Frigade · 2026-09-25 · notable
Jev Plays Pokémon Red — a decision model makes every move, live
Jev Plays Pokémon Red is an open-source live stream where TypeSafe's Jev decision model makes every menu, route and battle choice in the game and shows its odds for each one.
A typed-decision model plays a full RPG, one probability-scored choice at a time.
What is it?
Jev Plays Pokémon Red is a project by Christian Mathiesen at Frigade in which the Jev decision model plays the Game Boy game on a live stream. Jev picks menu answers, starter and team choices, where to go, when to heal or train, and every battle action. A side panel shows each decision and Jev's odds.
How does it work?
A Node.js harness runs a Game Boy emulator, reads game memory, and turns the current situation into a list of legal options with context. Jev is called through Vercel AI Gateway to pick one, and the harness only presses buttons — it never writes to game memory. The README counts 800 to 1,300 calls an hour, about 1,200 tokens each, at roughly $1 to $1.70 a day.
Why does it matter?
The project is a long, public stress test of the typed-decision idea behind Jev: can one-shot choices from a fixed list carry an agent through hours of play? It reached 221 points as a Show HN, and every call is logged with its probabilities and latency, so anyone can study where the approach gets stuck.
Who is it for?
developers experimenting with decision models and game agents
Try it
git clone https://github.com/christianmat/jev-pokemon && cd jev-pokemon && npm install