A few weeks ago I wrote about running OpenClaw on a single Mac Mini. That post ended with me saying it felt like the future.
I meant it. So I kept going.
Two Mac Minis. One Dell Optiplex 7060. Two macOS VMs on my home server. Five OpenClaw instances, all running 24/7, all with a job. Total hardware cost to get here: under $1,000.
Based on our math at Hyperflow, this setup will pay for itself five times over in the first 30 days. Not in some abstract “productivity gains” way. In actual dollars saved and actual revenue generated.
I want to walk you through every machine, every workflow, and exactly why I think this is 100% worth you testing yourself.
The Fleet
The Fleet
Mac Mini #1 is the original. Already had this one sitting in my office. Still my personal assistant. Morning briefs, daily scheduling, async tasks. The one I wrote about before. It hasn’t changed because it didn’t need to.
This is the machine that got me hooked. Every morning it sends me a brief through iMessage before I’m even out of bed. Thought for the day, weather, my schedule, jacuzzi temperature. Throughout the day I throw half-baked ideas at it and walk away. It finishes them async and pings me when they’re ready. It manages my calendar conflicts, handles research tasks, and keeps track of things I’d otherwise lose in a notes app I never open.
It replaced the habit of checking email first thing. That alone was worth the experiment.
Mac Mini #2 I bought specifically for this. It handles client work at Hyperflow. It monitors project repos, watches for deployment issues, and manages the repetitive parts of our client workflows. When a client pushes code and something breaks at 2am, this machine knows about it before I wake up.
Here’s what that actually looks like: a client pushes to their staging branch. GitHub fires a webhook. Mac Mini #2 sees it, checks the CI status, and if something fails, it parses the error logs, generates context about what broke and why, drops an alert in the client’s shared Slack channel, and creates a Linear issue. All before I’ve had my first cup of coffee. That’s not a hypothetical. That happened twice last week.
This machine alone probably saves us 20+ hours a month of reactive work. Hours we used to spend triaging issues that are now triaged before we even know they exist.
The Optiplex 7060. $90 off Facebook Marketplace. I stripped it down and put Linux on it (that’s a whole other post for another day). This thing runs OpenClaw without breaking a sweat.
Why Linux? Because this machine doesn’t need macOS. It doesn’t need a GUI. It needs to sit in a corner, stay on, and process content. Linux does that better, faster, and with fewer resources than macOS. Plus I wanted to see how far I could push OpenClaw on minimal hardware. Answer: pretty far. Ninety dollars.
This is the content engine. It watches a specific Google Drive folder for new assets. When I drop a raw video in, the Optiplex picks it up, runs it through our editing pipeline (auto-cut, captions, crop for different platforms), and outputs a finished product. I give feedback through OpenClaw via iMessage. “Make the intro shorter.” “Add captions.” “Crop for vertical.” No project files. No editing software on my laptop. Just text messages.
Same thing for social content. It coordinates with Postmatic for graphics, manages the publishing schedule, and queues everything up for my approval in our scheduling app. I review, I approve, it posts. That’s the whole workflow.
VM #1 and VM #2. These run on my already established home server (also another post, also another day). The VMs aren’t running OpenClaw for production workflows. They’re for testing local models.
Right now I’m running Kimi as a local model on VM #1. That means I can test things without burning API credits and without sending data to external servers. It’s my playground for evaluating different models, running benchmarks, and figuring out what works before I commit to anything in production.
VM #2 is the sandbox. When I want to try a new OpenClaw skill, test a new configuration, or prototype a workflow, it goes here first. If it works, I promote it to one of the dedicated boxes. If it breaks, nothing else is affected. Cheap experimentation with zero blast radius. That’s the whole point.
What’s Actually Running
Let me get specific. Here’s every major workflow running across the fleet right now.
Live Workflows
Morning Brief (Mac Mini #1)
Every day at 6am, Mac Mini #1 kicks off a sequence. It pulls my calendar from Google, checks the weather for my location, reads the jacuzzi temperature from my smart home setup, and generates a thought for the day through Claude. It composes all of that into a single formatted message and sends it to me via iMessage.
By the time I pick up my phone, everything I need to know about my day is already there. No checking three apps. No scrolling through notifications. One message. Done.
