The Quantified Self Stack in 2026: Hardware, Apps, and AI

Quantified self has been a niche since the late 2000s — people tracking sleep, steps, and heart rate before it was built into every wrist. In 2026 the niche has expanded dramatically: Apple Watch has 100M+ active users, Oura is mainstream, and LLMs can now analyze health data in natural language. The bottleneck has shifted from “how do I collect this data?” to “how do I actually use it?”

Here’s the full stack as it exists today — hardware, software, and AI layers.

Hardware Layer

Apple Watch is the highest-volume health wearable by a wide margin. It writes to HealthKit automatically: steps, heart rate (continuous), HRV (during sleep and wakeup), resting HR (daily), VO2 max (from outdoor runs), blood oxygen (spot checks), sleep stages (watchOS 9+), workouts.

Oura Ring focuses on sleep and recovery. It writes sleep stage data to HealthKit (core, deep, REM, awake segments per night) and is often more accurate for sleep tracking than Apple Watch for users who find wrist-worn devices disruptive. The HealthKit integration means Oura data appears in the same database as Apple Watch data.

Garmin, Polar, Wahoo — sports-focused devices. Many write workout data and some write HR/HRV to HealthKit. Useful if you train seriously and want GPS + power data in the same pipeline.

Whoop — subscription recovery tracker. Doesn’t write to HealthKit (deliberate product decision), which means it’s an island unless you use third-party integration tools.

For the pipeline described here, any device that writes to HealthKit is automatically included. Devices that don’t (Whoop, Fitbit in some configurations) require separate handling.

HealthKit Layer

HealthKit is Apple’s on-device health data store. Everything from any app or device that has been granted HealthKit write access ends up here — unified, deduplicated, and accessible to any app with read permission.

The fundamental constraint: HealthKit is on-device only. There is no API. There is no cloud sync endpoint you can call. Your health data exists on your iPhone, and getting it anywhere else requires an iOS app with explicit user permission.

This constraint is the reason every solution in this space requires an iOS app component. There’s no workaround.

Database Layer

Getting HealthKit data into an AI requires persistence. You need a database where the data lives so queries aren’t limited by what the iOS app can return in real time.

The right choice is Postgres. It’s the default choice for structured data, it handles time-series queries well, and there are multiple free hosted options (Supabase, Neon) that make setup trivial.

The schema for HealthKit data is straightforward:

healthkit_metrics — raw samples (metric type, value, unit, timestamps, source device)
healthkit_daily_summaries — pre-aggregated daily values for historical queries

The sync mechanism matters more than the schema. An iOS app using HKObserverQuery receives push notifications from HealthKit when new data is written — Apple Watch records a new HRV reading, HealthKit fires, the app writes it to Postgres. This is real-time. An app using BGProcessingTask runs on iOS’s background scheduling, which can be hours delayed and will miss samples when the phone is idle.

AI Layer

With data in Postgres, the AI layer has two approaches:

Direct database queries — the AI writes SQL, runs it against your database, interprets the results. This works but requires the AI to know your schema, handle aggregation logic correctly, and manage the tiered raw/summary table distinction. Error-prone for non-trivial queries.

MCP tools — a local MCP server exposes pre-built functions (get_hrv_trend, get_coaching_brief, query_metric) that the AI calls by name. The business logic lives in the server. The AI calls the right function with the right parameters. This is cleaner and more reliable.

health4ai takes the MCP approach. The MCP server handles schema knowledge, aggregation logic, and the raw/summary tier routing. The AI calls tools and interprets results.

The Full Pipeline

Apple Watch / Oura → HealthKit (on-device)
                          ↓ (HKObserverQuery, push)
                 health4ai iOS app (background)
                          ↓ (TLS connection)
              Your Postgres database (Supabase / Neon / local)
                          ↑ (SQL queries via psycopg2)
           health4ai MCP server (local process on your Mac)
                          ↑ (MCP protocol, stdio)
              Claude Code / Cursor / Ollama via mcphost

Every component in this stack is either your hardware, your database, or software running on your machine. The only third-party infrastructure is your Postgres host and the AI inference provider.

What This Enables

Once the pipeline is running, questions that were previously unanswerable become routine:

“What’s my HRV trend over the last 3 months and is it improving?”
“Show me every day in the last year where my sleep was under 6 hours, then correlate that with HRV the following day.”
“How has my resting heart rate changed year-over-year since I started training consistently?”
“My travel schedule was heavy in April — show me what the physiological impact looks like in the data.”

These questions require years of data, multiple metric types, and the ability to correlate across them. A dashboard shows you charts. An AI with tool access can answer questions in plain language.

What’s Still Hard

Multi-device deduplication. If you wear both an Apple Watch and an Oura Ring, both write sleep stage data to HealthKit. Getting meaningful sleep analysis requires knowing which source to trust for which metric — the MCP server handles this for sleep by filtering to Oura data, but the general deduplication problem is real for users with multiple overlapping devices.

Context outside HealthKit. Health data in isolation is limited. Your HRV dipped in April — was it travel, training load, illness, life stress? The AI doesn’t know unless you tell it. The data answers “what” more reliably than “why.”

Wearables that don’t use HealthKit. If your primary wearable is Whoop or an older Fitbit, you’re outside this pipeline. Workarounds exist but add complexity.

For Apple Watch users — which is most of the market — the pipeline described here covers the full data set.

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