community
FBddcz/embodied-jevPython
FBddcz/embodied-jevPythonFrom the Jev guides
Original write-ups on what this entry is for.
From the Jev AI model community
Posts from builders shipping with Jev right now.