Currently the best (most user-friendly) way to run models locally is to use Ollama with Continue.dev. This one is not available yet, though: https://github.com/ggerganov/llama.cpp/issues/8519