A developer embedded a Forth interpreter into NVIDIA Nemotron 3.5 Lightning's weights, inspired by the Qwenputer project, to test if a stack-based language suits weight-based compilation and if a 3B active parameter model can drive such a machine. The 30B MoE model (3B active) remains unmodified while a separate 27-layer network is integrated alongside it, with weights calculated via Python script. This experiment explores the feasibility of running programmable machines directly within smaller language model architectures.