services:
  ollama:
    image: ollama/ollama:latest
    ports:
      - "11434:11434"
    volumes:
      - ollama_models:/root/.ollama
    healthcheck:
      test: ["CMD", "ollama", "list"]
      interval: 5s
      timeout: 5s
      retries: 60
    # Uncomment the block below if you have an NVIDIA GPU and the
    # NVIDIA container toolkit installed. Vision models are dramatically
    # faster with a GPU, but everything in this example also runs on CPU.
    # deploy:
    #   resources:
    #     reservations:
    #       devices:
    #         - driver: nvidia
    #           count: all
    #           capabilities: [gpu]

  model-puller:
    image: ollama/ollama:latest
    depends_on:
      ollama:
        condition: service_healthy
    environment:
      - OLLAMA_HOST=http://ollama:11434
    entrypoint: ["/bin/bash", "-c"]
    # `ollama pull` shows a redrawing progress bar with carriage returns and
    # ANSI cursor-hide codes. Without a TTY (which is the case here) that
    # comes out of `docker logs` as invisible output plus a hijacked cursor.
    # The pipeline below converts \r into \n so each progress update becomes
    # its own log line, and strips ANSI escape sequences so the cursor is
    # left alone.
    command:
      - |
        set -eo pipefail
        echo "Pulling ${OLLAMA_MODEL:-qwen3.5:9b} (this may take a while on first run)..."
        ollama pull "${OLLAMA_MODEL:-qwen3.5:9b}" 2>&1 \
          | stdbuf -oL tr '\r' '\n' \
          | stdbuf -oL sed -E 's/\x1b\[[?0-9;]*[a-zA-Z]//g'
        echo "Model ${OLLAMA_MODEL:-qwen3.5:9b} is ready. You can now run the demo scripts on the host."
    restart: "no"

  web:
    build: ./web
    ports:
      - "8082:8080"
    environment:
      - OLLAMA_URL=http://ollama:11434
      - OLLAMA_MODEL=${OLLAMA_MODEL:-qwen3.5:9b}
    volumes:
      - ./scripts:/app/scripts:ro
      - ./sample_images:/app/sample_images:ro
    depends_on:
      ollama:
        condition: service_healthy

volumes:
  ollama_models: