name: mistral-local

services:
  local-ai:
    build:
      context: .
      dockerfile: Dockerfile
    image: local-ai
    container_name: local-ai
    networks:
      - mistral-net
    ports:
      - "127.0.0.1:8080:8080"
    # Limites de ressources
    deploy:
      resources:
        limits:
          memory: 12g
          cpus: "8"
    # Sécurité
    read_only: true
    security_opt:
      - no-new-privileges:true
    tmpfs:
      - /tmp:size=256m
    devices:
      - /dev/dri:/dev/dri
    group_add:
      - video
    environment:
      - GGML_VK_VISIBLE_DEVICES=0
    volumes:
      - ./ai-models/mistral-7b-instruct.gguf:/home/llama/models/mistral-7b-instruct.gguf:ro
      - ./workspace:/home/llama/workspace

    # Health check — vérifie que le serveur répond
    # healthcheck:
    #   test: ["CMD-SHELL", "curl -sf http://localhost:8080/health || exit 1"]
    #   interval: 30s
    #   timeout: 10s
    #   retries: 3
    #   start_period: 60s    # laisser le temps au modèle de charger

networks:
  mistral-net:
    name: mistral-net
    driver: bridge
    internal: false         # pas de passerelle vers Internet
    driver_opts:
      com.docker.network.bridge.name: br-mistral