Cache RubyLLM responses based on semantic similarity, not exact string matching. Reduces costs and latency by returning cached responses for semantically equivalent queries.