Built on Atome LM v2, (SuperESP Edition) it transforms a standard ESP32 into a tiny AI appliance capable of running twelve practical applications entirely offline.
A lightweight language model, Atome LM (944K parameters) has been successfully run on a $5 ESP32-WROOM-32 microcontroller not in simulation, but on real hardware. The model generates text offline at about 1 token per second, proving that LLM inference is possible on tiny chips without cloud support.
