How to Run AI Agents 24/7 on a Mac Mini — Complete Setup Guide
A step-by-step guide to setting up OpenClaw AI agents to run autonomously on a Mac mini, handling bug bounty research, concierge support, and security monitoring around the clock.
If you're running any kind of software business today, there's a version of your work that could be handled by a well-configured AI agent: monitoring, research, customer support, bug bounty hunting, security audits.
The problem is that most AI agents are either cloud-hosted (expensive, latency, dependency on third parties) or laptop-based (dies when you close the lid, depends on your home WiFi). The Mac mini changes that equation.
This guide covers how we run our entire agent operation — multiple specialized AI agents handling bug bounty research, concierge support, and social content — on a single Mac mini that runs 24/7 in our office without a display.
Why the Mac Mini Makes Sense
The Mac mini with the M4 chip is roughly $600-800 and provides:
Our current setup runs 5 agents simultaneously: Steve (smart contract audits), Atlas (social media), Spy (bounty pipeline), Janet (ops), and Henry (orchestrator).
Step 1: Install OpenClaw
``bash`
brew install openclaw
openclaw setup
OpenClaw handles the agent framework, credential management, scheduling, and messaging integrations. After installation, configure your first agent:
`bash`
openclaw agent add --name Steve --role "smart contract auditor"
openclaw agent add --name Atlas --role "social media manager"
openclaw agent add --name Henry --role "orchestrator"
Step 2: Configure LaunchAgents for 24/7 Operation
The key to headless Mac operation is launchd, macOS's built-in service manager. Unlike cron, launchd jobs survive sleep/wake cycles and restart automatically on reboot.
Create a LaunchAgent for each agent:
`bash`
mkdir -p ~/Library/LaunchAgents
Example for Henry (our orchestrator):
`xml`
The RunAtLoad key ensures the agent starts when the Mac boots — no login required.
Step 3: Connect Your Messaging Platform
We use Telegram as our control channel — it's fast, supports bots, and works well for agent communication. In OpenClaw:
`bash`
openclaw channel add --platform telegram --bot-token YOUR_BOT_TOKEN
openclaw channel test telegram
You now have a bot that your agents can message through, and you can send commands to them directly.
Step 4: Set Up Your First Agent Skill
For bug bounty hunting, we use a specialized skill that handles program discovery, scope validation, and report generation. Install it from ClawHub:
`bash`
openclaw skill install security-auditor
openclaw skill configure security-auditor
For social media management:
`bash``
openclaw skill install social-media-agent
openclaw skill configure social-media-agent
Step 5: Configure Overnight Runs
The real leverage is scheduling. Here's our daily schedule:
| Time | Agent | Task |
|------|-------|------|
| 1:00 AM | Spy | Research: find new bounty programs |
| 2:00 AM | Henry | Decide and execute highest-value task |
| 6:00 AM | (you) | Receive daily briefing via Telegram |
| 11:30 PM | Henry | Dump daily journal to Supabase |
This "agents running while you sleep" model is where the productivity multiplier lives. You wake up to a full briefing of what the team accomplished overnight.
What We Run on Our Setup
Our Mac mini runs:
Total cost: ~$40/month in electricity vs $300-500/month for equivalent cloud compute.
Getting Started
The concierge installation at [openclawlaunchpad.com/concierge](/concierge) handles the full setup for you — installation, configuration, agent deployment, and messaging setup. If you want to DIY, the OpenClaw documentation covers everything above in more detail.
The key insight: the agents aren't just time savers. They're a force multiplier. When your agent finds a $50K bug bounty while you're sleeping, or your social agent fills your content calendar while you're in a meeting, that's compounding value that traditional outsourcing doesn't match.