OpenClaw probes Mac mini M1 limits

Good starter rig, but not a magic autonomy box. The Mac mini can teach you the stack and keep data local, yet once model size and context grow, the experience turns into a memory-management exercise.

• –OpenClaw's docs explicitly support Ollama-backed local models, with 16GB as the floor for 7B+.
• –On Apple M1 16GB, the best bets are Llama 3.1 8B Instruct, Qwen3 8B, Mistral 7B, Qwen2.5 Coder 7B, or Gemma 2 9B in Q4 quantization.
• –The thread's lone suggestion of Qwen 3.5 35B A3B feels more like a stretch target than a default choice on this machine, since that class of model leaves very little headroom.
• –If you want privacy, offline use, and no API bills, local is worth trying. If you want the smoothest assistant experience, cloud still wins.
• –If you are buying hardware for local AI, more RAM buys more freedom than a slightly newer chip.