WHAT NEW IN X11.04 WED

AI collections from X · 22 collections · 66,612 posts

jevVIDEO

A collection of 10 open-source projects built on Jev

Ten weekend-sized projects to build with Jev access, including a 3D self-driving simulator, the 2B decider model, Hong Kong stock prediction, the WindTunnel browser agent benchmark and Vercel Labs' ai-cli.

Chen
@ch3nweiiiArchived 2026.10.02Originally posted 2026.10.02
Chen @ch3nweiiiOriginally posted 2026.10.02
Open original on X

Jev went from one API to a whole ecosystem in a few weeks. If you have Jev access and want something real to build this weekend, pick one of these 10: 1. typesafe-playground A self-driving car in a 3D world. Four camera feeds, radar and remembered speed signs, and Jev picks the lane and target speed. https://github.com/kavehmz/typesafe-playground 2. decider An open 2B model built to answer like Jev: one forward pass, no text generated. It plays Tetris, Breakout, Pong, Snake and Connect Four, and in Tetris it clears 20 lines against 0.6 for a random pick. https://github.com/Mapika/decider 3. jev_stock Type a Hong Kong stock code and Jev returns up, flat or down, with the probabilities in a clean HTML report. https://github.com/sosopop/jev_stock 4. WindTunnel A browser agent benchmark. Jev + Mercury 2.5 solved 49 of 49 tasks at about $0.0011 per attempt. https://github.com/nekuda-ai/WindTunnel 5. ai-cli Vercel Labs' terminal tool with Jev as the default evaluation model. Pipe a support ticket in and ask "refund requested?" in one line. https://github.com/vercel-labs/ai-cli 6. Safer with Jev A firewall for prompts. Jev checks the request for prompt injection first, and only clean requests get forwarded to your model. https://github.com/andrelandgraf/safer-with-jev 7. pi-fast-jev-compaction Keeps your agent's context lean. Two yes/no questions per tool call decide what stays, and your conversation text is never rewritten. https://github.com/joelhooks/pi-fast-jev-compaction 8. skillbox A self-hosted skills library for AI agents. Describe the task and Jev recommends which skill fits. https://github.com/kitze/skillbox 9. jev-superpowers Upgrades Claude Code, Cursor and Codex with typed decisions, package vetting and completion gates, so the agent can't just say "done". https://github.com/AkashPriyadarshii/jev-superpowers 10. Augustus A skill that decides where each judgment should live: code, a model, a classifier or a person. Jev is its default engine. https://github.com/24601/Augustus Copy these Jev blueprints, then read the full Jev setup below ↓ ↓

This post quotes another post: x.com/hanakoxbt/status/2101709924828934222

x.com/ch3nweiii/status/2105745919534907789Archived 2026.10.02 10:03