GenieMax

Free, open health-analytics engine — recovery, sleep staging, HRV, strain/load, nap detection,
calorie tracking, and an end-to-end-encrypted sync vault. Built for the community and for learning,
so the science of recovery and sleep isn't locked behind a $200+ device and a forever-subscription.

Recovery, sleep, and HRV insights are some of the most useful things a wearable can give you — and they're almost always trapped behind expensive hardware plus a monthly subscription, in a cloud you don't control. The math behind those numbers is well-published science. So why should seeing your own data cost you forever?

GenieMax is that math, free and open, built to give away — for anyone who wants to learn how recovery/sleep analytics actually work, run them on their own data, and own the result. No subscription, no lock-in, your data stays on your device.

A companion iOS app (the screens below) is in public beta on TestFlight: → Join the beta on TestFlight

The frame decoding currently targets the WHOOP 5.0 / WHOOP MG data format only — other devices, and an Android client, are on the roadmap.

It started as a simple frustration: a perfectly good fitness band on my wrist, streaming rich biometric data every second — and I couldn't see my own numbers without paying every month, forever. The data was mine. The science was public. Only the software stood in the way.

So I set out to close that gap, and it turned out to be a genuinely hard, fun problem:

1. Reverse-engineering the data. The band streams its sensor data in an undocumented binary format over Bluetooth. There's no spec — just bytes. Figuring out which bytes meant what took capturing real streams, diffing frames, and a lot of patient guess-and-check to line decoded numbers up against ground truth.
2. Binding raw bytes to real signals. Once the frames were cracked, each field had to be mapped to a meaningful biometric — and the scaling/units verified against the official app's values until they matched.
3. Turning signals into insight. Raw HR and motion aren't "recovery" or "a sleep score". That's a second layer: the published sports-science models, implemented and pinned with golden-vector tests so the numbers are trustworthy and reproducible.

Nothing here is a black box — a few examples (the full set of formulas is here →):

$$\mathrm{RMSSD}=\sqrt{\frac{1}{N-1}\sum_{i=1}^{N-1}\bigl(RR_{i+1}-RR_i\bigr)^2}$$

$$\textbf{Day strain}=21\left(1-e^{-\mathrm{TRIMP}/\tau}\right)\qquad \textbf{Recovery}=100,\Phi!\left(0.55,z_{HRV}-0.20,z_{RHR}-0.10,z_{RR}+0.15,z_{Sleep}\right)$$

Every formula — HRV, zones, strain, training load (CTL/ATL/TSB), recovery, sleep, calories, fitness age — is written out and matched by a unit test. See the full breakdown → docs/HOW-IT-WORKS.md

Real screens from the GenieMax iOS app (personal identifiers removed). This repository is the analytics engine behind them.

Today — recovery, strain & vitals	Calories — burn, Move & zones	Activities — training log	Training habits
Fitness — VO₂max, rhythm, load	Journal — the coach learns	Lock-screen Live Activity	…and more

What the engine enables: 🟢 Recovery, strain & readiness · 😴 Sleep staging + hypnogram, SRI, sleep debt · ❤️ Full biometrics (HR, RHR, HRV RMSSD & SDNN, respiratory, skin temp, stress) · 🫀 Cardio age / VO₂max · 🔥 Automatic calorie burn · 📸 Snap a meal photo → AI logs calories & macros · ⏱️ Workout & interval timers · 📲 Lock-screen Live Activity & Dynamic Island · 💬 AI coach grounded in your data · 🔒 End-to-end-encrypted sync.

flowchart LR
  subgraph dev["Your device"]
    S["Sensor samples<br/>HR · HRV · motion · resp · skin-temp"] --> WC["GenieMax<br/>(this package)"]
    WC --> M["Recovery · Sleep stages<br/>Strain/Load · HRV · Calories"]
    WC --> V["E2EE Vault<br/>Argon2id + XChaCha20-Poly1305"]
  end
  V -->|ciphertext only| B[("Zero-knowledge<br/>Cloudflare Worker")]
  B -.->|ciphertext| V2["Vault on another device"]
  V2 --> WC2["GenieMax"] --> M2["Same metrics, decrypted locally"]

Companion: backend/ — a zero-knowledge Cloudflare Worker that stores only end-to-end-encrypted blobs (the server never sees plaintext). Deploy your own; see backend/README.md.

// Package.swift
.package(url: "https://github.com/satayutata/geniemax-core", from: "1.0.0")
// target deps: .product(name: "GenieMax", package: "geniemax-core")

git clone https://github.com/satayutata/geniemax-core && cd geniemax-core
swift test          # runs the full golden-vector suite

import GenieMax

let samples: [SleepSample] = …            // per-minute (ts, hr, hrv, motion, respiratory, skinTemp)
let sleep = SleepStaging.stage(samples)   // stages, TST, efficiency, hypnogram
print(sleep.tst, sleep.deep, sleep.rem, sleep.light)

Validated by golden-vector tests — recorded input → expected output — so a refactor that changes a number fails CI. Run swift test. Fixtures use time-shifted, de-identified sample data (no real dates, no personal identifiers).

• No connection/transport code is included — this package only interprets data you already have.
• No secrets, no personal data: no API keys, tokens, accounts, or real health records in this repo.
• Rhythm screening is wellness/experimental and non-diagnostic — not a medical device.
• Independent project — not affiliated with, endorsed by, or connected to WHOOP (a trademark of its owner).

• Goose — the open-source companion project (for the same wearable) whose approach and public reverse-engineering this work studied and built on.
• Visual/UX design inspired by modern health-dashboard styles — no brand affiliation.
• swift-sodium / libsodium — the crypto behind the vault.
• BIP-39 — the standard English word list used for recovery phrases.
• Public HRV / sleep research, incl. heart-rate-volatility work separating true sleep from quiet wake.

See THIRD-PARTY-NOTICES.md for bundled-dependency licenses.

Issues and PRs welcome — see CONTRIBUTING.md. Parity with the golden vectors is the bar: keep swift test green. Security reports: SECURITY.md.

MIT © 2026 GenieMax Contributors — free to use, learn from, and share. Third-party components keep their own licenses.