Saving $4000+/month with self-hosted runners

For the longest time, Sidecar didn’t have any continuous integration 😭.

Sidecar does have tests, but given the scale of the codebase I was looking at a potential bill in the thousands of dollars per month just to run continuous integration on GitHub. The more tests I wanted to write, the bigger a cost I was looking at for CI. This was like having to pay $1 every time I wanted to do a push-up; the incentive structure is backwards.

#

The cost of testing Sidecar

Sidecar takes about 10 minutes to build on a decently fast machine, and with almost 1,100 Swift packages in the dependency graph that also have their own tests, the CI minutes add up quick.

#

The cost of GitHub macOS runners

GitHub’s free tier gives you 3000 minutes per month, but macOS minutes have a 10x multiplier.

That’s 300 actual macOS minutes, or about 5 hours per month.

Doing some simple math: building Sidecar alone, plus a handful of key test suites uses about 30-35 minutes, would eat up the free tier after just 10 pull requests 😭. Being a cost-conscious bootsrapped founder, I ended up just running tests manually and hoped I didn’t miss anything (narrator: he did).

But then I learned about self-hosted runners.

It’s now been about a month with my own build cluster, and the numbers are wild: I’m saving about $4000 per month (~$45,000 annually).

Let me walk you through the math, compare against the alternatives, and show you how to set this up yourself.

---

#

Sidecar in build metrics

From October 1-17, 2025, the Clutch Engineering macOS runners processed 28,723 minutes of build time across two main workloads:

• Valhalla builds: 11,393 minutes (40% of total)

• Sidecar CI: 17,330 minutes (60% of total)

If I were running these on GitHub’s hosted macOS runners, here’s what it would cost:

#

Provider costs

GitHub charges $0.08 per minute for standard macOS runners (or $4.80/hour). The free tier gives you 300 actual macOS minutes per month (3,000 minutes with the 10x multiplier). Some quick math:

28,723 minutes - 300 free minutes = 28,423 billable minutes
28,423 minutes x $0.08 = $2,273.84

For just 17 days. Extrapolate that to a full 31-day month:

((28,723 x 31/17) - 300) x $0.08 ≈ $4,167 per month

It’s also important to note that the base machines are just 3-core M1s, so the build times I quote above (and below) are likely much longer.

I looked at other macOS CI providers to see if there was a better option. Here’s how they compare for Clutch Engineering’s monthly usage (~52,377 minutes/month):

† Xcode Cloud pricing is per-hour, not per-minute; equivalent to ~$0.08/min for comparison.

‡ Xcode Cloud offers tiered pricing; 1,000 hours/month at $399.99 is the best option for this usage. The problem is that Xcode Cloud is unreasonably slow (30+ minutes to build Sidecar).

§ CircleCI offers 30,000 free credits/month, but macOS runners use these credits super fast.

¶ Bitrise offers unlimited build minutes but limits concurrent builds; Teams allows 3 concurrent, Pro allows 10. With 2,774 workflow runs in 17 days, queue times become significant.

♱ MacStadium rents individual Mac machines; pricing shown for 2x M4.M Mac mini (10 Core, 24GB) at $199/month each to match current setup.

★ Hardware amortization + power costs only; machines are owned, not rented.

#

Renting Macs

The closest equivalent to self-hosting is renting Mac hardware from MacStadium. At $398/month for two M4 Mac minis (matching the setup), it’s significantly cheaper than most CI services. But here’s the math on ownership:

MacStadium rental:

$398/month × 12 months = $4,776/year

Purchase equivalent hardware:

Mac mini M4 16GB: $599
Mac mini M4 Pro 24GB: $1,399
Total: $1,998

MacStadium’s rental cost exceeds the purchase price in just 5 months. After that, you’re paying $398/month for hardware you could have owned.

Over 3 years:

• MacStadium: $14,328

• Self-hosted: $1,998 (hardware) + $252 (power) = $2,250

That’s $12,078 in savings just by owning the hardware instead of renting it. And you still own the machines at the end.

The only real advantages of MacStadium are:

1. No upfront capital expenditure

2. Someone else handles hardware failures

3. Easier to scale up/down on demand

But for a stable workload like mine, those advantages don’t justify paying 6× more over three years.

