THE JEV GUIDE · CURATED EDITION
Launch news, explanations, demos, and community projects around TypeSafe Jev, organized by topic. Every entry links back to its X source.
Curated selection · Not exhaustive · Updated manually
Sponsors
These products pay for the directory. Want yours up here?
The author points out that the so-called self-improving playtesting agent is just an LLM writing a game bot, fixing it after errors, essentially a Claude Code wrapper running on a Jev model key, and promotes their own game QA tool.
This visual novel game supports free input in addition to regular choices. Internally, it uses SystemOneModels defined after Jev's debut, which constrain output to numeric values, making them cheaper and faster.
UEFN Ducky introduces a free, open-source Play Tester Plugin that lets Jev or any chosen model play-test games while building them. Download it to have your agent play-test its own work.
ForgeAI shares results from 2,010 moves across 8 runs and 4 models, with the top 5 finishing within 225 points of each other; the model writes the plan and JEV picks each move.
The author points out that Jev's blockages were not due to bad decisions, but to issues like undetected text boxes, different addresses in the Catalan ROM vs the Spanish ROM, door loops, and stuck menus.
One card in the feed, every sixteen, on every page of it. $50 a week, cancel any week.
Mojofull updated the Flash slot in the 3-AI spectator mode of "One Exit" to OpenAI's decision-dedicated API, showing a direct comparison between Jev and gpt as non-text-generating decision-specialized AIs, with 1-step 12-question batch judgment.
A user named JustViewNews says they have built a Battle Royale game powered by decision models and asks whether they could get developer access to JEV to test the integration.
Maincode showcases the capabilities of Matilda and Matilda Jev by pitting them against each other in a poker game. Matilda is powered by the Agent SDK, while Matilda Jev uses the Client SDK's new systemOne.evaluate() API, highlighting how the two models approach the same challenge differently.
A developer reverse-engineered Meta Ray-Bans and used Opus 5.5 and Jev to create 3000+ Elo chess AI glasses. The entire workflow is open source.
The post showcases a game built with Claude Opus where the player types anything and the AI makes it happen in the world—cars, planes, dragons, or a meteor on another player. The game can be modified while playing and runs on Jev, GPT-OSS-120b, and more.
A developer reverse-engineered Meta Ray-Bans 3, combining Opus 5.5 and Jev to build AI chess glasses, and open-sourced the entire workflow.
In this chess variant, orders travel at light speed: pieces near the king act instantly, while distant ones take 1–3 turns to receive orders. Claude played against TypeSafe's Jev, who lost its queen and fell 12 behind before coming back to lead by 8.
