Show HN: AutoThink – Boosts local LLM performance with adaptive reasoning