About Bein
Two apps. One model. No server.
Take anything here without asking. If you want a build, a demo, an interview, or detail on the on-device architecture, email dan@bein.ltd — it goes straight to the person who wrote the code.
The story, if you're short on time
Most "AI apps" are someone else's API.
A solo developer shipped two consumer apps that share a single 2.4 GB AI model on the device, so users only download it once. Neither app has a backend, an account system, or a server of any kind — the AI features run on the phone, and work in airplane mode.
Most "AI apps" are a thin client in front of somebody else's API. Bein's are the opposite: the model ships to the device and never phones home. The unusual part is what happens when you install both apps. Rather than downloading the same multi-gigabyte model twice, the second app detects the first one's copy and runs inference against it in place — over a signature-permission bound service on Android, and a shared App Group on iOS. Only apps signed by the same developer key can connect.
Devices too weak to run the model don't get a degraded cloud fallback — they get a deterministic rules engine, so the privacy guarantee holds on every device rather than only the expensive ones.
Boilerplate
Copy and paste.
Bein builds mobile apps for people who want to know what goes into their body — and own that record completely. Its apps run their AI on the device, have no backend and no user accounts, and are free. Bein is the work of one developer.
On the name
Bein is Being Intelligent. Read the logo the other way round and it's Intelligent Being — the words run in a circle, so neither is first. That's the whole product in two words: the apps make you better informed about your own body, and the intelligence doing it belongs to you, because it runs on your hardware.
It's also the parent case of how the apps are named — a verb you're doing, minus the g. Burning, cooking, being. Burnin, Cookin, Bein.
The apps
Side by side.
| Burnin | Cookin | |
|---|---|---|
| What it is | Adaptive calorie tracker | Recipe keeper |
| The hook | Measures your real metabolic rate from your own data instead of a formula | Turns TikToks, links and cookbook photos into structured recipes |
| Platforms | iOS, Android | iOS, Android |
| Price | Free (one banner, profile tab only) | Free, no ads |
| Account required | No | No |
| Works offline | Yes, including AI | Yes, including AI |
| Store | App Store · Play | App Store — in review · Play |
Technical facts worth checking
All of this is verifiable.
- Model: runs on LiteRT-LM, roughly 2.4 GB, downloaded once by the user.
- Android sharing: AIDL bound service guarded by a signature-level permission.
- iOS sharing: shared App Group between the two app sandboxes.
- Vision over the mesh: an app below its own vision capability floor can pass a file descriptor to the sibling and still log a meal from a photo.
- Fallback: a rules engine, not a cloud call, on incapable devices.
- Telemetry: anonymous product analytics only — no identity, no user content, no location. Nothing else leaves the phone.
- Built with: React Native (Expo) and TypeScript, with native Kotlin and Swift modules for the inference layer.
Press & contact
Ask for anything you need.
Assets
App icons, device-framed screenshots and a 30-second demo for each app. All cleared for editorial use.
Prefer something specific — a particular screen, a different aspect ratio, a longer demo? Ask and it'll be sent the same day.
Talk to us
Bein is one developer. Email dan@bein.ltd and it goes straight to the person who wrote the code — no press office in between.
Happy to go deep on running a multi-gigabyte model on-device in React Native, which was genuinely painful, or to hand over a build.