Verdict
A Mac mini M4 Pro with 48 or 64 GB can run ZeroClaw next to a large local model served by Ollama, one of ZeroClaw's providers, so prompts never leave the machine. Llama 3.3 70B at Q4 is a 43 GB file: take 64 GB for room. ZeroClaw's tools run under macOS's Seatbelt sandbox by default. Until October 2026 this page said ZeroClaw denies network egress at iptables and bundles an offline RAG over a documents folder; its docs describe neither, and its sandboxed tools keep network access unless you restrict them.
Setup notes
Install Ollama for macOS and pull the model. Install ZeroClaw with its install script or the prebuilt macOS binary, run zeroclaw quickstart and pick Ollama as the model provider. Network limits are set per tool (allowed domains for the HTTP, browser and fetch tools) or by a firewall on the Mac.
Performance
Apple gives 5 W idle and 140 W maximum for the Mac mini M4 Pro. No published speed for a 70B model on this chip; a 7B model writes about 50 tokens per second on llama.cpp's scoreboard.
What breaks
- Anything requiring frontier-tier reasoning, with a local model
- Network isolation by default: sandboxed tools keep internet access until you restrict them
Sources
The agent's defaults and settings above were read in its own documentation on 2026-10-18.
- ZeroClaw README (v0.8.6): providers, channels, install, licence
- ZeroClaw NOTICE: dual licence, MIT or Apache-2.0
- ZeroClaw docs (v0.8.6): security model, default posture
- ZeroClaw docs (v0.8.6): sandboxing (auto-detection, network egress)
- ZeroClaw docs (v0.8.6): provider-agnostic
Want to know more
See the ZeroClaw page and the Mac Mini M4 / M4 Pro spec sheet.