He spent two hours turning $199-a-year WHOOP into an open-source version

Someone canceled their $199-a-year WHOOP subscription, then used Opus 5.5 to spend two hours building their own version.

The project is called Pulsr and it’s open source on GitHub. It runs on Apple Watch and iPhone, is free, has no subscription, no account, and data never leaves the phone.

First, some background. WHOOP is a subscription-based health tracking service: the hardware is free, and it makes money from the annual fee, $199 a year. Its core is three things: recovery, strain, and sleep. Every morning you open the app, look at a number from 0 to 100, and decide whether to push hard or rest today. Apple Watch itself has HealthKit and can measure heart rate, HRV, and sleep stages, but it doesn’t have WHOOP’s algorithms and interface.

Broken down, what Codez did isn’t complicated: take apart WHOOP’s interface logic, map it to the metrics Apple Health can provide, and rewrite it in Swift.

He included the full prompt in his tweet. The prompt itself is worth looking at, because the approach is more rigorous than many serious projects.

The first step isn’t writing code; it’s mapping. Have the model look through 14 WHOOP screenshots one by one, list each feature, mark which Apple Health data can feed it, and skip anything that can’t be measured.

Second, the engine comes before the interface. Write all scoring as pure Swift functions, without touching SwiftUI, run them against 60 days of simulated data, and only start building the interface after confirming the numbers look like a real person’s.

Third, design comes before code. Create 4 different visual directions, each with an iPhone interface and an Apple Watch face, put them on an HTML page to choose from, and only write the code after picking one.

Finally, validation: run every screen in the simulator in both light and dark modes, fix issues, then install on a real device.

Feature list:

• Recovery from 0 to 100, based on overnight HRV, resting heart rate, respiratory rate, and sleep performance; z-scored against a personal 30-day baseline, with the first 7 nights shrunk toward the population prior
• Strain from 0 to 21, calculated from minute-level heart rate using Banister TRIMP, with a target for the day
• How long you need to sleep tonight, plus a recommended bedtime
• Stress monitoring, health monitoring, and body-age estimation
• A journal that records the relationship between habits and recovery
• An on-device AI coach that doesn’t need an API key
• An Apple Watch app with light and dark modes

Two details in the algorithms.

Recovery uses ln(HRV), not raw HRV, because HRV is skewed. Green ≥ 67, yellow 34 to 66, red below 34. When there isn’t enough data in the first 7 nights, it shrinks toward the population prior, which means the numbers in the first week aren’t as accurate, and then they increasingly resemble you.

Sleep need = goal + yesterday’s strain + sleep debt. Sleep debt is only counted once and doesn’t snowball.

There’s also something called orb. Instead of a ring, it uses a blob of light flowing inside a glowing volume, like Siri, pulsing with resting heart rate and changing color with the recovery zone. It was first prototyped with a WebGL shader, then the same math was ported to SwiftUI’s Metal colorEffect. watchOS doesn’t have custom Metal, so it’s drawn with Canvas.

The community reaction is interesting. Some users asked whether they could install it without a programming background and whether they need a Mac. That question is left hanging, because the project is open source and installing it requires Xcode and a Mac; a free Apple ID works, but you have to re-sign every 7 days. Someone asked how far the recovery score is from WHOOP’s, and Codez said it’s very close. Others asked how to install it on an iPhone, and no one in the replies gave a beginner-friendly tutorial.

The really interesting part isn’t the two hours. It’s that WHOOP’s black box was broken down into readable formulas and written in docs/ALGORITHMS.md, where anyone can read and modify them. The moat of a subscription health service is half hardware and half algorithms. When the algorithms can be re-derived by a patient person in two hours, all that’s left is the hardware.

And the Apple Watch has long been on a lot of people’s wrists.