Skip to content
aadya

Nothing leaves your device.

Cycle data is some of the most personal data there is. Aadya is designed so that there is nothing to collect: no server, no account, no tracking.

Privacy by architecture, not by promise.

A privacy policy can change. An architecture with no server cannot quietly start collecting data. The model is a file you download; it runs on your machine and answers locally.

  • No accounts. The only personal input an app needs is a name, stored locally.
  • No analytics, no telemetry, no third-party scripts, fonts or CDNs.
  • No network calls that carry user data.
  • Personalisation state is stored on the device and never uploaded.
Check it yourself

Do not trust, verify.

Go offline

Download the weights once. Then turn off your network and run the quickstart. It still works, because nothing in the model needs a connection.

Read the runtime

The released runtime is a few Python files you can read. It loads a weights file and returns an array of probabilities.

Watch this site

Open your browser’s network tab on any page here. Every request goes to this site’s own origin. No cookies are set.

The contract

Rules for anyone who builds with it.

If you embed Aadya in an app, these are the conditions the project asks you to keep.

  1. No network with user data. Do not send period dates, forecasts or names to a server.
  2. No analytics SDKs that phone home.
  3. Store locally (or in memory only), under the user’s control.
  4. Show uncertainty. Never a single date without its probability window.
  5. No fertility or contraception claims, and no diagnosis.
  6. State the evidence level. Simulated users plus small public cohorts.
  7. Respect the licence on the model card.
Safety

What Aadya is not.

Not a medical device

It does not diagnose, treat or screen for anything. If a cycle is unusual for you, talk to a clinician.

Not contraception or fertility planning

It forecasts when a period may start. It does not show “safe days”, an ovulation date or a fertile window.

Known weaker areas

Very irregular cycles, postpartum and perimenopause phases, hormonal contraception and very short histories are harder, and the forecast should be wider there. Forgotten logs hurt every method.

Evidence level

Results come from simulated users, two public real cohorts (364 people) and one small study for the optional ovulation-test input. This is not clinical validation.