I asked my AI agent for a private LLM client/server. It built one out of MQTT and 1,000 lines

A developer has created a lightweight, private client/server architecture for accessing a local Ollama-powered LLM from a mobile device without exposing home infrastructure to the internet. By utilizing MQTT as a message bus, the system allows both the client and the host PC to 'dial out' to a broker, eliminating the need for port forwarding, reverse proxies, or VPNs. The entire solution consists of approximately 1,000 lines of code, including a browser-based UI, a local service, and a Python plugin. The project, built using the AllSpeak language, emphasizes a minimal security footprint by using access tokens for authentication and MQTT ACLs for fencing. This approach provides a secure, low-latency way to interact with local GPU-accelerated models while maintaining complete control over the network environment, avoiding the complexity of traditional containerized or cloud-tunneling deployments.