deployment/docker-compose.yml

version: '3.8'

services:
  rag-service:
    build:
      context: ..
      dockerfile: deployment/Dockerfile.rag
    container_name: rag-service
    ports:
      - "8000:8000"
    volumes:
      - rag-chroma:/data/chroma
      - rag-documents:/data/documents
    deploy:
      resources:
        reservations:
          devices:
            - driver: nvidia
              count: 1
              capabilities: [gpu]
    environment:
      - LLM_MODEL=microsoft/Phi-3-mini-4k-instruct
      - EMBEDDING_MODEL=BAAI/bge-small-en-v1.5
      - CHROMA_PERSIST_DIR=/data/chroma
      - DOCUMENTS_DIR=/data/documents
    restart: unless-stopped

  web-app:
    build:
      context: ..
      dockerfile: deployment/Dockerfile.web
    container_name: web-app
    ports:
      - "5000:5000"
    environment:
      - RAG_SERVICE_URL=http://rag-service:8000
    depends_on:
      - rag-service
    restart: unless-stopped

volumes:
  rag-chroma:
  rag-documents: