docker-compose.yaml

services:
  transpiler-api:
    build:
      context: ./api
      dockerfile: Dockerfile
    container_name: transpiler-api
    ports:
      - "5001:5001"
    env_file:
      - .env
    environment:
      - INFERENCE_PROVIDER=${INFERENCE_PROVIDER:-remote}
      - INFERENCE_API_ENDPOINT=${INFERENCE_API_ENDPOINT}
      - INFERENCE_API_TOKEN=${INFERENCE_API_TOKEN}
      - INFERENCE_MODEL_NAME=${INFERENCE_MODEL_NAME}
      - LLM_TEMPERATURE=${LLM_TEMPERATURE:-0.2}
      - LLM_MAX_TOKENS=${LLM_MAX_TOKENS:-4096}
      - MAX_CODE_LENGTH=${MAX_CODE_LENGTH:-8000}
      - MAX_FILE_SIZE=${MAX_FILE_SIZE:-10485760}
    networks:
      - transpiler-network
    extra_hosts:
      # Allows the container to reach a custom domain mapped in /etc/hosts (remote API use case).
      # Leave as "not-needed" if your inference endpoint is a public URL.
      - "${LOCAL_URL_ENDPOINT:-not-needed}:host-gateway"
      # Allows the container to reach Ollama (or any service) running on the Docker host.
      # On macOS Docker Desktop this is automatic — the line below is a no-op but harmless.
      # On Linux with Docker Engine this is REQUIRED for host.docker.internal to resolve.
      - "host.docker.internal:host-gateway"
    restart: unless-stopped
    healthcheck:
      test: ["CMD", "curl", "-f", "http://localhost:5001/health"]
      interval: 30s
      timeout: 10s
      retries: 3
      start_period: 40s

  transpiler-ui:
    build:
      context: ./ui
      dockerfile: Dockerfile
    container_name: transpiler-ui
    ports:
      - "3000:8080"
    depends_on:
      - transpiler-api
    networks:
      - transpiler-network
    restart: unless-stopped

  # NOTE: Ollama is intentionally NOT defined as a Docker service here.
  #
  # On macOS (Apple Silicon / M-series), running Ollama inside Docker bypasses
  # Metal GPU acceleration, resulting in CPU-only inference that is extremely slow.
  # Ollama must be installed and run natively on the host machine so it can use
  # the Metal Performance Shaders (MPS) backend for full hardware acceleration.
  #
  # The backend container reaches host-side Ollama via host.docker.internal:11434.
  # See the INFERENCE_PROVIDER=ollama section in .env.example for setup instructions.

networks:
  transpiler-network:
    driver: bridge