Skip to content

Mac Mini M4 Hermes Agent Setup — Standalone AI Workstation

The Mac Mini M4 is the ideal single-machine Hermes Agent host for solo founders and developers. Everything runs on one box: LLM inference with Ollama and MLX, browser automation with Playwright, cron scheduling, and messaging — no worker nodes, no SSH keys, no multi-machine complexity.

Overview

The Mac Mini M4's unified memory architecture (16–32GB shared CPU/GPU) makes it uniquely suited for running local AI models. Combined with silent operation (~20W idle) and native macOS support for Playwright browser automation, it's the recommended Hermes Agent setup platform for solo operators who want a single-box solution.

How It Works

Component How It Runs on Mac Mini M4
Local Models Ollama + MLX; up to ~13B parameters comfortably with 16GB RAM
Cloud Models OpenRouter or direct Anthropic/OpenAI/DeepSeek API access
Browser Automation Playwright runs natively on macOS; patchright for Cloudflare-bypass
Memory Honcho (peer memory), GBrain (project knowledge), memcore-cloud (cross-session)
Crons Hermes cron scheduler with launchd for auto-restart
Messaging Native Telegram, Slack, Discord, and 17+ messaging platforms

Step-by-Step Installation

Step 1: macOS Prerequisites

# Install Homebrew if you don't have it
/bin/bash -c "$(curl -fsSL https://raw.githubusercontent.com/Homebrew/install/HEAD/install.sh)"

# Dependencies
brew install python@3.12 ffmpeg node git

Step 2: Install Hermes Agent

pip install hermes-agent
hermes --version

Step 3: Model Setup

Pick one or both strategies:

Local Models (Ollama — Free):

# Install Ollama
curl -fsSL https://ollama.com/install.sh | sh

# Pull models
ollama pull llama3.2          # Lightweight daily driver
ollama pull nomic-embed-text   # Embeddings for memory
ollama pull qwen2.5:14b        # Heavier tasks (if 24GB+ RAM)

# Configure Hermes to use Ollama
hermes config set model.default ollama/llama3.2

Cloud Models (OpenRouter — Pay-per-use):

hermes config set providers.openrouter.api_key "your-key"
hermes config set model.default openrouter/anthropic/claude-sonnet-4
hermes config set model.fallback "openrouter/qwen/qwen3-235b-a22b:free"

Strategy: Use local models for cron tasks and embeddings (free), cloud models for complex reasoning (pay-as-you-go). Set Ollama as primary and OpenRouter as fallback, or vice versa depending on budget. See our model selection guide for detailed tiering strategies.

Step 4: Browser Automation

Playwright runs directly on the Mac. No worker node needed.

pip install playwright
playwright install chromium

# Test
python3 -c "from playwright.sync_api import sync_playwright; print('OK')"

For sites with Cloudflare protection, add patchright:

pip install patchright
patchright install chromium

Step 5: Social Publishing (Postiz)

npm install -g postiz-cli
postiz auth
# → Opens browser: log in and authorize platform connectors

Step 6: Persistent Operation

Gateway (Background):

# Start as a launchd service for auto-restart
hermes gateway run --replace

# Verify
hermes gateway status

Cron Jobs:

# Email monitoring every 15 minutes
hermes cron create \
  --name "email-watch" \
  --prompt "Check inbox for unread. Summarize new emails. Silent if empty." \
  --schedule "*/15 * * * *"

# Daily summary at 6 PM
hermes cron create \
  --name "daily-report" \
  --prompt "Summarize today's activity. What happened? What's pending?" \
  --schedule "0 18 * * *"

See cron design best practices for production-grade scheduling patterns.

Keep Alive:

Prevent macOS sleep during operation:

# Keep system awake while Hermes runs
caffeinate -dims &

Or go to System Settings → Battery → Options → Prevent automatic sleeping on power adapter.

Step 7: Memory Stack

# Honcho — peer memory (2 min)
hermes mcp add honcho -- npx mcp-remote https://mcp.honcho.dev \
  --header "Authorization: Bearer your-h...ken" \
  --header "X-Honcho-Workspace-ID: your-workspace"

# GBrain — project memory (10 min)
git clone https://github.com/garrytan/gbrain && cd gbrain && ./setup.sh

# memcore-cloud — cross-session context (5 min)
pip install memcore-cloud && memcore-cloud init

Full details in the memory architecture guide.

Benefits of Mac Mini M4 + Hermes Agent

  • Single-box simplicity: No worker nodes, no SSH, no multi-machine coordination
  • Unified memory: 16–32GB shared between CPU and GPU — ideal for local LLM inference
  • Silent operation: ~20W idle, ~40W under load — leave it running 24/7
  • Native browser automation: Playwright and patchright run directly on macOS
  • Developer ecosystem: Homebrew, Python, Node.js — everything just works
  • Cost efficiency: $599 one-time hardware, free local models, ~$3/month electricity

Cost Summary

Item Cost
Mac Mini M4 hardware $599 (one-time)
Ollama models Free
OpenRouter API ~$5–20/month
Honcho Free tier available
Electricity ~$3/month at $0.15/kWh

Total: ~$600 upfront, $10–25/month ongoing.

FAQ

The Mac Mini M4 offers silent operation, lower power consumption (~20W vs 150W+), unified memory architecture that's ideal for LLM inference, and native macOS support for all Hermes Agent features. A gaming PC setup provides more raw GPU power but at higher cost, noise, and power draw.

Can I use only local models on Mac Mini?

Yes. With 16GB RAM you can comfortably run models up to ~8B parameters; with 24GB+ you can run 13B–14B models. For heavier workloads, supplement with cloud models via OpenRouter as a fallback.

How do I prevent my Mac Mini from sleeping?

Use caffeinate -dims & from the terminal or go to System Settings → Battery → Options → Prevent automatic sleeping on power adapter.

What if I need more GPU power?

Add a gaming PC worker node via SSH for GPU-heavy inference, or use cloud GPU instances for burst workloads.

Next: Cron Scheduling · MCP Integration