OpenHealth: stop hoarding your health data
Most of your health is decided by how you live, not by the clinic. OpenHealth turns tracker data and a light journal into personal, evidence-graded actions.
I wear a WHOOP. So does half the room around me. Almost none of us look past the morning recovery number. The band tracks HRV, heart rate, sleep, stages - data drips in for years. It produces exactly zero decisions.
That’s the trap. A blood panel is a snapshot of one morning. A tracker chart is a pretty line that changes nothing. Between two measurements, months of real life go by: what you ate, how you slept, how stressed you were, when you went to bed. Without a light journal, that stretch is a black box. And that black box is where everything actually happens.
Meanwhile most of your health isn’t decided in the clinic. In the classic estimate (Schroeder, NEJM 2007), behavior and lifestyle account for roughly 40% of outcomes, while medical care accounts for about 10%. Sleep, food, movement, light, stress - what you do every day outweighs the occasional doctor’s visit. It even holds against your genes: at high genetic risk for heart disease, a healthy lifestyle cut coronary risk by nearly half (Khera et al., NEJM 2016).
So I built OpenHealth - a personal, local-first health system on top of an AI agent. The interface is your Claude Code or Codex, not another dashboard to learn. Your data never leaves your machine.
Three layers, one loop
Data. Trackers give numbers: HRV, resting heart rate, sleep, strain. OpenHealth pulls them from Apple Health, WHOOP, Oura, Garmin - from where they already live, no manual entry.
Journal. This is the key link. Once a day, a light note about yesterday: sleep, food, training, light, how you felt. Minimal effort, like the single morning question WHOOP asks. The journal is what connects how you live to what the numbers show. And it isn’t a ritual for its own sake: people who log more consistently lose more weight - consistency beats perfection (Burke et al., 2011). Digital tracking of diet and activity has a measurable effect on its own (Obesity Reviews 2021, a systematic review of RCTs).
Action. Correlations between your journal and your recovery - on your own baseline, not a population - turn into 1-3 concrete steps for today. Each carries a C1-C5 confidence grade, from “this is an established fact” down to “raw hunch.” Not a diagnosis. A thinking partner, not a doctor.
Why personal, not “what works for everyone”
Averaged advice loses to personal observation. People’s responses to the same meal vary wildly: in the PREDICT study (Nature Medicine 2020), the spread in glucose and fat responses to identical meals ran into multiples, and directly measured genes explained only about 9% of the glucose difference. It’s less about genetics and more about each body simply reacting its own way.
HRV has no universal “normal” either: at rest, adults range from under 20 to over 70 ms. Comparing yourself to others is pointless - what matters is your own trend against your own baseline.
Hence n-of-1: change one variable, watch yourself, find the link. The most honest experiment on your health runs on a sample of one - you.
Data alone changes nothing
An archive of numbers doesn’t move your health an inch. The act does, and so does understanding the cause. Data changes behavior only when it answers you back, not when it piles up in logs (systematic review, IJBNPA 2023).
So OpenHealth doesn’t hand you another chart by default. It nudges gently but without letting you off the hook: “here’s one simple thing from the evidence-based basics - go do it. And what did you actually do for your health today?” Self-monitoring works not on its own but when it closes the loop: observe → goal → act → feedback.
Local-first and open
Everything stays local: raw data and keys go nowhere. The core is dependency-free Python, domain modules (sleep, recovery, cycle, labs, journal), C1-C5 confidence grades, and red-flag checks that stop interpretation and route you to a clinician when something looks off.
A desktop app is coming soon. The skills already work today, on plain files in your folder. The code is open: github.com/igindin/openhealth.
OpenHealth is a thinking partner, not a doctor. No diagnoses; medical decisions belong with a clinician, and anything alarming is a reason to stop and seek care. The figures cited from research are offered as a directional, population-level signal, not as clinical precision for any one person.