services:
  voice-chat:
    build: .
    ports:
      - "8000:8000"
    volumes:
      # Cache models on the host so they survive container rebuilds
      - huggingface-cache:/cache/huggingface
      # LoRA adapters — drop .safetensors files into ./loras on the host,
      # reference them from config.yml as /cache/loras/<file>.safetensors
      - ./loras:/cache/loras
      # Avatar images uploaded via the web UI persist between restarts
      - ./avatars:/app/avatars
      # Mount source so you can edit code/config without rebuilding the image
      - ./config.yml:/app/config.yml:ro
      - ./configs:/app/configs:ro
      - ./server:/app/server:ro
      - ./static:/app/static:ro
      - ./tests:/app/tests
      - ./run.py:/app/run.py:ro
    deploy:
      resources:
        reservations:
          devices:
            - driver: nvidia
              count: 1
              capabilities: [gpu]

volumes:
  huggingface-cache: