Training OS

Training OS is a personal training coach and analytics playground — a weekly-first training calendar that combines structured sessions (FIT/Strava) with a lightweight planning layer and an LLM-powered coach interface.

Highlights

• Week-centric UX: separate Actuals (sessions & notes) and Plan (weekly intent)
• Training Load pipeline: TL (per-hour based), ATL (7d), CTL (42d) and ACWR
• Live coaching tools: MCP-based LLM tools that can query your session history
• Polished visual reports: merged CTL+ACWR view, per-week CTL shown as-of-today

Screenshots

• Calendar (weekly sessions, plan & actuals, inline edit):

  

• Coach (LLM-powered coaching chat):

  

• Trends (CTL + ACWR, period-based analytics):

  

• Training Load (weekly load, daily breakdown, week stats):

  

• Weekly calendar workflow for sessions and day notes

• Weekly plan as a separate intent layer

• MCP tools for agentic coaching queries

• Analysis views (analysis.html, training-load.html, chat-history.html)

• Strava sync + FIT import

• Training load pipeline (TL / ATL / CTL / ACWR)

Quick start

One command (dev)

./scripts/dev_up.sh

Manual (backend + frontend)

# backend
cd backend
python3 -m venv venv
source venv/bin/activate
pip install -r requirements.txt
uvicorn app.main:app --reload

# frontend (new terminal)
cd ../frontend
python3 -m http.server 3000

• App: http://localhost:3000
• API docs: http://localhost:8000/docs

MCP server (coaching tools)

The LLM service runs in MCP mode and can call tool functions over your training data before composing an answer. Key tools include get_week_summary, get_day_details, get_session_details, and more.

Main implementation points:

• Tool schema + executor: backend/app/llm/mcp_tools.py
• LLM orchestration: backend/app/llm/service.py
• Provider adapters: backend/app/llm/providers.py
• Prompt loading/compilation: backend/app/llm/prompt_loader.py

Training load model (TL / ATL / CTL)

Training OS uses a moving-time based TL signal and derives:

• ATL: 7-day exponential response
• CTL: 42-day exponential response
• ACWR: ATL / CTL

Parameters live in backend/app/core/training_load_defaults.py and the propagation logic is implemented in backend/app/training_load.py.

Strava sync and import

# refresh new Strava activities
curl -X POST "http://localhost:8000/api/integrations/strava/import/refresh"

# full Strava backfill
curl -X POST "http://localhost:8000/api/integrations/strava/import/backfill?per_page=100&max_pages=40"

# FIT import
cd backend && source venv/bin/activate && python scripts/import_fit.py

Tech stack and libraries

Backend:

• FastAPI, Uvicorn
• SQLAlchemy (SQLite)
• Pydantic / pydantic-settings
• fitparse / fitdecode for FIT ingestion

Frontend:

• Alpine.js
• Tailwind CSS (CDN)
• Chart.js
• marked + DOMPurify for coach markdown rendering

Project direction

Roadmap is intentionally paused for now. See ROADMAP.md for the current maintenance-mode status.