Serious local LLM work needs better tools than Ollama

Ollama and llama.cpp are the default for local LLMs but fall short for serious workflows. vLLM offers an OpenAI-compatible API server with high-throughput inference, continuous batching, and prefix caching. The runtime matters as much as the model when building agents or deploying on Macs.