Chat with your health data

The AI Chat connects your Apple Health data to the AI model of your choice. Every number in an answer was computed on your iPhone before the model ever saw it — the AI reads the results, it never does the arithmetic.

The trends the Apple Health app can't show you

Apple Health has been quietly building a longitudinal dataset about your body. Every night of sleep, every workout, every resting measurement adds a data point. The Health app can surface that data as a chart, but its charts cap at recent windows and keep each metric in its own silo.

The questions worth asking need the full timeline and multiple metrics at once: Did your resting heart rate trend down the year you started running, then creep back up? Does your HRV drop predictably every winter? Are your highest-effort training weeks actually followed by lower sleep quality?

Choose your model

50+ AI models. Your data, the AI of your choice.

Tau. bridges your data to the AI; it doesn't host a model of its own. You bring the key, so the cost and privacy terms are the provider's, not Tau's — and nothing you send ever passes through a Tau server, because there isn't one.

Free-tier keys
Mistral and Gemini (Google) currently offer a free API tier — terms are the provider's and can change. ChatGPT (OpenAI) and Claude (Anthropic) need a paid API key from the provider's console — a subscription isn't enough.
Three speeds, no model names to learn
Fast, Normal and Deep thinking — Tau. resolves each one against the provider's live model list at runtime, so a model the provider retires is replaced automatically — and a newer generation is picked up without an update.
Or pin one yourself
Expert mode turns off the automatic mapping and lets you choose any chat model your key gives you access to — that's where the fifty-plus figure comes from. Your pick is remembered per provider.
The chat loop

Tau. handles the data work between questions.

A normal AI chat only knows what you paste into it. Tau. opens every session with a map of your record: every metric you actually have, its unit, the first and last day it was recorded, the devices behind it, its all-time high and low, and the last seven days in full. From there the model asks Tau. for whatever a question needs — any other day, any window, any statistic, any correlation — and your iPhone computes it and hands back the result.

Ask
Ask in plain language — whichever you write in. Quick check or long investigation, same box.
Find
Nothing to attach. The model asks Tau. for the days, the windows and the statistics your question calls for, and your iPhone computes them on the spot.
Answer
Your chosen AI answers from those numbers, attributes them to Apple Health, and tells you which date or window it read.
Follow up
Ask for the underlying figures, a different period, or — in Clinical tone — a summary written for a physician.
Tone

Two tones. Two different answers.

The same question, answered for two different readers.

Everyday
A warm coach. It explains what a metric measures and why it matters before interpreting it, and reaches for an analogy when one helps.
Clinical
Precise terminology, unexplained; findings first, no preamble; interpretations framed explicitly as hypotheses, never as diagnoses.
Privacy

What never leaves your phone

Your API key lives in the iOS Keychain. The app has no server, no account and no analytics.

Reproductive health is a setting, and it is off until you turn it on. While it is off, nothing in that category reaches any model, even a paid one — periods, ovulation and pregnancy tests, contraception, cervical mucus, sexual activity, lactation. The metrics are kept out of the context entirely, and every tool refuses to return them: the context says plainly that the data is withheld by a setting rather than missing, so the model cannot mistake your silence for an empty record. Turn the setting on and the category becomes readable like any other, with a dedicated menstrual block (bleeding dates and cycle lengths) added so cycle questions can be answered. You can turn it back off at any time.

Questions you can ask AI about your health data

These aren't aspirational examples. Each one maps to data Tau. actually holds and statistics it actually computes on demand — your full daily history, the tested trends, the weekly and yearly cycles, the rank correlations between any two metrics.

Heart & Cardiovascular
  • "How has my resting heart rate trended over the past 3 months?"
  • "Compare my VO₂ max this spring to the same months last year."
Sleep
  • "Does time spent in daylight track with how much I sleep?"
  • "Do I sleep longer at weekends?"
Activity & Fitness
  • "Which days of the week am I most and least active?"
  • "How has my running speed changed over the last six months?"
Cross-Metric Analysis
  • "Based on everything you can see, which patterns actually cleared the significance threshold?"
  • "Do any of my metrics show a yearly cycle — something that shifts with the seasons?"