User repository on GitHub
Jev Plays Minesweeper
- Forks
- 1
- Last pushed
- Updated Sep 2026
Repo stats from the GitHub API, cached Oct 2026.
Jev directory
CommunityJev Demos & Experiments
Fly vs Jev: Minesweeper
Jev plays Minesweeper against the full wiring of an adult fruit fly brain: both see only revealed clues on the same seeded 9×9 boards, Jev judging every covered square safe or mined. Over 16 held-out boards the fly averaged 59.8 safe squares and Jev 59.1, against 52.6 for random clicks.
- Stars
- 1
- Language
- Python
- Reading time
- 3 min
- Published by
- Community
- Author
- enesyilmazcode
- Last verified
- Oct 2026
Used inJev for Gaming
- Languages
- Python
- Capabilities
- Open source
Highlights
- Jev gets a coordinate-labelled text board and judges every covered square safe or mined, clicks its safest pick and flags the lowest-safety squares.
- The fly's 166,700 neurons and 6.2 million connections run as a deterministic spiking network for 150 ms per candidate click, never trained or changed.
- Sixteen held-out boards: fly 59.8 average safe squares, Jev 59.1, random clicks 52.6, and a constraint solver a perfect 71 with 16 clearances.
- The fly led on 9 boards, Jev on 6, one tied — not statistically clear, with the README's sign test at p = 0.61.
- A guaranteed-safe centre opening and a seed fix each 9×9, 10-mine board before either player moves, and neither is ever shown a hidden mine.
Watch out
Grounded on the truncated README and GitHub metadata read on 2026-10-15; the excerpt ends inside the controls table, so the full controls comparison is not covered. The repository declares no license in its GitHub metadata, and over larger pools behind the video the fly cleared 2 of 510 ten-mine boards and Jev 2 of 500.
Questions about this entry
Short answers drawn from the linked project or page.
A simulated fruit fly brain against an AI model at Minesweeper. The full wiring of an adult fly — all 166,700 neurons — sees only the revealed clues and picks every square, and TypeSafe Jev gets the same view on the same seeded boards.
Jev gets the rules and a coordinate-labelled text board. It judges every covered square as safe or mined, clicks its safest pick, and flags as many lowest-safety squares as the board has mines. Hidden mines never enter either player's view.
Over 16 held-out boards the fly averaged 59.8 out of 71 safe squares and Jev 59.1, against 52.6 for random clicks; the constraint solver cleared all 16 perfectly. The fly led on 9 boards, Jev on 6 and one tied — a sign test gives p = 0.61.
Each candidate click runs through the MaleCNS connectome as a deterministic spiking network for 150 ms, turning the visible board into photoreceptor spike rates. A readout over 2,048 L1/L2 neurons scores the candidates, and it was fit once from a constraint solver that saw only what a player sees.
The readout. It picked an optimal or score-tied click on 53.5% of held-out solver positions against 5.3% for chance, and the same readout wired straight to the 162 eye channels with no brain in between still reached 48.5%, so most of the skill comes from what was trained.
Neither cleared any of the 16 held-out boards. Over the larger pools behind the video's ladder the fly cleared 2 of 510 ten-mine boards and Jev 2 of 500 — ten mines is near the edge of what either can finish. Both still substantially outlast random clicking.
The fly's readout was fitted in a one-time setup on Modal GPUs over 500 positions and 10,064 candidate clicks, and during play the solver is never consulted. The captured README does not cover run commands or Jev-side credentials, and the GitHub metadata declares no license.
More Jev resources like this
A 12,000-neuron fly-brain subgraph of the maleCNS connectome fights Jev in mk.js: the fly simulates spikes in the browser at 10 decisions per second while Jev reads the fight as JSON and answers one typed choice among seven moves.
Jev Demos & Experiments#community#catalog-import#logicrw
Community
lavallee/mk-jev-fly-brainJavaScript
lavallee/mk-jev-fly-brainJavaScriptBrowser-only Laya demo suite: four Vite apps — a Traditional Chinese decision playground, Gomoku, Big Two and a 3D maze — run the Laya ONNX model on-device through Transformers.js, downloading fp16 weights from Hugging Face on first use.
Jev Demos & Experiments#community#catalog-import#madewithjev
Community
killkli/open-jev-layaPython
killkli/open-jev-layaPythonFire-evacuation simulation where every occupant gets a Jev-style typed decision engine: a local Ternary Bonsai 2 27B, Qwen3.5-4B or the TypeSafe Jev API reads choice probabilities, while movement runs on validated social-force physics.
Jev Demos & Experiments#community#catalog-import#madewithjev
Community
using76/typeevacsafePython
using76/typeevacsafePythonAn experiment where Jev plays Zork I: every turn Jericho supplies the valid actions and Jev answers one Choice of which command to type, with confidence bars, an anti-loop memory of the last 8 turns, a JSONL log and a French replay dashboard.
Jev Demos & Experiments#community#catalog-import#logicrw
Community
resadan-dev/jev-zorkPython
resadan-dev/jev-zorkPythonA Three.js first-person garden game where a glowing fly hunts goblins and TypeSafe's Jev steers: about five times per second it answers four typed questions on turning, sidestepping trunks, climbing and diving, and every flight is logged.
Jev Demos & Experiments#community#catalog-import#madewithjev
Community
webdevcody/jevs-flyJavaScript
webdevcody/jevs-flyJavaScriptAn autonomous ViZDoom agent that decouples Doom's simulation from Jev SystemOne inference: the game runs at 35 ticks/sec while Jev is queried asynchronously at about 10 Hz for macro goals, targets, movement and firing decisions.
Jev Demos & Experiments#community#catalog-import#logicrw
Community
amoghcreator/doom-jevPython
amoghcreator/doom-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.