Jev plays Snake
Typed AI decisions inside a Snake game loop, compared with code baselines through recorded runs.
Open the original ↗
Decisions inside a game loop
What happens when a model chooses Snake’s next move? This experiment compares TypeSafe’s Jev with random, greedy, and flood-fill code baselines. The engine owns movement and collision rules; the model chooses from the supplied directions.
The recorded runs compare raw game state with code-computed geometric features. They make the latency and decision-quality trade-offs visible without treating one seed as a general benchmark.
Watch the Snake recordings · Read the code and experiment notes
Playback uses saved runs and makes no model requests. The repository also includes offline baselines and optional local API experiments.
