Using LLMs for political questions is dumb in the first place. That said you can run distilled models like Deepseek or Mistral locally. That still doesn’t solve bias in their weights. Best is to not use AI for such questions at all.