🏋️ Fitness Coach — AI Personal Trainer

English | Русский

Open-source AI fitness coach with multi-provider support. Build personalized workout plans, track progress, and get intelligent coaching through a conversational interface.

Features

• Multi-Provider AI: Choose Anthropic Claude, OpenAI GPT, or self-hosted Ollama
• 27 MCP Tools: Comprehensive workout management with parallel execution
• Graph Knowledge Base: Exercise relationships, alternatives, and progressions
• Conversational Coaching: Natural language chat with tool-calling capabilities
• Smart Plan Creation: Efficient full-plan creation in a single tool call
• Workout Management: Create, track, and adapt workout plans
• Progress Tracking: Statistics, streaks, and completion metrics
• Offline Support: PWA with service worker caching
• Memory System: Optional RAG for personalized long-term context
• Docker Ready: Production-grade containerized deployment

Quick Start

1. Clone

2. Configure

3. Run

Open http://localhost:8000/docs for API documentation.

Architecture

fitness-coach/
├── backend/                 # FastAPI + SQLAlchemy
│   ├── app/
│   │   ├── api/fitness/     # REST endpoints (plans, workouts, chat)
│   │   ├── models/          # SQLAlchemy models
│   │   ├── providers/       # AI, embedding, RAG, memory providers
│   │   └── services/        # Business logic + AI agent
│   └── requirements/        # Modular dependencies
├── frontend/                # Next.js 16 PWA
│   ├── src/app/             # App Router pages
│   ├── src/components/      # React components
│   └── src/stores/          # Zustand state management
└── docker/                  # Docker Compose deployment
    ├── docker-compose.yml   # Production config
    └── docker-compose.dev.yml # Development overrides

┌─────────────┐     ┌─────────────┐     ┌─────────────┐
│   Frontend  │────▶│   Backend   │────▶│  PostgreSQL │
│  (Next.js)  │     │  (FastAPI)  │     │  + pgvector │
└─────────────┘     └──────┬──────┘     └─────────────┘
                           │
              ┌────────────┼────────────┐
              ▼            ▼            ▼
        ┌──────────┐ ┌──────────┐ ┌──────────┐
        │ Anthropic│ │  OpenAI  │ │  Ollama  │
        │  Claude  │ │   GPT    │ │  (Local) │
        └──────────┘ └──────────┘ └──────────┘

AI Provider Comparison

See backend/README.md for detailed provider configuration and available models.

Core Capabilities

27 MCP Tools organized by category:

• Basic (7): Plans, workouts, stats, history
• CRUD (8): Create/edit plans, weeks, days, exercises
• Batch (2):
  create_full_plan
  ,
  create_full_week
  for efficient creation
• Graph (4): Exercise alternatives, progressions, muscle-exercise mapping
• RAG (2): Search workout memory, store insights
• Status (3): Complete, skip, add notes
• Other (1): Training programs

Performance: Parallel tool execution via asyncio.gather for multi-tool operations.

Knowledge Base: NetworkX or Neo4j graph for exercise relationships and progressions.

Documentation

Development

Backend

Frontend

Contributing

We welcome contributions. See CONTRIBUTING.md for guidelines.

License

MIT License - see LICENSE for details.