The numbers work offline
The factor tables are compiled into the app bundle. Once Drop is installed, an estimate never waits on a network — the API is a refresh path, not a prerequisite.
Photograph it. Read the water behind it.
Drop is a water-footprint tracker for Android. Photograph a meal or a product, confirm what it is, and read the litres behind it — with the uncertainty range around them and the source they were computed from.
Your water · today
1,724litres
This week1,724 of 16,000 L
- Ham & Cheese Sandwich300 g1,046 L
- Cakes & Croissants60 g98.4 L
- Plate · 2 items×2580 L
Anise · 2 g
22.3–103L
Matched to a single published study
rough estimate
Three taps from a photo to a number.
The whole loop fits between putting the plate down and picking the fork up.
01
Photograph
Point the camera at one item, a whole plate, or something you are about to buy. You can also search the catalog by name.
02
Confirm
Drop names what it sees. You correct it if it is wrong and set the amount. Nothing is counted until you say so.
03
Read
Litres, the range around them, and the source the figure was computed from — 862–1,501 L for a 300 g ham and cheese sandwich.
Confirmed items build a history on the phone. Drop works out what is actually driving your footprint and suggests one realistic step for next week — not a lecture, one change you would plausibly make.
Make your next choice a little better than your last one.
Better than last time, every time.
Drop 1.0.0, for Android.
One file, installed by hand. Everything you need to check it before you run it is on this sheet.
Drop-1.0.0-preview.apk
177MB
- Version
- 1.0.0 · preview channel
- Package
- rocks.ourplanet.drop
- Architectures
- arm64-v8a · armeabi-v7a · x86
- Permissions
- Camera. Nothing else is asked for.
- Catalog
- v2026.08.2, bundled — estimates run offline
SHA-256
6f322649f4d8cd3d0499da0ec90b3e77b296e73236ba8c467d5dd075e8511231
Four steps, once.
Open this page on the phone and tap Download the APK.
Android will warn you that this file type can harm your device. That warning appears for every APK, signed or not.When the phone asks, allow this browser to install unknown apps.
Settings → Apps → your browser → Install unknown apps. You can switch it back off afterwards.Open the downloaded file and tap Install.
Verify the SHA-256 first if you want to:sha256sum Drop-1.0.0-preview.apk.Launch Drop and grant the camera when it asks.
Photograph a plate. The first estimate lands in about thirty seconds.
Prefer to build it yourself? Clone the repo
The model recognises. It never calculates.
Ask a language model for a water footprint and it will hand you one — confidently, and often invented. So Drop keeps it away from the arithmetic entirely and gives it the job it is genuinely good at: identifying what is in the frame.
What the model does
- Reads the frame
- Names the thing
- Maps it to a catalog identifier
What the engine does
- Matches the versioned factor
- Applies the amount
- Returns litres, a range, and the source
Everything after recognition is a deterministic engine running on versioned data that ships inside the download. Same input, same output, every single time — with provenance and an uncertainty range attached to the answer rather than bolted on afterwards.
Catalog v2026.08.2as shipped in this build
1,000curated entries across food, drink, transport and everyday products
476automated tests hold the pipeline to all of it
The catalog grows slowly on purpose. An entry gets in when there is data good enough to stand behind it. Unbacked breadth teaches people to trust a number nobody earned, and that is the one failure this product cannot survive.
What runs on the phone.
Your history stays on the device
Confirmed items are written to a local SQLite database on the phone. Photographs go out for recognition and are not kept.
Every figure carries its range
Litres arrive with the uncertainty band around them and the dataset they were computed from, listed assumption by assumption underneath.
Download Drop
The water data already exists. Drop is the shortest path between it and your hand.
177 MB, one file, four steps. Photograph the next thing you were going to eat anyway.
Invisible water use, made visible, trustworthy and actionable — one choice, one habit, one drop at a time.
