How on-device AI protects privacy in dementia care technology
By Spiro Khoury · Updated 7 August 2026
A memory aid that watches and listens all day raises an obvious question before it raises anything else: where does that footage go? For most consumer AI devices, the honest answer is “to a server somewhere,” processed in the cloud and often retained. For a device meant to sit with someone living with dementia or mild cognitive impairment (MCI), often in their own home, that answer isn't good enough. This guide is about the architectural choice that changes it: doing the processing on the device itself, instead of in the cloud.
Cloud processing vs. on-device processing
Most AI assistants work by sending what they capture — audio, video, or both — to a remote server, where a large model processes it and sends a response back. That's what makes them powerful, and it's also what makes them a data-collection product: the recording has to leave the device to be useful, and once it has left, what happens to it is largely out of your hands.
On-device (sometimes called “edge”) processing runs that same kind of model locally, on hardware inside the device itself. Nothing has to be transmitted for the assistant to work. The trade-off is real — on-device models are more constrained than the largest cloud models, and building for constrained hardware is harder engineering — but for a device that's on for most of someone's day, it's the difference between a private memory aid and a camera that streams a vulnerable person's home to a server.
Why the stakes are higher here specifically
Ordinary privacy advice — read the policy, adjust the settings — assumes the person being recorded is the one deciding whether to use the device, moment to moment. That assumption gets shakier as memory loss progresses. Someone with MCI or dementia may not reliably remember that a device is recording, may not be able to review a privacy policy meaningfully, and is often being observed inside their own home, historically the one place surveillance doesn't reach. Family members and carers see the output too, which is its own layer of exposure on an already difficult situation. None of that is an argument against this kind of technology — the research on why it can help is real, and we cover it in our guide to wearable cameras and memory loss — but it is an argument for holding the privacy bar higher, not lower, than a typical consumer gadget.
What this looks like in practice
A useful test for any device in this category: can it tell you exactly what it did with what it just recorded, in plain language, and does that answer hold up if the device is examined rather than taken on faith? For Syntropic Labs specifically, the answer is that everything runs on device — none of your data goes to the cloud or the internet, and you own your data. Video and photos are processed in real time and are not retained long term. Full detail on what's collected and why is on our Privacy page.
That's the practical meaning behind the line we use everywhere else on this site: it listens and watches in the moment, keeping the memory, never the recording. The memory — a text summary, an answer to “where did I leave my glasses,” a note that medication wasn't taken — is what's kept. The raw footage that produced it isn't.
Strictly memory support, no claim to treat, cure, or slow cognitive impairment. See Products for how this applies to each device, and Research for the studies behind the approach.