In development
Gizmo Guard
Edge AI that keeps inference on the device, with small local models doing useful work without shipping your data anywhere.
The idea
Most AI products assume a round trip to somebody else's datacenter. For a whole class of devices, including industrial sensors, cameras and embedded controllers, or anything in a regulated or disconnected environment, that assumption is a non-starter. The data is sensitive, the link is unreliable, or the latency budget simply does not allow it.
Gizmo Guard is our work on the other approach: run small, capable models directly on the hardware. Inference stays local, data stays where it was collected, and the device keeps working when the network does not.
Where it stands
Gizmo Guard began as an experiment in what current small models can actually do on constrained edge hardware. That exploration is public on GitHub, and it is the most accurate picture of where the work is today.
We publish product pages before the product ships on purpose, because it keeps us honest about what exists and what does not. Everything on this page describes work in progress, not a shipping release.
See it working
An early prototype, on camera
Worth being clear about what this is: a proof of concept, demoed live at CodeMash 2026. It is a long way from the product. What it does show is the core idea running end to end on real hardware.
A Raspberry Pi with a camera, a Spring Boot app, and Spring AI describing the scene in natural language, then reacting when something changes.
- Motion detection with no extra sensors
- Models running locally, no external API keys
- No data leaving the network
- Close to zero hardware budget
Building something adjacent?
If this overlaps with your roadmap, we would rather talk early than late.