customer4 — LLM chat demo

Running on Cozystack/Talos at gulfgenai.com. Model: qwen2.5:0.5b via Ollama, pure CPU inference.