8 Jev projects to clone: trading, Doom, drones, robots

Ryven @imryvenOriginally posted 2026.10.05

Open original on X

Jev is the weirdest model of 2026, and it's already trading, playing Doom and piloting simulated drones pick one and copy it: jev-trader - Jev gets the order book and answers buy or sell, once every 300ms block. the bot itself posts the limit order https://github.com/jarrodwatts/jev-trader jev-doom-agent - real Doom running in the browser, two engines on the same map, Jev picking the tactical move every frame https://github.com/lukaske/jev-doom-agent jev-drone - a simulated quadrotor flying a five-station obstacle course on camera alone. flight control keeps it stable, Jev makes the higher-level calls about twice a second https://github.com/RomanSlack/jev-drone typesafe-mario - Jev picks one of seven controller actions every eight emulator steps, reading structured game state instead of screenshots https://github.com/fhshaik/typesafe-mario jev-t-rex-runner - for every new obstacle in the Chrome dino game, Jev picks jump, duck or keep running, and a short or full jump. the browser keeps the timing and the input https://github.com/joshlarsen/jev-t-rex-runner jev-ultrafast - Zurich to London on Google Flights in 7.1 seconds in its own demo. Jev picks the action and the element, a small text model only writes when something has to be typed https://github.com/browser-use/jev-ultrafast jev-robot-control - a simulated xArm7 arm puts an apple on a plate in MuJoCo, Jev choosing the movement direction and the gripper command at every step https://github.com/openroboto-ai/jev-robot-control diffjury - paste a public pull request link, one Jev call scores merge risk, review depth and blast radius, then returns approve with nits, request changes or block https://github.com/raihankhan-rk/diffjury killmyidea - describe a startup idea, one request puts ten questions to Jev, and plain code turns the scores into kill, fix or ship https://github.com/monteduro/killmyidea jev-curate - a Rust tool that runs Parquet and JSONL data through Jev checks like circular logic, sycophancy and unfinished code, and splits the rows into clean and rejected https://github.com/AkashPriyadarshii/jev-curate start with the one closest to a decision your agent already pays a full model to make

This post quotes another post: x.com/imryven/status/2104255840819491030