I’ve talked about this before, but it’s worth repeating: this is the thing that changed my relationship with my phone in the morning. I stopped opening email first. I stopped doom-scrolling. The brief gives me what I need and I move on.
Content Pipeline (Optiplex 7060)
This is the one that saves the most time.
I record a video or take a photo. I drop it in a Google Drive folder. The Optiplex detects the new file, classifies what type of content it is, and runs it through the appropriate pipeline. For video that means auto-editing, adding captions, and cropping for different platforms. For images it means coordinating with Postmatic to generate social graphics in my brand.
When it’s done, it queues everything in our scheduling app and pings me on iMessage: “3 new pieces ready for review.” I open the app, scroll through, approve or tweak, and they’re scheduled. An entire content production workflow that used to take hours now takes me about five minutes of review time.
The Optiplex cost me ninety dollars and it’s producing more content than I could manage with a part-time hire.
Client Monitoring (Mac Mini #2)
This is the one that keeps clients happy.
Every Hyperflow client with a shared Slack channel is connected to this machine. It watches their GitHub repos for pushes, monitors CI pipelines, and checks deployment status on Vercel. When something breaks, it doesn’t just alert. It analyzes. It pulls the error logs, figures out what went wrong, generates a summary with context, and posts it in the client’s Slack channel. Then it creates a Linear issue with the details so nothing falls through the cracks.
Before this, we were reactive. A client would message us: “Hey, the staging site is broken.” Now they get a message from us first: “We noticed a deployment issue on staging. Here’s what happened and we’re looking into it.” That’s a completely different client experience.
Email Sequences (VM #2, promoted to production)
This one actually started as an experiment on VM #2 and worked well enough that we kept it running there.
When someone signs up through our site, a webhook fires to Convex. The system determines which email sequence they should get based on how they signed up. It generates personalized emails using Claude with our brand templates, sends them through Resend, tracks engagement (opens, clicks), and queues the next email in the drip based on their behavior.
It’s not a complex email marketing platform. It’s just enough automation to make sure nobody signs up and hears nothing from us for two weeks. The emails feel personal because they are. Claude writes them based on actual context, not a generic template.
The Stuff That’s Harder to Categorize
Beyond the big workflows, there are dozens of smaller things running across the fleet. Mac Mini #1 handles random research tasks I throw at it throughout the day. Mac Mini #2 generates weekly status reports for clients. The Optiplex manages our social posting schedule. VM #1 is currently benchmarking a new local model I’m evaluating.
None of these are individually groundbreaking. But stacked together, running 24/7, not needing my attention? They add up to something that fundamentally changes how I spend my time.
The Glue: Tailscale
Five machines spread across my office and my server rack don’t do much if they can’t talk to each other securely. That’s where Tailscale comes in.
Every machine in the fleet is on my Tailscale network. That means they can communicate with each other over encrypted connections without exposing anything to the public internet. I can access any machine from anywhere. My phone, my laptop, whatever. Doesn’t matter where I am.
Tailscale also makes the security story way simpler. No port forwarding. No public IPs. No VPN configuration headaches. Each machine has a stable address on the tailnet and can only be reached by devices I’ve authorized. That’s it.
When the Optiplex finishes editing a video, it notifies Mac Mini #1 through the tailnet. When Mac Mini #2 catches a deployment issue, it can trigger an alert that routes through the same network. The VMs pull model weights and configs from a shared location on the tailnet without ever touching the open internet.
If you’re running more than one machine for anything, Tailscale is the answer to “how do I connect them without exposing them?” I’d recommend it even if you’re not doing any of this AI stuff. It’s just good infrastructure.
One mesh network. Five machines. Zero exposure.
Under $1,000. All In.
People hear “five machines” and think expensive. Here’s the actual cost:
| Item | Cost |
|---|---|
| Mac Mini #1 | Already owned |
| Mac Mini #2 | ~$500 (bought used) |
| Optiplex 7060 | $90 (Facebook Marketplace) |
| VM #1 and VM #2 | $0 (existing server) |
| OpenClaw | Free |
| Tailscale | Free tier |
| Total hardware | Under $1,000 |
Monthly recurring:
| Service | Cost |
|---|---|
| Claude (new dedicated account) | $100/month |
| Electricity | Negligible |
| Total monthly | ~$100/month |
I signed up for a separate Claude account at $100/month just for the fleet. I wanted clean separation between my personal Claude usage and what the machines are doing. At $100/month for the backbone that powers five autonomous workflows? That’s the best money I spend each month.
