pulsebase
Quantified self, without the walled garden

Your health data, as files you actually own.

Pulsebase turns your Apple Health history into open files on your device, syncs them to a cloud you control, and lets you ask real questions about your training.

See how it works How your data is handled

Built for iPhone · Your health data never touches our servers.

A look inside

Five screens, one local dataset

Every number here is read from files on the phone. Nothing is fetched from a Pulsebase server, because there isn't one.

The Today screen: cloud sync listing Dropbox, OneDrive and Google Drive all up to date, a readiness score of 75 with an explanation of the underlying HRV and resting heart rate, and a coach's note below it.
TodayReadiness, a coach's note, and where your data has synced — in one glance.
The Charts screen: the question "Compare my workout minutes with my resting heart rate rolling 6 months" has produced a dual-axis chart of weekly workout minutes against average resting heart rate.
ChartsDescribe what you want to see. The AI writes the query; your phone runs it.
The Coach screen: a chat where the reply recommends a 45 to 50 minute aerobic base run, citing the user's seven-day HRV, resting heart rate and workload ratio.
CoachAnswers that cite your own numbers, not generic advice.
The Activities list: runs and walks named by the places they happened, each showing distance, duration, temperature, the shoes worn, and a trophy on personal bests.
ActivitiesEvery session named by place, with the shoes and the weather.
The route planner: a 9.75 kilometre route drawn on a map with numbered waypoints, and an elevation profile showing 269 metres of ascent below it.
PlanDraw a route, snapped to real roads and paths, then share it as a file.

Scroll sideways for the rest →

What it does

Your data, working for you

A companion for people who want their data as data — not locked inside another app's servers.

Your data as open files

Workouts and health metrics export to open, portable files on your device — readable by any tool, never trapped in a proprietary format.

Sync to your own cloud

Mirror your dataset to your own iCloud, Google Drive, Dropbox, or OneDrive. It goes to your account — Pulsebase runs no servers of its own.

Ask anything

An on-device query engine and an optional AI coach let you ask real questions about your training and get answers, charts, and summaries.

Routes & workouts

Review runs with location, weather, and per-shoe mileage, and plan new routes — all built on the same data you already own.

How it works

From Apple Health to answers

Three steps, and your data stays in your hands the whole way.

1

Connect

Grant read access to Apple Health. Pulsebase exports your history to open files, right on your device.

2

Sync

Mirror those files to your own iCloud, Google Drive, Dropbox, or OneDrive over official sign-in. Nothing routes through us.

3

Ask

Explore with charts and an AI coach that answers real questions about your training — on your terms.

AI, on your terms

A built-in AI coach — powered by your own key

Pulsebase has an AI layer that reads your data and answers real questions. It runs on your API key, so you're in control of it — and it's completely optional.

Bring your own key

The AI features use Google Gemini. You paste your own Gemini API key once in Settings → AI; it's stored in the iPhone's Keychain and used to talk to Google directly. Getting a key is free from Google AI Studio. No key means the AI features simply stay off — everything else in the app works exactly the same.

Google Gemini Free key · AI Studio Stored in your Keychain Fully optional

Today's readiness

A morning readiness score with the reasoning behind it — built from your recovery signals: HRV, resting heart rate, sleep, and recent training load.

Coach chat

Ask in plain language — "how's my training trending?", "am I recovered?". The coach queries your data on-device and answers from the results.

Per-workout debrief

Every workout gets a short written summary — pace, heart-rate zones, splits, and how the session fits into your recent training.

AI charts

Describe the chart you want; the AI writes the query and chart spec, your device fills in the numbers. Save it and it refreshes with new data forever.

What actually leaves your device

Only summaries — never your raw recordsThe coach and readiness analysis do the number-crunching on your device first, then send Google just the small summaries they need (like a 7-day average heart rate) plus your question.
Charts send no data at allFor AI charts, Google only writes the chart's instructions. Your device supplies the actual numbers — so none of your health data is sent.
Google is the only outside partyRequests go straight from your phone to Google over an encrypted connection, authorised with your own key. Pulsebase runs no server in between.

AI is entirely opt-in — no key, no AI, and every other feature keeps working. See the privacy policy for the full detail.

For the data nerds

Not an export — a queryable dataset

Pulsebase writes your history as Apache Parquet: columnar, partitioned, and self-describing. Any tool — DuckDB, pandas, or an AI CLI — can analyze it with zero setup.

Apache Parquet, Snappy-compressedTyped, columnar, and tiny — years of samples in a few megabytes. Hive-partitioned by type and year, so queries scan only what they need.
Full resolution — nothing thrown awayOne row per Apple Health sample, never downsampled. Per-workout series keep native resolution: beat-to-beat heart rate, per-second speed, and every GPS trackpoint.
Self-describingEvery sync writes a README.md data dictionary, an AGENTS.md guide for AI agents, and a _manifest.json — all generated from the real schema, so the docs can't drift.
Documented conventionsTimestamps in UTC with a timezone offset, stable join keys like workout_id, and an append-only + tombstone model the docs spell out for correct reads.
.parquet DuckDB-ready Hive-partitioned Snappy README.md AGENTS.md
zsh — ask anything about your own data 
$ cd ~/Dropbox/HealthData
$ claude "my longest run this year — distance and average heart rate?"

  Read AGENTS.md → this folder is a DuckDB-ready Parquet dataset.
  Longest run: 21.4 km on 2026-05-18 · avg HR 156 bpm · 1h47m.

# Prefer SQL? It's just Parquet — point DuckDB straight at the files:
$ duckdb -c "SELECT activity_label, round(max(total_distance_m)/1000, 1) AS km
           FROM 'workouts/*.parquet' GROUP BY 1 ORDER BY km DESC LIMIT 3"

Because AGENTS.md and README.md ship inside the folder, an AI CLI knows your schema the moment you point it there — no explaining, no setup, no data leaving your machine.

Privacy by design

Your data stays yours

Pulsebase is built the opposite way to most health apps: there's no backend collecting your data, because there's no backend at all.

Read the full privacy policy →

No serversPulsebase runs no backend. Your health data never touches infrastructure we operate.
Your cloud, your accountSync goes straight to storage you own, authorised through the provider's official sign-in.
On-device by defaultYour files and queries live on your device. AI features send only what's needed, only when you ask.
No trackingNo advertising SDKs, no data selling, no profiling. Ever.

Your data. Your files. Your call.

Pulsebase is a solo-built, quantified-self companion for people who want ownership — not another walled garden.

Read the privacy policy

Coming soon to iPhone.