Memory doesn’t just replay on Engram – it mutates, regenerates, and answers back. Evan King describes this second-revision prototype as a hardware sampler that runs embedded generative audio models directly in the box, without needing an internet connection or subscription.
The familiar side of the instrument is a sampler workflow with a step sequencer, multiple audio channels, sample slicing, FX, and CV/MIDI I/O. The unfamiliar side is the generation layer: blank-slate generation from a prompt, guided generation that builds from existing sounds, and “model bending,” where knobs intentionally push the MusicGen decoder into glitchy, unstable territory. Voice prompting is powered by Moonshine, making Engram less like a static sample player and more like a compact, experimental sound-design platform.