On‑device Apple Health AI
- Andres Trevino shared an iOS app that analyzes 90 days of Apple Health data locally using Claude AI for pattern detection. - The app looks for correlations across 20 metrics while processing everything on device to preserve privacy. - The prototype is presented as a privacy‑first approach to symptom tracking and personal pattern detection, offering a design reference for local inference (x.com).
Apple’s HealthKit keeps health data in a central store on iPhone and Apple Watch, and apps can read it only after users grant permission for each data type. Andres Trevino’s new iOS prototype applies that model to artificial intelligence by analyzing Apple Health data on the device instead of sending it to a server. (developer.apple.com) (x.com) Trevino said the app reviews 90 days of Apple Health history and looks for patterns across 20 metrics, including symptom and activity signals, using Claude for local analysis. He presented it as a prototype for symptom tracking and personal pattern detection rather than a diagnostic tool. (x.com) Apple’s rules for HealthKit already require explicit permission to read and write health data, and users can allow one category while blocking another. Apple also says apps using HealthKit cannot use that data for advertising and cannot disclose it to third parties without express user permission. (developer.apple.com) Apple says HealthKit data is stored locally on the device and encrypted when the phone is locked, and its privacy white paper says the company tries to process as much health data on-device as possible. Trevino’s prototype follows that same privacy-first architecture at a moment when artificial intelligence companies are pushing deeper into consumer health. (developer.apple.com) (apple.com) That timing is specific. Anthropic began rolling out Apple Health integration in the Claude iPhone app in January 2026 for U.S. Pro and Max subscribers, letting users opt in to share movement, sleep, and activity data with the chatbot. (macrumors.com) Anthropic said that cloud-connected feature can summarize medical history, explain test results, detect patterns across fitness metrics, and help users prepare questions for doctor appointments. The company also said health data from that feature is not used to train models. (macrumors.com) The difference in Trevino’s demo is where the computation happens. Instead of exporting health records to a remote assistant, the prototype keeps the analysis inside the phone, which matches a broader push among Apple developers to treat local inference as a way to reduce exposure of sensitive data. (x.com) (apple.com) That approach also answers a practical problem with Apple Health itself: the app collects and stores large volumes of data, but the default interface mostly shows snapshots and trend cards rather than cross-metric explanations over time. Developers and users have started building separate tools to turn those exports into dashboards and pattern reports. (apple.com) (medium.com) Trevino’s prototype does not change Apple’s permissions system or Claude’s medical limits. It shows what a privacy-first health assistant could look like when the phone already holds the data, the operating system already gates access, and the artificial intelligence layer stays on the device. (developer.apple.com) (x.com)
