curl -fsSL https://deb.nodesource.com/setup_22.x | sudo -E bash - && sudo apt -y install nodejs
sudo npm i -g pm2
pm2 startup systemd # then run the line it prints
curl -fsSL https://pkg.cloudflare.com/cloudflared-main.gpg | sudo tee /usr/share/keyrings/cloudflared.gpg >/dev/null
echo "deb [signed-by=/usr/share/keyrings/cloudflared.gpg] https://pkg.cloudflare.com/cloudflared any main" | sudo tee /etc/apt/sources.list.d/cloudflared.list
sudo apt update && sudo apt -y install cloudflared
Local model4 · ollama tier online
curl -fsSL https://ollama.com/install.sh | sh && ollama pull qwen3:4b
Your three machines. Tap the key repeatedly the instant power hits — don't hold it.
ASUS NUC
14 Pro and the rest of the mini-PC fleet
Want
Key
BIOS setup
F2boot order, secure boot, memory info
Boot menu
F10pick the USB once, without touching boot order — the installer key
Network / PXE
F12rarely needed
Fast Boot + a USB keyboard = missed window. Already inside Ubuntu? Drop straight to firmware with systemctl reboot --firmware-setup.
Lenovo laptop
the Novo button is the reliable way in
Want
Key
Novo menu
Novothe pinhole/arrow button by the power key or on the side. Power off, press it (not power) → menu with BIOS Setup, Boot Menu, Recovery. Works when F-key timing fails.
BIOS setup
F2/Fn+F2ThinkPad: tap Enter at the logo, then F1
Boot menu
F12/Fn+F12
If F-keys do nothing, it's the FnLock — hold Fn while tapping, or just use the Novo button and forget the timing game entirely.
Raspberry Pi 5
no BIOS — it's all software and EEPROM
The Pi 5 has no firmware setup screen. There's nothing to press at power-on. Boot config lives in the EEPROM and on the SD card, so you change it from inside a running OS.
Power button
short press = on / graceful shutdown · hold = force offfirst Pi with a real power button — use it instead of yanking USB-C
Boot order
edit in the EEPROMUSB-before-SD, network boot, etc.
Bricked bootloader
flash recovery to a spare SD with Raspberry Pi Imager → Misc → Bootloader, boot it once
sudo raspi-config # Advanced → Boot Order, the friendly way
sudo rpi-eeprom-config --edit # or edit BOOT_ORDER directly
vcgencmd bootloader_version # check what firmware you're on
GRUB & the console
getting into a Linux box that won't behave — NUC or Pi
Want
How
Show GRUB
hold Shift (BIOS) or tap Esc (UEFI)
Edit an entry
e → change the linux line → Ctrl+x to boot it
Rescue shell
append systemd.unit=rescue.target to that line — root shell, no services
Switch TTY
Ctrl+Alt+F3a frozen desktop isn't a frozen kernel — this usually still logs in
Safe forced reboot
Alt+SysRq then R E I S U Bone letter a second — syncs discs first, always beats holding power
The commands that turn "it's broken" into a diagnosis. Everything here runs the same on the NUC, a Pi 5, or the Lenovo under Linux.
systemd
is it running, and why did it stop
systemctl status ollama # state, PID, last log lines
sudo systemctl restart ollama # bounce it
journalctl -u ollama -f # live tail its logs
systemctl list-units --failed # everything red on the box
Diagnosing a crash after the fact: journalctl -b -1 -p err shows errors from the previous boot only.
storage
the thing that actually killed the last install
Know the disks before you touch them
Run lsblk before every mount, dd, or install. No exceptions.
lsblk -f # disks, partitions, filesystems, labels
df -h # free space per mount — watch / and /var
sudo ncdu -x / # interactive "what's eating root" (apt install ncdu)
When / fills up
sudo journalctl --vacuum-size=200M # logs are usually the culprit
sudo apt clean
pm2 flush # PM2 logs grow forever untended
docker system prune # if docker's on here, likely this
Permanent PM2 log fix: pm2 install pm2-logrotate once, never think about it again.
Mount a foreign disk safely
Read-only first — you can't corrupt what you can't write to. This is the 1TB Pi-data drill.
lsblk -f # find it by size + label
sudo mkdir -p /mnt/x && sudo mount -o ro /dev/sdX1 /mnt/x
sudo mount -o remount,rw /mnt/x # only once you're sure
Drive health
sudo smartctl -a /dev/nvme0 | grep -Ei "percentage_used|temperature|critical"
NVMe percentage_used is wear — under 10% in a year is fine. SSDs die by wear, HDDs by reallocated sectors.
network
who's listening, who's reachable
Ports & processes
"Address already in use" and "connection refused" both start here.
ss -tulpn # every listening port + the process behind it
ss -tulpn | grep 11434 # who owns ollama's port
ip a # interfaces and addresses
Tailscale
tailscale status # every node + reachability
tailscale ping kay2nuc # direct path vs DERP relay
Stuck on "via DERP"? UDP's blocked somewhere — works, just slower. Direct is the goal.
processes
pm2, driven properly
pm2 ls # fleet at a glance — watch the ↺ column
pm2 logs clive --lines 50
pm2 restart clive && pm2 save # save = survives reboot
pm2 monit # live CPU/RAM per process
A climbing ↺ counter is a crash loop, not uptime. Read pm2 logs <name> --err for the cause — never just restart harder.
htop by keyboard
M sort by memory · P by CPU · F4 filter · F9 kill · t tree view
Load average = 1/5/15-min run-queue length. Sustained above nproc means CPU-bound; high load but low CPU means stuck I/O.
recovery
when it won't boot
The chroot rescue recipe
Boot the install USB → shell → become root inside the broken system. Fix bootloaders, clear a full disc, reset a password.
sudo mount /dev/nvme0n1p2 /mnt # root partition (lsblk to confirm)
sudo mount /dev/nvme0n1p1 /mnt/boot/efi # EFI partition
for d in dev proc sys run; do sudo mount --bind /$d /mnt/$d; done
sudo chroot /mnt # now "inside" the broken OS
From there: apt --fix-broken install, update-grub, or delete whatever filled the disk. Exit and reboot.
vitals
one-liners worth memorising
uptime # load at a glance
free -h # RAM — "available" is the number that matters
sensors # temps (apt install lm-sensors; run sensors-detect once)
last -x | head # recent logins, reboots, shutdowns