Coach AI

Your coach, wherever your agent lives.

Coach AI is not an app you run — it's a portable coaching agent definition that installs into agent harnesses you already use (currently Claude Code and Codex CLI). A small coach CLI configures a coach personality, a set of skills (SKILL.md files), and live MCP connections to your fitness data. The harness's own reasoning, memory, and scheduling (/loop, /schedule) do the rest.

📖 Full documentation — architecture, capabilities, skills catalog, source integrations, daily workflow, development & testing, and roadmap.

---

Quickstart

Requires Python 3.12+ and uv.

git clone https://github.com/AchintyaX/Coach.ai.git
cd Coach.ai
uv sync

Pick a path and authenticate its source(s) — see Installation for the full one-time auth steps:

# Garmin path
coach setup --source garmin

# — or — Strava + Calendar path (pick one calendar)
coach setup --source strava
coach setup --source google_calendar   # or: coach setup --source outlook_calendar

Then install into your harness and (optionally) set up the daily loop:

coach install --harness claude   # or: codex, all
coach setup --schedule           # interactive — asks what time to run readiness-check + generate-daily-workout

Everything above is idempotent — re-run any step any time your sources, skills, or personality change. For what each coach command writes and where, see Architecture & principles and Installation.

---

Two functional paths

Coach AI ships two functional paths — a path is one metrics source plus one workout-calendar source (Garmin covers both). The installer resolves your sources' combined capabilities and installs only the skills (and tool resolutions) that capability set supports — nothing is faked when a capability is missing. See Capabilities & paths for the full comparison.

• Garmin — full readiness/HRV/sleep/body-battery/training-load data, plus structured workouts pushed to and rescheduled on the Garmin calendar.
• Strava + Google/Outlook Calendar — Strava supplies activity streams and PRs; the calendar is the workout system-of-record, with each session written as a free-text event.

---

Development & testing

uv sync
uv run pytest -q

See Development & testing for the repository structure, fixture-driven functional tests, and the full skill × path test matrix.

---

Contributing

Issues and PRs are welcome — please open an issue first for new sources, skills, or larger changes so we can align on approach. Priority areas: new SourceSpecs (Apple Health, Whoop, Polar, Oura), additional harness support beyond Claude Code/Codex, and refinements to the skill catalog.

License

MIT