GymCoach 1.0

GymCoach 1.0 is the first tagged release of the self-hosted workout tracker with an AI coach that runs on your own LLM key.

Self-host in one minute

mkdir gymcoach && cd gymcoach
curl -fsSLo docker-compose.yml \
  https://raw.githubusercontent.com/Julien-Au/gymcoach/main/docker-compose.selfhost.yml
printf 'JWT_SECRET=%s\nPOSTGRES_PASSWORD=%s\n' \
  "$(openssl rand -base64 48)" "$(openssl rand -hex 16)" > .env
docker compose up -d

Then open http://localhost:3000. Live demo: https://demo-gymcoach.mesureprivee.com (demo@gymcoach.app / gymcoachdemo).

Highlights

Log and train

• Offline-first set logging with shorthand (100x8@9) and AI-parsed free text
• Rest timer, plate calculator with a barbell diagram, warm-up ramps, supersets, drop sets
• A weight picker limited to the loads your own gym can make, and the equipment recorded on every set
• Readiness check-in and return-to-training calibration that adjust the suggested load and say why

Progress

• e1RM and max-load trends, weekly volume per muscle against MEV/MRV, a muscle heat map
• Records board, consistency calendar, stall detection and one-tap deloads
• Bodyweight, body measurements and private progress photos

Programs and cardio

• Established programs as written: 5/3/1 BBB, GZCLP, nSuns, PPL, Starting Strength, StrongLifts 5x5, Madcow, PHUL, PHAT
• First-class cardio, with FIT, TCX and GPX import and no cloud account. Imports from Strong and Hevy CSV, full export.

AI coach (bring your own model)

• Weekly debrief with one-tap program adjustments, a streaming chat coach, and program generation from a sentence
• Anthropic or any OpenRouter model. The coach only ever sees a structured summary of your own data, and a card shows exactly what it sees.
• An MCP connector for ChatGPT and other MCP clients

Self-hosting

• A prebuilt image on GHCR, one compose file, migrations on start, multi-user with strict data isolation
• Interface in English, French and Russian

Built by loops

Most of the 250+ merged PRs behind this release were written, reviewed and merged by Claude Code running in documented loops. The playbook, guardrails and lessons learned are in docs/loops/.

Thanks to the contributors: @SHAREN, @Retsumdk, @shaurya703, @mvnixon, @gogoku121.

Full list of changes: CHANGELOG.md