Edge AI for smart glasses
Heavy AI for light glasses.
Edge Offload is a small box on the local network of a café, office or home. AI glasses hand it the models too large to run on your face, and it answers in real time.
Camera frames stay in the building.
Where should the AI run?
The models that make glasses useful need more computer than a pair of glasses can carry. Today that work goes to one of three places, and each has a cost.
Too small
Large models on the frame mean heat, weight and battery drain, so the glasses run small ones.
Too hot
The phone in your pocket throttles as it heats up and drains its battery while it works.
Too far
Every request travels to a data center and back, adding delay, cost and a copy of what you saw.
Edge Offload
A dedicated GPU a few meters away answers in real time, and what your camera sees stays in the building.
How it works
The box is an appliance: it needs power and a network cable.
-
Plug it in
The box connects to the network by Ethernet. In a venue it also runs its own Wi-Fi network for guests' phones.
-
Your phone finds it
The phone app that drives your glasses discovers the box on the local network and connects to it securely.
-
The box does the heavy work
Vision, speech and language models run on the box's own GPU, and results stream back to the app and onto the lens.
What it does
The first two jobs we're building for.
Visual assistant
Look at something and ask about it. The box keeps a running picture of the scene, so the answer starts from what is in front of you.
Live translation
Speech in one language becomes captions in another on the lens, including on glasses that have no camera.
Private by default
A camera on someone's face is a serious thing to handle. The box is designed so that it keeps very little.
- Camera frames and audio are processed in memory and never written to disk.
- Transcripts and detections last for the session, then they are deleted.
- A venue sees health and usage counts, never what its guests saw or said.
- Cloud processing is off unless you turn it on, and is never used for guests at a venue.
Built for the places people wear glasses
Cafés and shops
One box behind the counter serves every guest's glasses on the venue's own network.
Offices
A team's glasses share one box on the office network, and camera feeds stay inside the building.
Homes
A box on the home network serves the household's glasses and phones.
Talk to us
Edge Offload is in development. If you run a venue, make AI glasses, or operate a wireless network, we would like to hear from you.