// the find
StanfordBDHG/HealthGPT
Query your Apple Health data with natural language 💬 🩺
An iOS app from Stanford that lets you ask natural language questions about your Apple Health data, backed by GPT-3.5/4, local Llama 3 inference, or a network fog node on your own hardware. Built on the Spezi health framework, so it's more of a reference implementation than a drop-in tool. Aimed at iOS developers who want a starting point for HealthKit + LLM work.
The three inference backends (cloud OpenAI, on-device Llama 3, local network fog node via mDNS) are genuinely well-thought-out — most projects like this just hardcode the cloud API and call it done. The Spezi foundation means HealthKit permissions, onboarding, and chat UI are already handled so you're not writing boilerplate from scratch. CI includes static analysis, link checking, and TestFlight deployment pipelines — the project hygiene is noticeably better than typical research code. The fog node setup (Docker + avahi mDNS) is a practical middle ground between on-device constraints and full cloud data exposure.
Only 6 HealthKit data types are supported out of the box (sleep, steps, active energy, exercise minutes, heart rate, body mass) — adding anything else requires touching three separate files manually, which is friction that should be data-driven. The context window approach — dumping 14 days of aggregated stats into a prompt — won't scale to richer queries or longer history without hitting token limits fast. There's no RAG or structured query layer; everything is prompt engineering, so complex questions about patterns over time will just get hallucinated answers. Last meaningful activity suggests this is more of a research artifact than actively maintained software.