services:
  voice-chat:
    build: .
    ports:
      - "8000:8000"
    volumes:
      # Cache models on the host so they survive container rebuilds
      - huggingface-cache:/cache/huggingface
    deploy:
      resources:
        reservations:
          devices:
            - driver: nvidia
              count: 1
              capabilities: [gpu]

volumes:
  huggingface-cache: