Record and analyze pump foiling sessions from your sports watch — GPS track, foiling distance, pump cadence and glide phases. Garmin, Wear OS, Apple Watch and Amazfit (Zepp OS), with native iOS/Android apps and a web PWA.
pumpfoil.org · License: AGPL-3.0
Pump foiling means riding a hydrofoil with no wind, waves or motor — you stay up purely by pumping the board rhythmically. Pumpfoil records these sessions on your watch and turns the raw GPS + accelerometer data into a detailed analysis: how long you were actually foiling, how efficiently you pumped, and where you rode.
The signature feature: your track is colored exactly over the foiling phases (by speed, heart rate or pump cadence) with every detected pump stroke marked right on the route.
Watch recording on every platform — a Connect IQ app for all ~78 Garmin
devices (fēnix, Forerunner, epix, Instinct, vívoactive, …), plus native Wear OS,
Apple Watch and Amazfit (Zepp OS) recorder apps. They capture GPS + heart rate (Garmin
also raw acceleration) and upload the raw data.
Native companion apps — full iOS and Android apps (sessions, map, track preview,
per-run stats, history with trend charts, community) alongside the installable web PWA.
Automatic analysis — foiling phases, distance, pump cadence and glide phases per session,
detected server-side from GPS + accelerometer (ML model + GPS state machine).
Color-coded track — foiling segments colored by speed / heart rate / pump frequency, with pump markers.
FIT upload — analyze old activities too: drop in a .fitfile (or Garmin's ZIP export).
Labeling UI — mark pump / glide / not-foiling ranges to build training data for the ML model.
Community & history — compare sessions over time, share runs; web UI in 7 languages.
One-click watch download — the site detects your watch model and serves the matching build.
| Track your progress over time | Community records |
 |
 |
| Community records per spot | Sessions — yours & everyone's |
 |
 |
| Detailed per-run stats | |
 |
Native apps on every platform — one account, one analysis backend.
📱 Android phone
📱 iOS phone
⌚ Wear OS
⌚ Apple Watch
⌚ Garmin (all ~78 Connect IQ devices)
⌚ Amazfit (Zepp OS)
| Directory | Stack | Purpose |
|---|---|---|
watch/ |
Monkey C (Connect IQ) | Garmin recorder (all ~78 devices): records GPS + raw accelerometer, uploads raw data |
android/ |
Kotlin · Jetpack Compose | One Gradle project: Android phone app (:app) + Wear OS recorder (:wear) |
watch-apple/ |
Swift · SwiftUI | iOS companion app (Sources-iOS/) with embedded Apple Watch recorder (Sources/) |
watch-zepp/ |
JavaScript · Zepp OS (@zeppos/zml) |
Amazfit recorder: records GPS + heart rate, uploads via the Zepp app |
server/ |
Python · FastAPI · PostgreSQL · numpy/scipy/scikit-learn | Ingest, immutable raw storage, foiling/pump detection, REST API |
web/ |
React · Vite · TypeScript · Tailwind · Leaflet | Installable PWA: sessions, map, charts, labeling, community |
deploy/ |
systemd · Apache | Service unit, reverse-proxy config, backup timers |
Raw session data is always stored complete and unmodified, so any future detection model can be re-run on old sessions. Detection runs server-side (fast iteration in Python, no watch recompile).
Server
cd server
python3 -m venv .venv && . .venv/bin/activate
pip install -e .
cp .env.example .env # dev defaults to SQLite — no Postgres required
uvicorn app.main:app --reload --port 8090
# API docs: http://localhost:8090/api/docsWeb
cd web
npm install
npm run dev # Vite dev server, proxies /api to :8090Garmin watch — requires the Garmin Connect IQ SDK
and a developer key (see watch/setup-sdk.sh). The SDK is not included (distributed by Garmin
under its own license). Build one device with watch/build.sh <device>, or all manifest devices
with watch/build-all.sh (also generates the download catalog the website serves).
Android phone + Wear OS — one Gradle project under android/ (modules :app and
:wear), Kotlin + Jetpack Compose. Build/verify with ./gradlew :app:compileDebugKotlin /
:wear:compileDebugKotlin (JDK 21).
iOS app + Apple Watch — watch-apple/, Swift + SwiftUI, generated with
xcodegen (project.yml); built in Xcode on macOS.
The raw upload format (GPS + int16 accelerometer chunks) is specified in
docs/data-format.md.
Pumpfoil is a community platform: by design, your sessions are visible to other users in the community feed. What it will never do is sell your data or hand it to third parties for advertising or tracking. This repository contains source code only — no user accounts, recordings, databases or Garmin SDK files.
Pumpfoil is free software, licensed under the GNU Affero General Public License v3.0 (AGPL-3.0). In short: you may use, study, modify and redistribute it, but if you run a modified version as a network service, you must make your source available to its users.
It is a community project — contributions are welcome.