A runner-game prototype uses TypeSafe AI's Jev decision model to choose terrain from options defined by the developer. The game sends the player's position, movement and nearby terrain, then asks for platform widths, gaps, heights and surface types. The developer's own game code places the selected blocks.
The prototype's write-up reports API response times of 319 to 375 milliseconds across five requests. This applies the model's structured-output design, in which possible outputs and their structure are defined in advance, to a bounded task during play: the model chooses among permitted terrain configurations while the game retains control of how those choices become a level.
Terms in this piece · Glossary
LLM — A large language model — the neural network behind tools like Claude and ChatGPT, trained on huge amounts of text to predict what comes next.
