It depends on whether you're running the models locally. If you're just using a Claude or OpenAI token (as probably 95%+ of OpenClaw users are), the RAM requirements are minimal. My first-gen M1 Mac Mini runs it just fine.