Engram Turns AI Hallucinations Into Experimental Music

Overview

Thoughtful Things has launched a Kickstarter campaign for Engram, a new hardware instrument that reimagines how creators interact with artificial intelligence. Rather than serving as a generic music generator, Engram functions as a sampler and groovebox designed to exploit the imperfections of small-scale AI models. The device aims to transform what are typically considered errors in machine learning—hallucinations and artifacts—into intentional, experimental soundscapes.

This approach marks a distinct departure from popular "push-button" AI music tools like Suno or Udio. While those platforms prioritize producing polished, radio-ready tracks, Engram is engineered for uncanny, glitch-heavy audio design. It invites users to explore the boundaries of local neural networks, treating the latent space of an AI model not as a source of coherent output, but as a raw, malleable terrain for sonic exploration.

"Think of Engram as a field recorder for latent space."

The device operates entirely offline, running custom tiny AI models directly on the hardware. This local-first architecture ensures that the processing happens in real-time without relying on cloud servers, offering musicians a self-contained unit for live performance and studio experimentation. By focusing on the breakdowns and distortions inherent in these small models, Engram positions itself as a tool for the avant-garde rather than the mainstream producer.

Technical Approach and Data Ethics

A central concern for many artists and developers is the legal and ethical sourcing of training data. Thoughtful Things has explicitly addressed this by stating that its audio models are trained exclusively on open datasets containing audio licensed for commercial use. The company emphasizes that it has not used, and will not use, non-commercial, pirated, or otherwise stolen data.

"We have not trained and will never train our models on non-commercial, pirated, or otherwise stolen data."

This commitment to ethical sourcing is particularly significant given the ongoing litigation in the music technology sector. Major labels such as Sony Music and Universal Music Group have recently filed lawsuits against companies like Suno, alleging blatant copyright infringement and deceptive trade practices. By contrast, Engram’s reliance on openly licensed material aims to sidestep these legal pitfalls while still delivering high-fidelity, albeit stylized, audio output.

The AI model driving Engram is custom-trained in-house. Because it runs locally, the device does not require an internet connection to function. This isolation not only enhances privacy for users who may be working with sensitive or unreleased material, but it also guarantees latency-free performance. Musicians can manipulate the AI’s output in real-time without the buffering delays often associated with cloud-based generative tools.

Sonic Philosophy and User Experience

The core philosophy behind Engram is rooted in the tradition of circuit bending. Originally a practice where engineers short-circuit consumer electronics to create unexpected audio feedback, circuit bending is adapted here for the digital age. Engram allows users to "bend" tiny AI models, tweaking their parameters until they break or hallucinate new sounds.

In promotional demonstrations, founder Evan King speaks into the device, requesting simple instruments like "piano." Instead of producing a clear recording, Engram responds with glitchy, fragmented audio that vaguely resembles the requested sound. This interaction highlights the device's primary appeal: it does not replicate reality but rather distorts it, creating textures that are difficult to achieve with traditional synthesizers or sample libraries.

Interactive Exploration

Users can interact with Engram through vocal commands and physical controls, encouraging a trial-and-error approach to sound design. The device is designed to be intuitive for those familiar with hardware samplers, yet accessible enough for explorers of experimental music. The goal is to make the internal state of a neural network audible, turning abstract mathematical probabilities into tangible sonic events.

Market Position and Availability

Engram is being introduced through a limited-run Kickstarter campaign, with early pledges starting at $675. This price point includes a 30 percent discount compared to the anticipated retail price, which is expected to fall between $850 and $900. While exact figures for the standard edition have not been finalized, the tiered pledge structure suggests that Thoughtful Things is targeting dedicated hardware enthusiasts and experimental musicians willing to invest in niche gear.

One of the key differentiators for Engram is its openness. Thoughtful Things plans to release the firmware, allowing the community to modify the device or load custom models. This modularity is intended to extend the lifespan of the product and foster a collaborative ecosystem around the hardware. By empowering users to tweak the underlying AI, the company hopes to inspire a diverse range of creative applications beyond its original design intent.

The launch of Engram arrives amid a growing interest in AI-powered music plugins and instruments. Recent developments, such as the Lunacy Audio Nova platform for building custom AI effects, indicate a broader trend toward tools that put creative control back in the hands of producers. However, Engram distinguishes itself by focusing on the hardware experience and the aesthetic of failure, offering a tactile way to engage with the unpredictable nature of machine learning.

As the tech industry grapples with questions of copyright and the role of AI in creative industries, Engram offers a pragmatic alternative. By prioritizing local processing, ethical data sources, and experimental outputs, Thoughtful Things is carving out a space for AI as a tool for discovery rather than mere automation. Whether Engram will resonate with the broader market remains to be seen, but it certainly provides a compelling case for the artistic potential of broken machines.