I'm Building a Sobriety App That Donates My Money If I Fail

I'm attempting sobriety myself right now. This app came out of that attempt.

There's one method that actually worked for me before. I bought a breathalyzer and made a deal with a friend: send a capture of every reading, and pay a fine if I didn't. It looks like a silly contraption, but it worked surprisingly well. Looking back, the reason it worked wasn't the precision of the device. There was money to lose, and someone watching. The thing behavioral economics calls a commitment device — I had duct-taped one together out of a breathalyzer and a friend.

Building that clumsy contraption properly is where this app started.

The structure: stake money, prove it daily

The structure is simple. You put up an amount as a stake and check in every day with a selfie. Keep the streak, and nothing happens. Fail, and the stake is automatically donated to charity. You don't just lose the money — it gets taken by a good cause. A loss that stings but that you can stand behind: that distinction mattered to me.

I'm building the mobile app with Expo and the backend with Fastify, with 313 tests and CI already in place. It's still in progress — I'm writing that down as plainly as everything else.

A design that never touches the money

The most important design decision in this app: the platform never holds the user's money. It's a non-custodial Stripe Connect flow (manual capture), so the stake is processed without ever passing through the platform's account.

Honestly, this wasn't the picture I drew first. It was the structure Claude Code — which I was consulting on the design — recommended most strongly. Once I heard it, it was obvious. The moment a solo developer holds other people's money, the weight that follows — trust, refund disputes, security — isn't the weight of one feature. If you never hold the money, most of those problems never exist. Even when the recommendation comes from an AI, a right recommendation is right, and I'm writing that down.

The weakest link: a selfie can be faked

This is the part of the post that has to be the most honest. The selfie check-in can be beaten. Someone else can take the photo for you, and I haven't found a technical way to stop someone determined to cheat.

Come to think of it, the breathalyzer-era "send a capture" was technically flimsy too. It still worked — probably because the person receiving the capture was a friend. I could have cheated, but the cost of lying to a friend was higher than the fine. The app has to fill that friend's seat, and the more I build, the clearer it gets that this isn't a technology problem. For now, all I can write is that I'm still wrestling with it.

One line I'm drawing

To be clear about one thing: a device like this can help with a casual commitment to drink less, but anything that looks like alcohol dependence is not an app's territory. Professional counseling and treatment come first.

For now, I'm user number one. Whether the contraption I once assembled out of a breathalyzer and a friend still works as an app — I'm finding out on myself first.