Music startup Thoughtful Things has launched a Kickstarter campaign for Engram, a sampler and groovebox built around a small local AI model. Rather than generating polished songs from text, the instrument is designed to distort recorded audio and produce unstable, unfamiliar sounds that musicians can shape.
Engram works without an internet connection. The company says its model was trained in-house on open datasets containing audio licensed for commercial use, such as CC-BY material, and not on pirated or non-commercial recordings. It also plans to open the firmware so owners can alter the device or load custom models. In a demonstration, a spoken request for “piano” produces a glitchy approximation rather than a conventional sampled instrument.
The limited Kickstarter run starts at $675, described as a 30% discount; the company has not fixed a retail price but indicates a likely range of $850 to $900. Crowdfunding carries delivery and product risks, and the claims come from the maker rather than independent testing. Engram’s distinctive choice is to treat model errors as a sound-design surface, combining local generation with the hands-on unpredictability associated with circuit bending instead of offering another automated song service.