Thoughtful Things launches Engram, an offline AI sampler built to hallucinate
The 16-pad groovebox runs custom local models to turn audio and voice prompts into deliberately uncanny sounds.
Thoughtful Things has launched a Kickstarter campaign for Engram, a 16-pad hardware sampler and groovebox that processes incoming audio with a local AI model designed to distort it into unfamiliar sounds. The instrument is aimed at experimental sound design rather than polished or radio-ready music.
Engram runs without an internet connection, using what the company calls a custom-trained “tiny AI” model. The Verge described the device as Thoughtful Things’ first instrument and reported that the model was designed in-house and custom-trained.
The device treats AI hallucination as part of the creative process. In a demonstration, founder Evan King made a spoken request for “piano”; Engram produced a glitchy sound that was recognisably, but only vaguely, piano-like. The company has described the instrument as a “field recorder for latent space.”
Thoughtful Things says users can feed audio into the machine and get results that fall between intention and accident. The approach draws on circuit bending: users are meant to be able to alter, and potentially break, the small models in pursuit of unusual results.
The company plans to open Engram’s firmware, allowing users to modify it and load their own custom models. It says its audio models were trained on open datasets containing commercially licensed audio, including CC-BY or similar licences, and says it has not used and will not use non-commercial, pirated or stolen data.
The Kickstarter campaign is for a limited production run, with pledges starting at $675. Thoughtful Things has not announced a retail price; the campaign presents that pledge as a 30% discount, with an anticipated retail range of roughly $850 to $900.
Tech Buzz characterised the launch as a test of whether musicians want a deliberately broken AI music tool, rather than one built to generate coherent professional output. Engram’s offline processing also keeps sound experiments on the device rather than sending them to a cloud service.