Been tinkering with local LLM setups on my machine since April. Ditched the API dependencies from the big players—Anthropic, OpenAI, and others. Running models locally gives you actual control and privacy. Just wrapped up a year of experimenting in 2025 and picked up some solid insights along the way. Here's what I figured out.