What it does
The local suite includes rule and random baselines alongside Jev and nuclis models. Its driving world credits a permitted JevPilot rewrite; image levels require another model because the Jev adapter rejects pictures.
Playable demos & bots
Compare decision models in simulated driving, airport and support-desk worlds, watching their choices or collecting results from the command line.
Only you can see your notes.
Screenshot unavailable. Open the experiment ↗
The local suite includes rule and random baselines alongside Jev and nuclis models. Its driving world credits a permitted JevPilot rewrite; image levels require another model because the Jev adapter rejects pictures.
Before using AI in your app, compare its decisions with simple rules on the same practice cases. A starting number called a seed keeps those cases fixed. A side panel shows what was asked and answered. A developer can also collect result tables without opening the game's screen.
Start with simple rules, which need no AI access key. To compare Jev, you need a key for its paid service and a level without picture questions. Keep a record of which model answered. Doing well in these made-up worlds does not show that AI can drive a real car.