Video summary
I Bought 4 Mac Minis to Run My AI Agents (OpenClaw Setup & Tutorial)
Main summary
Key takeaways
Tech setup & purpose (OpenClaw on multiple Mac minis)
- The creator (Kevin Builds) launches OpenClaw on four Mac minis to run AI agent workflows continuously.
- They report OpenClaw previously handling a full “video production system”:
- research → scripting → voiceover → B-roll/video editing
- Claimed results: ~4 million views in a month.
Why use 4 Mac minis
- The goal is scaling parallel tasks by running multiple agent instances/workflows instead of relying on a single machine.
Tutorial: Set up a Mac mini from scratch for OpenClaw
Browser/account isolation for security
- Download Chrome
- Sign in using a dummy/burner email to avoid personal information
Remote access with Jump Desktop (streaming multiple minis)
- Install Jump Desktop
- Purchased from the Mac App Store for about $35 one-time
- Use Jump Desktop to stream/control all Mac minis from a MacBook Pro
Prevent sleep/disconnect
- Configure Screen Lock / Screen Saver settings so the Mac mini never turns off
- Ensure the system stays awake (changes require a password)
No-monitor workaround (dummy HDMI adapters)
- Order a dummy HDMI adapter (example: 4K 120 Hz)
- This makes the Mac minis behave as if a monitor is connected, keeping them usable remotely
Tutorial: Install OpenClaw via Terminal + Homebrew
- Steps include:
- Open Terminal via Spotlight
- Copy the OpenClaw website install command/script into Terminal
- Use
sudobefore the command
- Install process:
- Includes Homebrew and OpenClaw
- Takes roughly 5–10 minutes
Important installer prompt guidance
- If the installer asks a key question:
- Select “yes”
- Choosing “no” may force a restart/recovery, and the creator recommends getting guidance to return to setup.
Tutorial: Quick start + choosing models via OpenRouter
Quick start configuration
- Select OpenRouter as the model provider
- Create an OpenRouter account (free) and generate an API key
- Add the OpenRouter API key into OpenClaw on each Mac mini
Model selection
- Uses “4.6” as an onboarding option on all machines
- Cost and preference notes:
- Avoid Opus due to expense
- Prefer Sonnet for onboarding
- Plan to switch later after training/onboarding:
- Haiku / Gemini Flash
API key handling / cost note
- Creating keys is straightforward, but the creator emphasizes:
- Store keys securely
- Avoid leaving them in plain/unprotected locations
Telegram bot integration (for talking to the agent)
- Recommends using Telegram for channels/communication
Bot setup via BotFather
- Install Telegram on your phone
- Start/mention @BotFather
- Send
/newbot - Provide:
- bot name
- bot username (underscore + “bot” format)
- Copy the resulting bot token/key into OpenClaw
Skipped/optional setup steps initially
- Provider and Skills configuration can be deferred
“Hatching” / interface
- Select Open WebUI as the local chat/testing interface
- Telegram pairing code may be required, then Telegram chat is enabled inside Open WebUI
Optional monetizable “skills” (group/community)
- Mentions a pre-built “skills” package available in a group/community
- Claim: it can help users avoid spending thousands in API credits by not building skills from scratch
- Creator’s use case: OpenClaw for real agency automation tasks, not just simple email checking
Main speakers / sources
- Speaker: Kevin (Kevin Builds)