What is openGym?
openGym is a workout tracker you can run on your own server with Docker or use on the phone alone. It plans routines over a library of 5,632 exercises with animated demos, runs guided sessions with pre-filled weights and rest timers, applies progression rules such as double progression or Greyskull LP, and charts estimated one-rep maxes, body weight and muscle coverage. There are apps for iPhone with Apple Watch, iPad and Android, plus a web app, in 26 languages, and imports from Strong, Hevy and FitNotes.
AI is an optional extra rather than the core. An AI coach drafts and revises routines from your log using your own provider key, and an MCP server lets an assistant such as Claude Desktop read your training history and log a workout you describe. Both are off by default, and the README states that nothing in a default install calls an AI service.
The repository is AGPL-3.0, has about 22,700 GitHub stars, 3,017 forks and 348 open issues, and is at v1.4.5 (14 November 2026), with releases every one to two weeks.
Is openGym worth it?
Judged as an AI project, openGym is mostly not one. You can plan, log, sync and chart training for years without any model involved; the coach and MCP server exist only if an admin switches them on. That is the right order for an app holding body weight, measurements and heart rate.
The coach is well fenced. It works with Anthropic, OpenAI, Gemini, MiniMax or any OpenAI-compatible endpoint, Ollama included, and the payload is built from an allowlist: plan, logged sets, weigh-ins, intake answers and preferences, capped at 12 weeks or 60 sessions. Health check-ins, sleep, resting heart rate, HRV and blood pressure go only if you flip a separate switch that is off by default. A consent sheet names the provider and lists the categories before anything leaves, and a model on your own network skips it because nothing reaches a third party. Proposed changes come back as cards you approve.
Health data handling is plain and worth knowing. A self-hosted instance stores everything as JSON under ./data, unencrypted, so anyone who can read that folder or its backups can read every user's workout history and body-weight log, and admins can see users' history by design. On the phone alone, data stays on the device; Apple Health and Health Connect sync in both directions, and the project says there is no telemetry. Passkeys are the default sign-in.
The project is maintained by one person alongside school, with no SLA for security fixes, and the exercise animations are licensed separately from the AGPL code. For a self-hoster who wants Hevy-style tracking without a subscription, those are acceptable trade-offs.
Where it earns its place
A home-lab hobbyist who trains four days a week
Move my training history off a subscription app onto my own server
Run docker compose up, import a Strong or Hevy CSV, and sign in with a passkey from the iPhone over HTTPS. Cost is the existing server; there is no subscription, and backing up ./data backs up everything.
A lifter running Ollama on a home server
Get a weekly review of my programme without sending health data to a cloud AI
An admin points the coach at the Ollama endpoint; the docs report about a minute per review for qwen2.5:3b on a 4-core box with no GPU. Requests to a local network address skip the third-party consent prompt because nothing leaves the network.
A Claude Desktop user who forgets to log sets at the gym
Ask my assistant about my progress and log sessions in plain words
Add the stdio MCP server pointed at ./data; 11 read tools answer questions such as estimated bench 1RM using the same functions as the app, and log_workout writes a session after pairing with the server.
Before you commit
- Setup
- Clone, copy .env.example, docker compose pull and up, then open localhost:8080 and create a profile. Phone sign-in with passkeys needs HTTPS on a domain; the guides cover Cloudflare Tunnel, Caddy, Traefik, nginx and Tailscale. The iPhone app is on the App Store and Android ships as a signed APK.
- Integrations
- Apple Health and Health Connect both ways, imports from Strong, Hevy and FitNotes, signed webhooks to Home Assistant, n8n or ntfy, an OpenAPI-documented HTTP API, a stdio MCP server and an optional remote MCP container using scoped tokens.
- Data and privacy
- Self-hosted data is JSON in ./data, not encrypted at rest; only the coach's provider key is encrypted. Phone-only use keeps data on the device. No telemetry. Coach requests carry allowlisted categories to your chosen provider; health metrics only with a separate opt-in.
- Cost at scale
- No paid tier. A household or small gym pays for its own hosting plus model tokens if the coach is on; the admin card sets daily limits on what a shared API key can spend.
- Limits
- 348 open issues; one maintainer; rate limiting only on password sign-in and device codes, the rest left to your reverse proxy; Google Play still in closed testing; Wear OS in beta.
What’s good
- Full tracker works with no AI and no account
- AI coach payload is an allowlist; health data off by default
- Coach runs on your own key or a local model
- Phone, watch and self-hosted server sync with field-level merges
What’s not
- Data in ./data is not encrypted at rest
- One maintainer; no security response SLA
- Exercise media not covered by the open-source licence
- Android APKs are large, from about 315 MB
Best for
Lifters who want a full-featured, self-hostable training log they control, with an AI coach only if they choose to wire one up.
openGym: common questions
- Is openGym worth it?
- openGym is a serious workout tracker first, and its AI coach is the rare optional extra that tells you exactly which of your health data it sends and asks before it does. We rate it 4.0 out of 5.
- How much does openGym cost?
- Free and open source (AGPL-3.0), with no paid tier. The AI coach uses your own provider key or a local model. ElfHosted offers managed hosting with a $1, 7-day trial.
- What are the downsides of openGym?
- Data in ./data is not encrypted at rest; one maintainer; no security response SLA; exercise media not covered by the open-source licence; android APKs are large, from about 315 MB.
- Who is openGym best for?
- Lifters who want a full-featured, self-hostable training log they control, with an AI coach only if they choose to wire one up.
- How much does openGym cost for a team or in production?
- No paid tier. A household or small gym pays for its own hosting plus model tokens if the coach is on; the admin card sets daily limits on what a shared API key can spend.
- Is openGym safe for company data?
- Self-hosted data is JSON in ./data, not encrypted at rest; only the coach's provider key is encrypted. Phone-only use keeps data on the device. No telemetry. Coach requests carry allowlisted categories to your chosen provider; health metrics only with a separate opt-in.
Answers checked Nov 18, 2026.