#

The solution: self-hosted runners

Here’s my current setup:

• 1x MacBook Pro M1 Max 32GB (2021, repurposed laptop)

• 1x Mac mini M4 Pro 24GB (2024, purchased for Clutch Engineering)

• 1x Mac mini M4 16GB (2024, purchased for Clutch Engineering)

#

The Real Costs

Hardware:

• Mac mini M4 16GB: $599 (base model)

• Mac mini M4 Pro 24GB: ~$1,399

• MacBook Pro M1 Max: Already owned, sunk cost

Power: At ~10W idle and ~40W under load per Mac mini, averaging 25W each over a month:

25W x 2 minis x 24h x 31 days = 37.2 kWh
37.2 kWh x $0.26/kWh ≈ ~$10 per month

The MacBook Pro adds maybe another $3/month.

Internet: I already have business internet for other services, so this is effectively $0 marginal cost.

Amortization: Spread the Mac mini costs over 3 years:

($599 + $1,399) / 36 months ≈ $55.50 per month

Total recurring cost: ~$64/month (Mac mini amortization + power for all three machines)

Compare that to the $4,191/month for GitHub, or even the $628/month for CircleCI’s 4 CPU runners.

#

Multiple runners per machine

One optimization worth mentioning: you can run multiple GitHub Actions runner instances on a single machine. The Clutch Engineering build cluster runs 3-5 runners per machine, which means those three physical machines provide 12+ concurrent build slots. This is super useful for shorter test suites that don’t max out the CPU; you can run multiple tests in parallel on the same hardware.

#

Maintenance cost

Setup took me about an hour for the new Mac Mini. Maintenance has been essentially zero; these machines just work. The GitHub Actions runner software auto-updates, macOS is stable, and builds are reliable.

Even if you value your time at $200/hour, that’s a one-time $400 cost that pays for itself in the first month.

---

#

How to set it up

Setup is hella easy. Like, shockingly easy.

#

Step 1: Prepare your Mac

1. Install the latest macOS

2. Enable automatic login (Settings > Users & Groups)

3. Configure energy settings to prevent sleep

4. Install Xcode and any other build dependencies you need

#

Step 2: Register the runner

1. Go to your organization’s settings: https://github.com/organizations/<YourOrganization>/settings/actions/runners/new

2. Select macOS as the runner type

3. Follow the download and configuration instructions

The commands look something like this:

# Create a folder for the runner
mkdir actions-runner && cd actions-runner

# Download and install the latest runner
# (see the GitHub instructions for the latest commands)

# Configure the runner
./config.sh --url https://github.com/<YourOrg> --token <YourToken>

# Install and start the service
./svc.sh install
./svc.sh start

#

Step 3: Configure Your workflows

Update your workflow files to use your self-hosted runner:

jobs:
  build:
    runs-on: self-hosted
    # or be more specific:
    runs-on: [self-hosted, macOS]

#

Step 4: Monitor and maintain

GitHub provides a dashboard showing runner status, job queue, and utilization. You can access this at:

https://github.com/organizations/<YourOrganization>/settings/actions/runners

The runners auto-update themselves, so maintenance is minimal.

#

Security considerations

A few important notes on security:

1. Network isolation: Keep your runners on a separate VLAN if possible.

2. No public repos: Only use self-hosted runners for private repositories.

GitHub’s documentation has more details: About self-hosted runners

#

Verdict

For the Clutch Engineering use case (lots of iOS builds across multiple projects) self-hosted runners are a no-brainer:

• Savings: $2,000-$4,000+ per month

• Setup time: ~2 hours one-time

• Maintenance: Effectively zero

• Performance: Faster than GitHub’s hosted runners (no boot time, cached dependencies)

• Flexibility: Full control over environment, installed tools, and configurations

But the most important result of this: costs are no longer directly associated with the scale of testing and deploying Sidecar. This is a huge psychological win.

If you need any amount of non-trivial macOS CI workload, the math is clear: buying even one Mac Mini can save you tens of thousands of dollars. 🥳

---