vLLM is an open-source inference and serving engine designed for large language models (LLMs), offering fast and memory-efficient inference through GPU optimizations like PagedAttention and continuous batching. This tutorial provides a step-by-step guide on how to run LLMs using vLLM on a local machine, with the ability to expose them via a secure public API without relying on cloud services. By using Clarifai Local Runners and the Clarifai CLI, users can initialize, configure, and execute models locally, maintaining full control over the environment while leveraging GPU acceleration. The setup involves creating a model directory with essential files, customizing scripts for model interaction, and configuring runtime settings. The process culminates in starting a Local Runner that connects to the vLLM runtime, enabling secure routing of API requests to your machine for local execution. This setup facilitates testing, integration, and real-time streaming of model outputs, offering flexibility and security, with options to use the free tier or a paid developer plan for extended features.