Put the food on the scale
The one you own. The glasses read the display, not a Bluetooth radio, so nothing to pair and nothing to buy.
You eat.
It logs.
Ray-Ban Meta glasses look at the kitchen scale you already own, read the food and the grams off its display, and write the line. No app on the scale. No pairing. No hands.
- Works with Ray-Ban Meta
- Any kitchen scale
- Muse Spark reads the frame
- Muse Voice hears you
- Every entry stays on your phone
How it works
Three seconds. Zero typing.
Food apps make you type. Smart scales make you pick. itlogs.food reads the number the scale already shows, and asks only when it has to.
Say log it
Or press the button on the frame. Muse Spark names the food and reads the grams. A second look 1.5 s later checks the first.
Hear it back
218 grams of chicken breast, logged. When two reads disagree you get a question in your ear instead, and your answer is the label.
The log
A log that writes itself.
Every line carries the grams, the calories and protein from a nutrient table, and the label the loop produced: agreed, confirmed, or corrected. Those labels are tomorrow's eval set.
- Grams to the unit, straight off the display
- Kcal and protein from the grams, nothing else
- Apple Health on your phone, nowhere else
- 18:42grilled chicken breast218 g360 kcalagreed
- 18:43cooked white rice187 g243 kcalconfirmed
- 18:44olive oil9 g80 kcalagreed
- 18:45roasted pumpkin120 g53 kcalcorrected
- 18:47
dinner534 g736 kcal74 g protein
In the kitchen
Glasses on. Scale on. That's the setup.
“Every other health stream fills itself: the ring, the watch, the body scale. Food was the one log still typed by hand. The typing is what kills the habit.”
Pictures generated with Muse Image, Meta's image model.
Built on
Meta's models, in the loop.
Ray-Ban MetaThe glasses, through the Wearables Device Access Toolkit: camera stream, microphone, the button.
Muse SparkReads the frame: the food, the number on the display, and how sure it is. Structured output, every time.
Muse Voice TranscribeHears your answer when the two reads disagree. Realtime, endpointing, 24 kHz.
Muse ImageMade every picture on this page from a text prompt and one reference frame.
Muse GlimmerThe open-weight path: the same reader on a Mac, recalibrated on the labels the loop produces.
FastAPI + SwiftA thin phone client streams frames up a WebSocket; one Python loop does the rest. Open source.
Submission 2026-NOVEMBER-04.
The build is public.
Ten days, one kitchen, every commit in the open. Watch the repository to see the film land.