They probably hope to run a LLM locally on the Mac mini but they don't realize that decent models require much more computational power
Most don't realize that until they come across configuring openclaw or other agentic frameworks and then realize they've to use anthropic or openai via API (additional cost)