./play d1

You're playing d1-omni-600M, a small decision model I fine-tuned on Stockfish-labelled positions (how). It runs in your browser: the first visit downloads a 339 MB 4-bit copy from Hugging Face, then it's cached. Policy mode plays its first instinct; search mode looks 2–3 moves ahead with its own move and value heads (best with WebGPU).