Not born. Made.
OpenWOD is a self-hosted virtual training coach for home workouts, solo or in pairs. It executes structured sessions in six formats (timed circuits, AMRAP, EMOM, for time, you-go-I-go and strength sets) on a deterministic clock, announces every transition by voice, and maintains an extensible workout library. It runs as a static site straight from GitHub Pages, or as a single container on your own hardware, and requires no account, cloud service or telemetry.
OpenWOD plays a workout the way a coach would run it: it tells you what is next, counts you in, calls the boundaries, and otherwise stays out of the way. Pick a session from the library, press Start, and put the phone down.
| Format | What the clock does |
|---|---|
| Timed circuit | Work and rest slots on the clock, rounds and blocks with their own rests, A and B lanes that can alternate. |
| AMRAP | The whole block stays on screen with a round and rep counter. The timer never tells you what to do next, you do. |
| EMOM | Work starts on every boundary and never waits. An early Done shows the rest you earned. |
| For time | Counts up, ends on Done, with an optional cap. |
| You go, I go | Hands over on Switch and counts team rounds. |
| Strength | Rest on a timer between Done taps, set by set. |
- Proper CrossFit and HYROX timers with voice cues and beeps for work, rest and the countdown
- Train alone or with a partner, with the timer telling each of you who is up
- 54 ready-made workouts, from Fran and Murph to HYROX sessions and partner circuits, each with a difficulty rating
- Scaled loads for every weighted movement: light, medium and heavy, with women's weights alongside
- Find a workout that fits the time you have, the gear you own and how hard you want to go
- Write your own workouts or share them with friends as a single file
- Let your AI assistant write one: copy the guide into ChatGPT, Claude or any other assistant, describe the workout and paste its answer back
- See what you trained and when, and repeat a session with one tap
- Make it yours: pick one of five offline voices, show weights in kg or lb, tune the difficulty scale to your fitness, and choose your theme and accent colour
- Open it in the browser or add it to your phone's home screen, no install needed
- Optionally self-host it, so every device at home shares the same library and settings
- No account, no cloud, no tracking
Note
On the static site everything you build, import or log stays in that browser. The household server has no login yet, so keep a shared instance on your own network.
OpenWOD runs in one of two shapes built from the same code. The static site needs nothing but a browser and keeps the library in that browser. The container adds a household server, so every device in the home shares one library and one session log.
The current release is published at romanboegli.ch/OpenWOD. Open it, pick a workout and press Start. On a phone or tablet it installs as an app, see On a phone or tablet.
Everything you build, import or log is stored in that browser's storage, separately for each browser and browser profile. Download a backup now and then under Settings → General → This browser, and use Keep data on this device on the same card so the browser does not clear the library when space runs low. A backup restores into any OpenWOD, static or server.
To host the site yourself, download openwod-static-vX.Y.Z.zip from the latest release and unpack it into any web root.
Serve it over HTTPS (or localhost), because the screen wake lock and persistent storage require a secure context.
No rewrites or server configuration are needed.
Tip
Press Test sound once after opening the app. Browsers only play audio after a user gesture, and that tap unlocks the beeps and voices for the session.
The container serves the app and the household API on one port and stores its data in a SQLite database in the /data volume.
It needs Docker on any amd64 or arm64 host, for example a NAS.
Images are published to ghcr.io/romanboegli/openwod for every release.
curl -O https://raw.githubusercontent.com/RomanBoegli/OpenWOD/main/compose.yaml
APP_IMAGE=ghcr.io/romanboegli/openwod:v1.5.3 HTTP_PORT=8080 docker compose up -dOpen http://<host>:8080.
In Portainer, create a stack from this repository with the Compose path compose.yaml and set APP_IMAGE as a stack variable.
The settings, HTTPS with your own certificate, backup and restore, upgrades and rollback are described in docs/deploy.md.
Important
The household server has no login yet. Run it on your own network or behind an authenticating proxy, never on the public internet.
OpenWOD is a progressive web app (PWA), so either shape can be installed on a phone's or tablet's home screen. It then opens full screen under its own icon, without the browser bars, and the training view is laid out to fit the screen.
| Device | How to install |
|---|---|
| iPhone and iPad | Open the app in Safari, tap Share, then Add to Home Screen. |
| Android | Open the app in Chrome, open the menu, then Install app or Add to Home screen. |
Install it from an HTTPS address, so the screen stays awake during a workout. The installed app still loads from its address on every start, so it needs a connection to Pages or to your server and is not an offline copy. On iOS the installed app keeps its own storage, separate from Safari's, so a static-site library built in Safari does not appear in the installed app. Move it with Download backup in Safari and Restore from backup in the installed app.
Download a backup first if you want to keep the library, since every step below deletes it.
| Setup | Remove the app | Remove its data |
|---|---|---|
| Static site | Delete the home-screen icon, or the unpacked files on your web server. | Clear the site's data in the browser settings for romanboegli.ch or your own address. |
| Docker Compose | docker compose down in the folder holding compose.yaml, then docker image rm ghcr.io/romanboegli/openwod:<tag>. |
docker compose down -v, which also deletes the openwod_openwod-data volume. |
| Portainer | Remove the openwod stack. |
Delete the volume openwod_openwod-data under Volumes. |
A workout is a single YAML file with the suffix .wod.yaml.
The file is self-contained: prescriptions, targets, loads, repeats, rests and section rules all travel with it, and it never refers to another library.
The same file is used for the bundled workouts, for workouts built in the editor, and for workouts exchanged between installations.
