Thoughtful Things has unveiled Engram, a hardware sampler and groovebox with a small audio model built in. Where most AI music products promise a finished song from a text prompt, Engram treats the odd, glitchy sounds such models produce as raw material for musicians to work with.
The company calls the idea "audio model bending", borrowing from circuit bending. The unit works offline, with no app, account or subscription. Users can feed in their own recordings, generate short sounds and then slice, sequence and distort them like any other sample. The listed hardware includes an embedded neural audio codec, dedicated model-bending controls, swappable audio models, MIDI in and out, CV sync and gate, a stereo line input, a headphone output and a built-in microphone.
In the launch material, founder Evan King asks the device for piano and gets something only loosely piano-like, which is the intended effect: sound close enough to spark an idea, wrong enough to stay interesting. WeRSM reads this as a move away from AI tools that aim to complete the work and towards ones that keep the musician in control of how the model misbehaves. Running offline makes the model feel like studio gear rather than a remote service, and the firmware is meant to accept new audio models, so the hardware could stay open to experiment.
There is a cultural tension here too: AI firms have spent years making output smoother and more human, and Engram bets some creators want visible seams instead. The company describes Engram as a research prototype, and a limited Kickstarter run reportedly starts at $675, with a higher retail price to follow. It is not aimed at the mass market, and WeRSM says its uptake will depend on whether musicians find the hands-on approach worth paying for dedicated hardware.
Sources · 1