Data and processing
How does TriElite handle training data and AI?
TriElite is designed around local-first training records. Plan generation, Coach, and food-photo estimates run on your device using Apple Intelligence and Apple Vision; prompts, training context, and photos are not sent to a cloud AI provider. Deterministic code validates proposals and calculations, while missing, estimated, and disconnected evidence remains explicitly identified.
The on-device AI features described here apply to version 1.2.4 and later. Installed versions 1.2.3 and earlier retain their optional remote AI services; see processing and retention for older versions.
How is information processed step by step?
- Training plans, activity records, Fuel entries, recovery context, settings, and Coach history are designed to remain primarily in app storage on the iPhone.
- Plan generation and Coach use Apple Intelligence FoundationModels on-device when available, with built-in deterministic behavior when Apple Intelligence cannot run.
- Food-photo estimates use Apple Vision and Apple Intelligence on-device. Photos are not sent to a cloud AI provider; barcode lookup and manual logging remain available.
- Apple Intelligence requires supported Apple hardware and software, a supported system language, regional availability, and the feature to be enabled.
Other services still process data for the features you use, including App Store and RevenueCat purchases, connected Strava, barcode lookup, and analytics or attribution where enabled. The privacy policy describes those boundaries.
Common mistakes about AI training tools
- Assuming AI output is an observed fact rather than a bounded suggestion.
- Assuming local-first means every operation is local or that device backups and third-party services are controlled by the app.
- Entering emergency information or highly sensitive medical details into a coaching or free-text route.
- Using an estimate as if it were measured evidence without checking its provenance.
AI may assist with strategy and wording, but deterministic validation owns safety, prerequisites, completion, provenance, and final calculations.
Why do people struggle to interpret incomplete data?
A training record can be missing distance, heart rate, power, sleep, nutrition, or recovery information. Treating each gap as a zero creates a false sense of precision, while treating every estimate as measured evidence hides uncertainty. TriElite keeps evidence states visible so that an athlete can decide what the record supports.
For the connected product model, see What is TriElite? and How planning works. The iPhone download page links to the public App Store listing.