The short version
- Thoughtful Things has launched a crowdfunding campaign for Engram, a hardware sampler designed to manipulate audio using local artificial intelligence models.
- The device operates without an internet connection and relies on in-house trained models built exclusively from commercially licensed open datasets.
- Pledges begin at $675, with the company indicating that future retail pricing will likely range between $850 and $900.
A new musical instrument designed to harness the unpredictable nature of artificial intelligence has entered the crowdfunding stage. Thoughtful Things, a music technology startup, has initiated a campaign for Engram, a sampler and groovebox that processes incoming audio through local AI models. The device is not intended to function as an automated song generator or a tool for creating radio-ready pop tracks. Instead, it focuses on experimental sound design, allowing users to distort existing recordings and generate entirely new, often glitchy, sonic textures.
The core appeal of Engram lies in its ability to produce what the developers describe as hallucinations. Rather than striving for fidelity or realistic replication, the system is engineered to push audio models beyond their standard operational limits. This approach results in uncanny, distorted outputs that may only vaguely resemble the original input. For example, a user might speak the word piano into the device and receive a fragmented, glitch-laden approximation of the instrument rather than a clear recording. The goal is to treat the AI not as a precise synthesizer but as a source of creative error and unexpected texture.
Privacy and data integrity are central to the device’s architecture. Engram functions entirely offline, meaning it does not require an internet connection to operate or process audio. This design choice ensures that user inputs and generated sounds remain local to the hardware. Furthermore, the company has emphasized its commitment to ethical data sourcing. The internal models were trained using open datasets containing only audio licensed for commercial use, such as those under Creative Commons Attribution licenses. The developers have explicitly stated they will not train their systems on pirated, non-commercial, or stolen data.
The technical foundation of Engram relies on what the creators call tiny AI. These are compact models designed to run efficiently on the device’s local hardware without cloud dependency. Unlike larger generative models that require significant computational power and server-side processing, these smaller iterations allow for real-time manipulation of audio streams. The in-house design allows for specific tuning toward experimental outcomes rather than general-purpose audio generation. This localized processing also contributes to the immediate feedback loop essential for live performance and improvisation.
Inspiration for Engram draws from the subculture of circuit bending, a practice where electronic devices are modified to create new sounds by altering their internal circuits. The device encourages users to tweak and even break its AI models in similar ways. By adjusting parameters and pushing the system into unstable states, musicians can discover unique sonic artifacts. This methodology shifts the focus from perfect reproduction to exploratory interaction with the machine’s latent space. The founder describes the instrument as a field recorder for this abstract digital environment, capturing the erratic behavior of neural networks.
Community involvement is planned through open firmware access. Thoughtful Things intends to allow users to modify the device’s software and load custom models. This openness supports a ecosystem where developers and musicians can experiment with different AI architectures or training sets. It also aligns with the broader ethos of hardware instruments that value user customization and technical transparency. By providing tools for firmware tweaking, the company positions Engram as a platform for ongoing innovation rather than a closed, static product.
The initial release will be limited to backers of the crowdfunding campaign. Pledges start at $675, which represents a significant discount compared to anticipated retail prices. The company has not finalized the standard market price but suggests it will likely fall between $850 and $900. This pricing structure reflects the specialized nature of the hardware and the costs associated with developing custom AI models for local deployment. Early adopters will receive units as part of this limited run, while broader availability remains uncertain until after the campaign concludes.
The launch of Engram highlights a growing niche in music technology that embraces imperfection and algorithmic unpredictability. While many AI audio tools aim to streamline production or mimic human performance, this device seeks to complicate the creative process by introducing controlled chaos. It offers musicians a way to interact with the raw, unpolished outputs of neural networks. As AI becomes more integrated into creative workflows, instruments like Engram provide a physical interface for exploring the boundaries between intentional composition and machine-generated error.
Sources behind this briefing
Go to the original reporting
- The Verge↗Engram is a sampler that turns broken AI hallucinations into music
