How to Get 10 Years of Apple Health Data Out of Your iPhone

If you’ve been wearing an Apple Watch for several years, there’s a substantial dataset in HealthKit — steps, heart rate, HRV, sleep, workouts, VO2 max, and more. Apple provides a way to export this as XML, and health4ai imports it all into Postgres on first launch. Here’s what’s in the export and how the import works.

The Apple Health Export

On your iPhone, go to Health → Profile photo (top right) → Export All Health Data. The export generates a zip file containing:

export.xml — the main file with all HealthKit samples
export_cda.xml — CDA-format clinical records (if you have any)
A workout-routes/ folder with GPX files for outdoor workouts

The export.xml file is large. Several years of Apple Watch data produces a file in the range of 500MB to several GB. It’s an XML document where each sample is a record element with attributes for type, value, unit, start/end date, and source device.

What’s in the Export

Every HealthKit record is included:

All quantity types (heart rate, steps, HRV, VO2 max, weight, etc.)
All category types (sleep analysis, mindful sessions, etc.)
All workout records with their metadata
Correlations (blood pressure — systolic and diastolic as paired records)
All sources and their device metadata

The date range goes back to the earliest data recorded — typically when you first set up your Apple Watch or Health app. If you’ve had an iPhone since 2015 and tracked steps via the pedometer, that’s in there.

health4ai’s Backfill Import

On first launch, the health4ai iOS app runs a full backfill of your HealthKit history. This doesn’t use the XML export — it queries HealthKit directly via the HKSampleQuery API to pull historical samples for each metric type. The results go into your Postgres database.

The backfill process:

Queries each registered metric type with no date limit (full history)
Writes batches to your Postgres healthkit_metrics table
Shows progress on the Home screen (sync status card)
Populates healthkit_daily_summaries for the historical data

For a dataset with 5+ years of Apple Watch data, the backfill might write 4-6 million rows. This takes a few minutes on a typical home WiFi connection to Supabase or Neon. The app runs the backfill in the background.

What the XML Export Is Useful For

The XML export is useful as a backup and for cases where you want to work with the raw data outside health4ai. If you’re building custom analysis in Python, the export gives you everything in a portable format.

For getting your data into health4ai, the iOS app’s built-in backfill is simpler — no manual export, no file handling. The backfill runs automatically.

Verifying the Backfill Completed

After the backfill completes, you can check the database to confirm the data range:

In the Supabase SQL Editor (or any Postgres client):

SELECT 
  metric_type,
  COUNT(*) as rows,
  MIN(started_at::date) as earliest,
  MAX(started_at::date) as latest
FROM healthkit_metrics
WHERE user_id = 'your_user_id'
GROUP BY metric_type
ORDER BY rows DESC
LIMIT 20;

You’ll see the row count and date range for each metric type. For steps, you should see data going back to whenever you first had an iPhone with a motion coprocessor. For HRV, back to when you first got an Apple Watch capable of measuring it.

In Claude Code, after backfill:

Give me a health summary for the last 30 days.

Then:

What's my VO2 max trend over the last 3 years?

The second query is the one that demonstrates the backfill working — you’re getting 3 years of historical data from your database, not just what the iOS app has seen since install.

Data Volume After Backfill

For reference, typical dataset sizes after backfill:

A user with 5 years of Apple Watch (Series 4+) might see:

~3M+ rows in healthkit_metrics (heart rate alone is high-volume — continuous sampling)
~500K rows of sleep, workout, and activity data
~150K rows in healthkit_daily_summaries (aggregated from the raw data)

The MCP server handles this volume efficiently because queries beyond 30 days route to the pre-aggregated summary table. A 5-year trend query scans ~1,800 rows in the summary table, not 3 million rows in the metrics table.

Historical Data You Might Not Know You Have

The export often contains data people didn’t know they were tracking:

Audiogram data — if you’ve taken a hearing test on AirPods
Handwashing — if you enabled handwashing detection on Apple Watch
Environmental sound levels — background noise data from Apple Watch
Cycle tracking — if enabled
Medication logging — if tracked in Health
Mental health scores — if you’ve done any Apple Health mental wellbeing assessments

All of this ends up in HealthKit. After backfill, it’s all queryable via query_metric with the appropriate HKIdentifier.

health4ai is free through July. Everyone in the founding batch gets lifetime access at $0.
Download on the App Store →