HealthGPT is an experimental iOS application that enables users to query their Apple Health data using natural language. Developed by the Stanford Biodesign Digital Health team, it leverages the Stanford Spezi framework to provide an extensible platform for LLM-powered health applications, targeting researchers and developers interested in integrating health data with AI.

How It Works

HealthGPT utilizes the Stanford Spezi framework, a modular system for building digital health applications. It integrates with Apple HealthKit for data access and supports both cloud-based LLMs (GPT-3.5/GPT-4 via SpeziLLM) and local LLM execution (Llama3 8B via SpeziLLMLocal) for enhanced privacy. The app features a chat interface with speech-to-text and text-to-speech capabilities, allowing for conversational interaction and data export.

Quick Start & Requirements

• Install: Clone the repository and open HealthGPT.xcodeproj in Xcode 16.2+.
• Prerequisites: macOS, Xcode 16.2+, iOS device (for local LLM).
• Notes: Requires manual data entry in the simulator. Local LLM execution requires a physical iOS device with Metal GPU support.
• Docs: HealthGPT

Highlighted Details

• Extensible architecture built on Stanford Spezi.
• Chat interface with speech-to-text and text-to-speech.
• Supports OpenAI GPT-3.5/GPT-4 and local Llama3 8B execution.
• Out-of-the-box support for sleep, steps, active energy, exercise minutes, heart rate, and body mass.

Maintenance & Community

• Open-source project of the Stanford Biodesign Digital Health team.
• Contributions are welcome; refer to contribution guidelines.

Licensing & Compatibility

• MIT License.
• Compatible with commercial use and closed-source linking.

Limitations & Caveats

The application is experimental and LLM outputs may contain inaccuracies ("hallucinations"). Aggregated HealthKit data for the past 14 days is uploaded to OpenAI when using their models.