Now here’s where it gets interesting. We sat down and ran the numbers at Hyperflow. Between the client monitoring that catches issues before they become emergencies, the content production running on autopilot, the email sequences that used to eat hours of manual work, and the async task handling across the board, this fleet saves us roughly 80+ hours a month of work that we were either doing ourselves or would need to hire for.
At our rates, that’s a 5x return in the first 30 days. Not projected. Not optimistic. Conservative.
A $90 Dell from Facebook Marketplace is doing the job of a part-time content producer. A used Mac Mini is doing the job of a junior DevOps engineer monitoring deployments. Two VMs on a server I already own are running local models and prototyping workflows for free.
The total investment is less than one month of a single part-time contractor. And it runs 24/7 without PTO, sick days, or Slack messages asking for clarification.
That’s not hype. That’s just math.
The Security Part (Read This)
I’m going to say this louder than last time because now I have five of these things running.
Use common sense.
Five machines running AI agents that can control a mouse, keyboard, and browser is not something to be casual about. Here’s exactly what I do:
Network isolation. Every machine is on Tailscale. Nothing is exposed to the public internet. Each machine can only communicate with authorized devices on my tailnet. No port forwarding. No public IPs.
Credential separation. The client machine doesn’t have access to my personal accounts. The content box doesn’t touch deployment pipelines. Each machine only has the credentials it needs for its specific job. Nothing more.
Dedicated hardware. I don’t run OpenClaw on my daily driver. My main laptop, where I log into my bank, where I have personal files, where I browse the internet? OpenClaw doesn’t touch it. Full stop.
Monitoring. I keep an eye on what each machine is doing. Not obsessively, but regularly. If something looks off, I investigate. These are powerful tools. Treat them that way.
Don’t run OpenClaw on your main machine with your bank logged in. Don’t expose it to the internet and walk away. Don’t skip the basics because the tech is exciting. Don’t store sensitive credentials on machines that don’t need them.
The security concerns I talked about in my last post haven’t gone away. They’ve multiplied by five. So I multiplied my precautions to match.
Going all in doesn’t mean going all in on being reckless. Set it up right or don’t set it up at all.
What Could You Build?
I’m running content production, client monitoring, and email sequences. That’s my world. But this setup isn’t limited to what I’m doing with it.
Pick an industry. Look at what a fleet of cheap machines and OpenClaw could do.
What could you build?
Pick an industry. See the fleet.
The point is: if you’re doing repetitive work that follows a pattern, a $90 computer and OpenClaw can probably do it for you. The only limit is what you decide to automate.
Dream big. Start small. Scale fast.
Why You Should Try This
I’m not saying you need five machines. Start with one. A cheap refurb off Facebook Marketplace. An old laptop. Whatever you have sitting around.
Install OpenClaw. Set up one workflow. Maybe it’s a morning brief. Maybe it’s monitoring a repo. Maybe it’s just a research assistant you can text from your phone. Start small. See what happens.
Here’s what I think will happen: you’ll start finding more things for it to do. Not because the technology is addictive, but because once you experience async, autonomous AI that actually does things, you’ll start seeing all the places in your day where you’re doing repetitive work that doesn’t need your brain.
That’s what happened to me. One Mac Mini turned into five machines in a few weeks. Not because I planned it. Because each one proved its value so fast that the next one was obvious.
The total investment is under a grand and $100 a month. The return, based on our real numbers at Hyperflow, is 5x in the first month. That’s the kind of bet I’ll make every single time.
Want to Know More?
Everything I talked about in how I replaced 7 roles with AI tools is now running autonomously across this fleet. The video editing. The content scheduling. The email marketing. The client monitoring. All of it, 24/7, pinging me only when something needs my attention.
That’s not the future. That’s my Tuesday.
If you’re curious about OpenClaw, start here. If you’re already running it and want to compare setups, reach out. I want to hear what you’re building.
The AI Content Engine: 100+ Posts From One Idea
That content pipeline running on the $90 Optiplex? I'm breaking down the entire system live. Watch me use Claude Code to create a month of content from a single idea in 60 minutes. No fluff. Just the real system I use every day.
Under a grand, 5x return, and I’m just getting started. Convince me I’m wrong.