HealthLens

AI health analysis from your Apple Watch.
Private · On-device · No backend · No account.

       

中文文档

       
_{Home · Trends · Settings}

---

What it does

HealthLens reads your Apple Watch data from HealthKit and runs AI analysis entirely on your device — no server, no account, no data leaves your iCloud.

	Feature
🧠	AI Insight — Claude or GPT-4o summarizes your last 7 / 14 / 30 days in plain language
🔔	Smart Alerts — model-generated, not hand-written if/else rules
📈	Trend Charts — Sleep · HRV · Resting HR · Steps · Active Energy · Weight
☁️	iCloud Export — daily JSON syncs to Mac automatically
⏰	Shortcuts Automation — set once, runs every night
🔐	Private by design — API key in Keychain, data stays in your iCloud
💸	Free account — no $99/year Apple Developer subscription needed

---

Quick start

Requires Xcode 16+ · iPhone with Apple Watch data · API key from Anthropic or OpenAI

git clone https://github.com/whwhw/HealthLens.git
open HealthExport.xcodeproj

1. Signing & Capabilities — set your Team, change Bundle ID to com.yourname.healthlens
2. Connect iPhone → ▶ Run
3. Grant HealthKit permissions on first launch
4. Settings → AI Analysis → paste your API key
5. Tap Generate Insight

Free account note: Xcode re-signs every 7 days. Reconnect and hit Run again when the app expires.

---

iCloud Drive sync

Export daily JSON snapshots to iCloud Drive — they auto-sync to your Mac for deeper analysis or AI agent integration.

1. Settings → Export Storage → Change — pick any iCloud Drive folder
2. App stores a security-scoped bookmark (no iCloud Capability required)
3. Files land at ~/Library/Mobile Documents/com~apple~CloudDocs/<folder>/

→ See SPEC.md for the full JSON schema.

---

Use as an AI agent data source

Once data is on your Mac, any AI agent can read it and answer questions like:

"Should I train hard today?"   "Why have I been tired this week?"   "Best time for deep work this afternoon?"

Integration guide (Python · Claude Code · MCP)

How it works

iPhone HealthKit
  ↓  HealthLens exports daily JSON
iCloud Drive ──sync──▶ Mac ~/Library/Mobile Documents/.../
                             ↓
                        AI Agent reads last N days
                             ↓
              answers grounded in your actual body data

Step 1 — verify your sync path

ls ~/Library/Mobile\ Documents/com~apple~CloudDocs/HealthLens/ | tail -5
# 2026-05-10.json  2026-05-11.json  ...  2026-05-14.json

Step 2 — Python helper

import json
from datetime import date, timedelta
from pathlib import Path

HEALTH_DIR = Path.home() / "Library/Mobile Documents/com~apple~CloudDocs/HealthLens"

def read_health(days: int = 7) -> list[dict]:
    snapshots = []
    for i in range(days):
        d = date.today() - timedelta(days=i)
        p = HEALTH_DIR / f"{d}.json"
        if p.exists():
            snapshots.append(json.loads(p.read_text()))
    return snapshots

def health_summary(days: int = 7) -> str:
    data = read_health(days)
    if not data:
        return "No health data available."
    lines = []
    for s in data:
        parts = [s["date"]]
        if s.get("sleep", {}).get("asleep_minutes"):
            parts.append(f"sleep {s['sleep']['asleep_minutes']//60}h{s['sleep']['asleep_minutes']%60}m")
        if s.get("heart", {}).get("hrv_sdnn_ms"):
            parts.append(f"HRV {s['heart']['hrv_sdnn_ms']:.0f}ms")
        if s.get("heart", {}).get("resting_bpm"):
            parts.append(f"RHR {s['heart']['resting_bpm']}bpm")
        if s.get("activity", {}).get("steps"):
            parts.append(f"steps {s['activity']['steps']:,}")
        lines.append("  " + " · ".join(parts))
    return "Recent health data:\n" + "\n".join(lines)

Step 3 — inject into system prompt

system_prompt = f"""
You are a personal assistant with full context about the user.

{health_summary(days=7)}

Use this to inform advice about energy, focus, training, and recovery.
Interpret patterns — don't just repeat raw numbers.
"""

Works with any framework

Framework	How
Claude Code	Add read_health() as a tool in CLAUDE.md
MCP server	Expose read_health as an MCP resource
LangChain / LlamaIndex	Wrap as a Tool
Any local agent	Call health_summary() at conversation start

---

Architecture

Show architecture diagram

AppTabView
└── @EnvironmentObject: HealthStore · APIConfig · FolderStore · NotificationManager
    │
    ├── HomeView
    │   ├── HomeDataLoader   — metrics + mini-trends
    │   └── InsightGenerator — prompt builder + API call
    │
    ├── ChartsView
    │   └── ChartDataLoader  — 6 time-series charts
    │
    ├── ContentView (Export tab)
    │   └── Orchestrator     — date range → JSON pipeline
    │
    └── SettingsView

HealthStore is created once at the root and passed as a method parameter to all data-loading classes — no hidden singletons.

Zero third-party dependencies — SwiftUI · HealthKit · Charts · UserNotifications · AppIntents · Security

---

AI providers

Provider	Default model	Note
Anthropic	claude-opus-4-7	Recommended
OpenAI	gpt-4o
Custom	any	OpenAI-compatible base URL — works with DeepSeek, Ollama, etc.

---

Why I built this

Apple Health has the data. What it lacks is interpretation.

I wanted something that looks at 7 days of my sleep + HRV + steps and tells me what it means — not a dashboard of numbers, but a paragraph that says "your recovery is solid but sleep consistency is slipping."

Also wanted to prove a non-iOS developer can ship a real iOS app in a weekend with AI coding tools. Built in 3 days with zero prior Swift experience, using Claude Code to write all the Swift.

---

Contributing

See CONTRIBUTING.md — one concern per PR, test on real device, keep zero-dependency rule.

License

MIT