There's a particular kind of gadget that makes you lean in, squint, and ask, "What in the name of all that is glitchy is that thing?" Engram is exactly that. Thoughtful Things has put a Kickstarter out for a sampler and groovebox that doesn't just chop up your audio, it feeds it to a local, in-house AI model and records whatever crawls back. No cloud, no Suno-in-a-box, no push-button pop hits. Just a tiny, custom-trained model living inside a box with 16 pads and 8 knobs, designed to produce the uncanny, the experimental, and the wonderfully weird. It's a bold swing, and frankly, we're here for it.
This thing is the anti-smart-curtain. We recently covered The sun does your drapes' dirty work now, no motor required, a product that uses passive solar mechanics to remove a mundane task from your life. Engram does the opposite: it takes a mundane input, like a sample of your voice or a drum hit, and introduces chaos. It's not about convenience; it's about surrendering control to a machine that hallucinates. That's a fundamentally different relationship with technology, and it's a refreshing one. The smart curtain promises a smoother day. Engram promises a stranger one. We'll take the stranger.
But let's get to the razor clam in the spec sheet, because there's always one. The model is "designed in-house and custom-trained," and it runs entirely locally. That's a significant engineering claim for a crowdfunded first instrument. It means no latency from a cloud round-trip, no subscription fee, and no privacy concerns about your audio leaving the room. It also means the entire promise of the device hinges on whether Thoughtful Things has actually built an AI that's compelling enough to justify the hardware. A sampler that just plays back your samples is a dime a dozen. A sampler that mangles them in interesting, unpredictable ways? That's a reason to exist. The catch, as with any Kickstarter, is that we're betting on the execution, not the render. The listing says it's real, but until it ships, it's a promise with a very interesting proof of concept.
If a reader asked us whether to back this, we'd say this: only if you're comfortable with the idea that the glitch gods are fickle. This isn't a tool for finishing a track; it's a tool for starting one, for finding a sound you didn't know you were looking for. The practical value is in the creative disruption, not the workflow. And the specific detail to watch isn't the AI itself, it's the training data. A custom model is only as good as its corpus, and Thoughtful Things hasn't told us what they fed it. If it's a few hundred hours of circuit-bent toys and field recordings, great. If it's a few thousand hours of generic synth presets, you'll know within the first five minutes of playing. That's the question we'd want answered before we hand over our money, and it's the one detail that will determine whether Engram becomes a beloved instrument or a very expensive paperweight that occasionally burps. Stay spooty.
