AI health coach for your wearable data

Imagine your Oura readiness reads 71 most mornings since February. Your Apple Watch puts your VO2 max at 44, and it has been 44 for four months. The rings close green, the readiness band says “good,” and nothing in either app explains the plateau.

That’s the gap. The dashboard reports each number in its own panel: HRV on one screen, sleep stages on another, glucose in a third app that doesn’t know the first two exist. Five apps, five charts, zero instructions on what to do next.

The thing worth paying for isn’t more data. You already collect more than you read. It’s an AI health coach for your wearable data: something that reasons across the streams you’ve already got.

What an AI health coach for wearable data does that a dashboard does not

A dashboard plots a stream. A careful coach reads streams together, surfaces possible explanations, and suggests one thing to test. That’s the whole difference, and it’s a difference in behavior, not branding.

Three jobs a dashboard structurally can’t do. It can’t connect two systems you’d never think to line up, like glucose swings on a stressful day and lower HRV the next morning. It can’t test whether zone 2 work tracks your fitness more closely than HIIT. And it can’t put a recheck date on a suggestion.

A chart shows the line. A coach can suggest why it may have bent and how to test that idea.

There’s a second source the coach almost always has, and it’s free: time. A single reading is noise. The meaning is in the slope across weeks, and a coach reads the slope where a snapshot reads the dot.

Reading across Oura, Apple Watch and CGM at once

Here is the move a dashboard can’t make. A flat number becomes a better question when a second stream adds context.

Take that VO2 max stuck at 44. On its own it is a plateau. Line the training log up against your HRV and one hypothesis appears: every logged session ran in zone 4 to 5, and HRV stayed below baseline between them. If the training change is appropriate for you, try two zone 2 sessions a week while keeping hard days hard, then recheck VO2 max next month.

Now cross a CGM export with HRV, two signals most people read on separate screens. Your glucose swings widest, a 50 mg/dL range across the day, before the mornings when HRV reads lowest. Those days also line up with high stress and short sleep. The data cannot tell you which factor drove the change, but it gives you hypotheses to test one at a time. Start with the largest post-lunch swing, then watch whether next-morning HRV changes over the following two weeks.

One more, shorter. Your resting heart rate is up 6 bpm over six weeks. The climb lines up more closely with two or three weeknight drinks than with the training block. That association makes alcohol one factor to test: remove it for two weeks and see whether resting heart rate changes.

In each case the second stream sharpened the question and suggested a follow-up.

How the coach tests a hypothesis, not another chart

The mechanism is simple to describe. The coach holds your streams on one timeline and looks for one signal that moves when another moves, then right-sizes the claim to what the data can carry.

The right-sizing is where trust is won or lost. When the data is too thin to call a cause, the honest answer is “not yet, here’s what I’m watching,” not an invented reason. Two late dinners and one short night isn’t a pattern; it’s a coincidence you note and wait on. A coach that says “associated with” when it means a correlation, and waits for the pattern to repeat before it tells you to act, is a coach you can act on. The n-of-one signal is noisy by nature. Caution in the verb is the credibility.

And the reference that matters is yours, not the population’s. A 55 ms HRV night means nothing in the abstract. Read against your own 70 ms floor, it’s a real dip worth a question; read against someone else’s average, it’s a number from a stranger’s body. Your baseline is the only yardstick that fits you.

Why this isn’t a generic chatbot

The fair objection: “I can paste my Oura export into ChatGPT and ask.” You can, and it’ll summarize the file you pasted. That’s genuinely useful for one afternoon. It also stops there.

A general chatbot sees one snapshot, with no memory of last month and no second stream sitting beside it. It pattern-matches the text you handed it. It doesn’t watch your line, because it has no line to watch. Ask it again in March and you’re re-explaining yourself to something that has already forgotten you.

A hypothetical coach with access to live streams could carry their history and compare blood and wearable data over time. That is the case for continuous context rather than a one-off prompt. Any system offering fitness and wellness guidance needs sourced numbers, such as the actual 110 mg/dL from a sensor export, not a plausible-sounding average it filled in.

How Depth turns your streams into one read

Apple Health is Depth’s only current health-data connection. Depth reads user-authorized fitness and wellness records from Apple Health and can place supported records alongside your bloodwork. It does not directly pull from or continuously watch Oura, Apple Watch, WHOOP, CGMs, or other hardware.

If another app or device writes records to Apple Health, Depth may read supported records from Apple Health. That is not a direct Depth integration with the app, vendor, or device, and Depth does not communicate with, control, or certify that hardware. Supported Apple Health records can add day-to-day context to blood markers such as ApoB, hs-CRP, HbA1c, fasting insulin, ferritin, and Lp(a), producing observations and follow-up suggestions rather than diagnoses or confirmed causes.

The next step is small. Early access and the Founders Edition let you authorize supported Apple Health records and compare them with the data already in Depth.

Which brings us back to that VO2 max of 44, flat for four months. It is a question no single signal can answer. Read together with the training log and HRV, it becomes a hypothesis you can test rather than a cause the data has proved.