A workout has one or more sections. Each section has a format, which defines how the clock behaves, and a body of nodes.
| Node | Purpose |
|---|---|
timed |
Work for a fixed duration, used in timed circuits. |
task |
Work that ends when the athletes tap Done, with targets such as reps, calories, distance or time. |
rest |
A timed rest. |
sequence |
Child nodes in order, with optional rests between them. |
repeat |
A body repeated a number of times, optionally swapping the A and B assignments every round. |
Each piece of work assigns a prescription to lane A, and in pair workouts to lane B.
A prescription names the movement and carries its targets, an optional load as one value or as light, medium and heavy levels with a women's amount, and free-text notes that are never executed.
Durations are written as 40s, 1m30s or 10m.
format: openwod.workout
formatVersion: 3
workout:
schemaVersion: 2
id: fran
revision: 1
title: Fran
participants: 1
prep: 10s
sections:
- id: fran-for-time
title: 21-15-9
format: forTime
transitionAfter: manualReady
cap: 10m
body:
id: fran-ladder
type: sequence
children:
- id: thrusters-21
type: task
assignments:
A:
name: Thrusters
reps: 21
load:
heavy: { value: 43, women: 29, unit: kg }
- id: pull-ups-21
type: task
assignments:
A: { name: Pull-ups, reps: 21 }Workouts are imported and exported in the library, one file at a time or as a .wods.yaml pack with collections.
Create with AI in the library hands you a guide to this format for any AI assistant, and takes the assistant's answer back into the import preview.
A sharing identity in each file lets an installation recognise a reimport and offer a newer revision as a replacement.
The complete format is specified in docs/formats.md and published as JSON Schemas under schemas/.
The bundled workouts, their sources and the load convention are documented in workouts/README.md.
| Tool | Version | Needed for |
|---|---|---|
| Node.js | 24 (pinned in .nvmrc), with npm |
Everything. The server uses the built-in node:sqlite module. |
| Playwright browsers | as installed by npx playwright install |
Browser tests. |
| Docker | with docker compose (Compose v2) |
Building and running the image. |
uv, or python3 with piper-tts==1.8.0, and ffmpeg |
Rendering the recorded voices locally. The Docker build and the release workflow render them on their own. |
git clone https://github.com/RomanBoegli/OpenWOD.git
cd OpenWOD
npm ci
npm run dev:full # app on http://127.0.0.1:5173, household API on port 8787
npm run dev:static # the static shape: no API, the library in IndexedDBdev:full keeps its data in .data/app.sqlite.
Run npm run cues:generate once to render the five voices into apps/web/public/cues, otherwise the app falls back to the device voice.
| Command | Purpose |
|---|---|
npm run check |
Typecheck, schema and workout bundle drift checks, unit tests. |
npm run test:e2e |
Browser tests in both shapes. Select engines with --project=firefox --project=webkit, or the static shape with --project=static-webkit. |
npm run build |
Production build of the web app and the server. |
npm run build:static |
The static site into dist/static, with relative paths. |
npm run preview:static |
Serves dist/static on port 4173. |
npm run test:production |
Smoke test against a running container or static preview, set by OPENWOD_PRODUCTION_URL and OPENWOD_PRODUCTION_MODE. |
npm run workouts:bundle |
Rebuilds the bundled workouts from workouts/. |
npm run schemas:generate |
Regenerates the published JSON Schemas from the TypeBox source. |
To build and run the container from source, use docker compose -f deploy/compose.dev.yaml up --build -d.
It binds to 127.0.0.1:8080 and keeps its own volume.
apps/
server/ Fastify: the household API over SQLite, migrations, health, the built app
web/ React 19 + Vite: library, editor, player, history, settings, the data layer under src/data/
packages/
domain/ Schema (TypeBox), compiler, difficulty, loads, movement catalogue, YAML exchange
engine/ Deterministic session engine
household/ Library rules (revisions, imports, sessions, export and restore) over SQLite or IndexedDB
deploy/ Compose file for building from source
docs/ Architecture, deployment, development, formats, validation, release notes
schemas/ Published JSON Schemas (generated)
scripts/ Bundling, schema generation, voice rendering, checks
tests/ Unit (Vitest), browser (Playwright) and production smoke tests
workouts/ The bundled workouts, one folder per collection
A pushed v* tag runs the Release image workflow.
It reuses the green Check of the tagged commit, builds and smoke-tests the amd64 and arm64 images, and publishes the multi-architecture image.
It also builds and smoke-tests the static site, creates the GitHub Release from docs/releases/<tag>.md with the static zip attached, and deploys the site to GitHub Pages.
The version bump, the one-time repository settings and the complete procedure are in docs/development.md.
- docs/architecture.md: design decisions behind the engine, the formats, the catalogue and the two deployment shapes.
- docs/development.md: repository layout, test strategy and the voice pipeline.
- docs/validation.md: what has been verified for each release.
- docs/plans/: designs of planned features.
To reset a working copy, run rm -rf node_modules .cache .data dist test-results and docker compose -f deploy/compose.dev.yaml down -v.
The code is licensed under the GNU GPL version 3. The bundled workout files are data under CC BY 4.0, each citing its source (see workouts/README.md). The benchmark workouts and station standards are those published by CrossFit and HYROX. CrossFit® is a registered trademark of CrossFit, LLC, and HYROX® of Upsolut Sports AG. OpenWOD is an independent project with no affiliation to or endorsement by either. The cue voices are rendered with Piper.
If OpenWOD is useful in your work, cite it via the Cite this repository button on GitHub (it reads CITATION.cff) or with:
@software{boegli_openwod_2026,
author = {B{\"o}gli, Roman},
title = {{OpenWOD}},
year = {2026},
url = {https://github.com/RomanBoegli/OpenWOD},
license = {GPL-3.0}
}