ATOM

A personal operating system for one body.

ATOM builds a continuously evolving, evidence-based model of its user — sleep, recovery, training, nutrition, symptoms, medication, habits and context — and reasons across them to turn personal data into clear decisions that say why.

Resting heart rate · against you, not the population Drag across the chart

the day's value your usual range (prior 28 days) outside your range

Standard in

Data enters through Apple Health, so any watch, ring or strap that writes there works. No single brand is the foundation. More signals make ATOM more precise; fewer still work, and ATOM says what is missing.

Its own numbers

Recovery, load and sleep are computed by ATOM from raw signals with a declared method: each signal against the person's previous 28 days (median and MAD). It works without HRV and states the basis it used.

Every claim has a source

Advice comes from guidelines, meta-analyses and position stands, cited every time. ATOM connects and explains; it does not diagnose, and it sends you to a doctor when that is the answer.

One gesture to tell it

A symptom, a photo of a meal, how you slept, a supplement: said in a moment, from anywhere. ATOM works out what it is and where it belongs in the model.

How it is built

Sensor: Native iOS app (SwiftUI, HealthKit background delivery, Core Motion). The phone collects; the cloud reasons.
Memory: Append-only event log on Postgres with hybrid retrieval (pgvector + full text). Nothing is forgotten; every insight keeps the context it was built from.
Reasoning: Language models plan, a validator checks every action against the person's constraints, and deterministic code computes the numbers.
Safety: A composable safety layer on every free-text input: a deterministic floor plus a semantic classifier, where silence is treated as uncertainty, not as safe.
Status: Private beta on TestFlight.

Building the model of a person, one signal at a time.

brandorossi@atom-health.com