# GPU host (.9, RTX 5050): the computers talk and play like people with the local model (docs/06 §12.1, docs/11 §8). # Ollama runs on this host as a system service on the GPU (OLLAMA_HOST=0.0.0.0:11434, model gemma3:4b); # the app container reaches it through the host gateway. Use together with the proxy file: # cd deploy && docker --context default compose -f docker-compose.proxy.yml -f docker-compose.ai.yml up -d --build services: app: environment: AI_URL: http://host.docker.internal:11434 AI_MODEL: gemma3:4b extra_hosts: - "host.docker.internal:host-gateway" # .9 also runs the chat bots and desktop services: keep the site small. mem_limit: 1g cpus: 2