OpenClaw AI agent framework: system requirements and Apple Silicon hardware selection

Evaluating whether OpenClaw (the open-source AI agent messaging bridge) can run on a MacBook Pro M1 Pro 16GB vs a Mac Mini M4 Pro, and understanding what hardware resources it actually consumes.

OpenClaw is a Docker-based AI agent framework that bridges messaging apps (Signal, WhatsApp, Telegram, Discord) to LLM providers (Anthropic, OpenAI, Ollama). It runs natively on Apple Silicon and fully supports macOS. Resource usage splits into two modes: (1) Cloud/Gateway Mode — very lightweight, 2–4GB RAM, stable internet matters more than compute; (2) Local Inference Mode (Ollama) — extremely RAM-heavy, 16GB minimum for small 7B models, 32–48GB recommended for useful 14B–32B models. MacBook Pro M1 Pro 16GB works fine for cloud mode but is constrained for local inference: only ~10–11GB is free for the GPU after OS overhead, limiting you to 7–8B models with small context windows. OpenClaw's own processes (Node.js gateway, SQLite, messaging daemons, Playwright for web skills) compete with the LLM for that limited memory. Mac Mini M4 Pro is objectively better for persistent deployment: no battery degradation from continuous inference, superior thermal envelope for sustained GPU loads, built-in Ethernet for stable WebSocket connections, and headless operation without sleep workarounds. Recommended M4 Pro configs: 24GB (good for 14B models), 48GB (sweet spot — runs 32B models comfortably with room for concurrent agents), 64GB (for quantized 70B models). Docker requirements are lightweight: 4GB RAM allocated, 10GB storage for images/data.

The bottleneck is unified memory. In cloud mode, the LLM runs remotely so local RAM is nearly irrelevant. In local mode, unified memory serves as both system RAM and GPU VRAM simultaneously, so the full model weight must fit alongside OS and app overhead. 16GB is genuinely too tight for anything beyond toy models when also running OpenClaw's service layer.

Choosing hardware to run OpenClaw as a persistent 24/7 AI agent service on macOS, deciding between a laptop (M1 Pro 16GB) and a desktop (M4 Pro Mac Mini).