Why ?

I’ve had this M1 MacBook Air sitting on my desk for ages. 8GB RAM, perfectly fine little machine. Like most devs I barely touch it — the company laptop runs the show and the personal one mostly just exists.

Felt wasteful. So I plugged it into my work MacBook Pro over USB-C, folded both lids on a vertical stand, and turned the Air into a tiny homelab I never look at.

Setup

Two Macs on a vertical stand, lids closed. The Pro is the one I actually use (external display + keyboard). The Air sits next to it in clamshell mode. Single USB-C cable between them.

I never touch the M1 physically. Screen Sharing if I want the GUI, SSH for everything else. The keyboard might as well not exist. There’s something pleasing about a computer that just sits there and works.

Quick sanity check that both are actually on the network:

1
2
3
4
5

arp -a | grep -v incomplete | awk '{print $2, $1}' | column -t
# (192.168.178.23)   macbookair.fritz.box        ← the M1, over Wi-Fi
# (169.254.214.49)   kirans-macbook-pro.local    ← the MBP, over USB-C
# (192.168.178.1)    fritz.box
# ...

Two paths to the M1: Wi-Fi via the router and link-local over the USB-C cable. The cable is faster, the router is more stable across sleep. I use whichever responds.

How much faster is the cable? About 6×:

1
2
3
4
5

$ ping -c 10 169.254.165.143       # USB-C link-local
round-trip min/avg/max/stddev = 0.947/1.111/1.236/0.107 ms

$ ping -c 10 192.168.178.23        # Wi-Fi via the router
round-trip min/avg/max/stddev = 5.943/6.691/8.153/0.787 ms

Not life-changing for kubectl, but it matters when I’m pushing images or copying big files.

Plan

Right now it runs a kind cluster. That’s it. Maybe later:

• Move my Photos library onto it so my main disk stops crying.
• Run Ollama locally as a fallback in case Claude / Cursor decide to triple their pricing one Monday or have a bad outage day. A worse local model is still a model.
• Whatever else I think of. It’s more hack than need.

The cluster

colima + kind. Docker Desktop’s idle footprint is rude on 8GB so I skipped it.

1
2
3

brew install colima kind
colima start --cpu 4 --memory 5 --disk 50 --vm-type vz --vz-rosetta
kind create cluster --config kind-dev.yaml

Reaching it from the MBP

The fun bit. Kind binds the API server to 127.0.0.1 inside the VM, colima forwards it to 127.0.0.1 on the host. Cool. Useless if “the host” is a different laptop.

SSH tunnel works but you have to remember to start it. Skip. Instead: socat as a launchd agent on the M1, rebinding port 6443 onto 0.0.0.0, starting at login, restarting if it dies. Add the M1’s LAN IP to the kind cert SANs so the cert is happy.

1
2
3
4
5

# on the MBP
scp m1:~/kubeconfig-dev-for-mbp.yaml ~/.kube/kind-dev.yaml
kubectl --kubeconfig ~/.kube/kind-dev.yaml get nodes
# NAME                STATUS   ROLES           AGE   VERSION
# dev-control-plane   Ready    control-plane   2m    v1.35.0

Tradeoff: anyone on my home Wi-Fi who has my kubeconfig can hit the API. Risk: my apartment, my problem.

What’s actually running

I run glassflow — the thing I work on — without burning my MBP’s memory. Local testing, debugging, and the cluster I’ll throw at my Kubernetes exam prep all live here now.

From the MBP:

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19

$ kubectl config current-context
kind-dev

$ kubectl get nodes -o wide
NAME                STATUS   ROLES           AGE   VERSION   INTERNAL-IP   OS-IMAGE                         KERNEL              CONTAINER-RUNTIME
dev-control-plane   Ready    control-plane   65m   v1.35.0   172.18.0.2    Debian GNU/Linux 12 (bookworm)   6.8.0-100-generic   containerd://2.2.0

$ helm list -n glassflow
NAME       NAMESPACE  REVISION  STATUS    CHART                 APP VERSION
glassflow  glassflow  1         deployed  glassflow-etl-0.5.18  3.1.0

$ kubectl get pods -n glassflow
NAME                                            READY   STATUS    RESTARTS   AGE
glassflow-api-6df5dd58cf-5dcns                  1/1     Running   0          26m
glassflow-controller-manager-5494bb5f7c-bcpsl   1/1     Running   0          26m
glassflow-nats-0                                2/2     Running   0          26m
glassflow-nats-box-7dbfc4f47f-w7ckt             1/1     Running   0          26m
glassflow-postgresql-0                          1/1     Running   0          26m
glassflow-ui-5db85889b9-zdwgt                   2/2     Running   0          34m

API, controller, NATS, Postgres, UI — six real pods doing real work on a laptop I can’t see from where I’m sitting. Packets leave the MBP, cross the USB-C cable, get NAT’d into a Linux VM, land in containers on the Air, come back. A small Rube Goldberg and I love it.

Result: a kube cluster I never have to look at 🚀

Next

1. Photos library on the M1, off my main disk.
2. Ollama, just in case.
3. Whatever else looks like fun.

Zero ROI, full nerd points. I’ll take it.