Local gesture control for Home Assistant.
A handful of intentional shapes, chosen so they never trigger from regular hand movements around the room. Easy to remember, hard to fire by accident.
open_palm
Open the palm: the selected entity toggles. Example: a lamp turns on, a switch flips.
Keep the palm open: the selected entity cycles through its preset values. Release to lock the one you want. Example: dim a light through its brightness levels.
l_shape
Form an L with thumb and index: cycle through the entities you've assigned. Release to lock the selection. Example: pick which light in the room to control next.
From the couch, across the room, in the dark, with friends. The camera and the model are designed for real-room distances, not the close-up demos you see in product videos.
Off-the-shelf hand-detection models are tuned for the conditions of public datasets: close range, good light. They fall apart the moment they meet a couch, a dimmer, or an infrared camera.
We use a proprietary training pipeline with data captured in real homes, specifically tuned for this use case. The result is small enough to run on low-resource hardware, with a Raspberry Pi 5 as the minimum.
Mini-PC, x86 desktop, ARM board, Raspberry Pi. If it runs Linux or Windows with Docker, or Home Assistant OS, Superhome runs on it. CPU-only inference, no GPU, no external service.
rtsp://192.168.1.42/stream0
Camera, model, and Home Assistant all run on your network. Nothing leaves the house.
A small ESP32-S3 display that shows what Superhome sees, in real time. It subscribes to HA sensors over the native API, so feedback lands in 20–40 ms. Optional. Bring your own panel.
JC3248W535C_I GUITION on AliExpress →Add the Superhome app repository to your Home Assistant Supervisor in one click.
Sign up to get a free beta license. The app installs everything it needs on first run, with no manual setup.
Paste your license key and the RTSP URL of any camera you trust into the app's Options tab. Pick the lights and switches you want to control, save, and you're done.
Everything Superhome ships today, at no cost while we polish the product.
The permanent free tier that takes over when beta ends. Reduced scope, no expiry.
For households and tinkerers running multiple cameras and devices.
Yes. Superhome runs entirely on your local network: the camera, the computer-vision models, and Home Assistant all stay in your home, and no video ever leaves the house. The only thing online is a free account to obtain your license.
Any RTSP camera works (IP cam, NVR, or USB webcam, infrared or visible); we recommend a Tapo C210 (2K) or at least a 1080p/HD camera for reliable detection. You also need a machine that runs Home Assistant OS or Docker. A Raspberry Pi 5 is the minimum; inference is CPU-only, so no GPU is needed.
Yes. The models are trained for infrared and low-light conditions, so gestures are recognised even in total darkness, unlike off-the-shelf models.
Superhome reliably detects gestures at 9–10 metres, designed for real-room distances like across a living room, where off-the-shelf hand-tracking models stop working at about 1–1.5 metres.
Superhome is free during the public beta, with the full gesture set, infrared and low-light support, and one camera per device. A permanent free tier and a paid Pro plan for multi-camera setups arrive after beta.
Two shapes drive three commands: an open palm taps to toggle (air_click) or holds to cycle through values, for example brightness levels (air_hold), and an L-shape with thumb and index selects which entity to control (l_select).
Yes. Superhome installs as a Home Assistant OS add-on or a Docker container on your LAN and pairs with Home Assistant out of the box, driving any entity type: lights, switches, covers and more.