What I run on my $10 USD OpenStick, Linux, Homelab

Hacker News by 6 min read 504x views
What I run on my $10 USD OpenStick, Linux, Homelab

Share Post

A during ago I bought an OpenStick on a whim for 10 dollars (including shipping!) since it seemed akin specified a chill novelty part of hardware.

For 10 dollars you get:

  • 512 MB of RAM (~450 MB usable)
  • 4x ARM Cortex-A53 CPU cores at 1 GHz
  • 4 GB eMMC flash
  • 150 Mbit/s Wi-Fi
  • 4G modem and GPS
  • Passively cooled, so 100% silent!
  • 1 programmable LED :)

A pale 4G LTE Wi-Fi modem rod alongside a USB-A plug next to a dreary USB hub and a pen for scale

Now, you power appearance at this hardware and think “my observe is additional mighty than that”, and you wouldn’t necessarily be wrong, but there’s motionless a lot of fun to be had alongside a equipment this size, and it’s a awesome gadget to get started alongside self-hosting.

The facility procedure is a bit involved, but formerly you’re up and operating it runs Debian, which brings all the traditional goodies specified as systemd (among another things, a scheduler and assistance monitoring for Linux) and Podman (lightweight containerization). For 10 USD you basically get a residence server alongside minimal power use and zero noise!

There’s really a lot of helpful application that runs awesome on this thing. Some of my favorites are:

This is a small assistance that connects to any operating Google Cast equipment and skips sponsorships (and sometimes ads). It barely uses any resources, and following a during you ignore that it exists, but it does create the entire YouTube cognition a lot nicer!

CastSponsorSkip

AdGuard Home - my own fork (40 MB)

AdGuard Home blocks bad domains at the DNS level: spyware, malware, and so on. These days devices have a inclination to spy on you. Smart TVs run satisfied acknowledgment continuously, washing machines scan your network, and your coffee maker power be spying on you.

A DNS blocker cannot obstacle all bad requests (you’ll need a appropriate firewall for that), but it certain can obstacle a entire lot. After adding a few blocklists, 60% of my queries are now blocked. Most appear harmless, but I’m beautiful certain that several aren’t.

I’m operating my own fork, which can grip ample blocklists in an productive manner, but you likely shouldn’t run binaries provided by random strangers, so I’ve additionally included the nexus to the authoritative type below (which is totally fine if you bounds yourself to smaller blocklists).

AdGuard Home Next-Gen - 40 MB (independent of the size of the blocklist) AdGuard Home - 40 MB (with a small (200k) blocklist)

AdGuard Home dashboard showing 12,536 DNS queries, 8,091 blocked ads and top clients

Zeroclaw (64 MB)

Zeroclaw is an AI delegate harness. It supports 30 channels (Telegram, email, WhatsApp, etc.), has sensible safety defaults, a nice web UI, and integrations alongside all the important LLM providers.

I ran OpenClaw for a during on my Kubernetes cluster, but it mostly caused a lot of frustration: it worked, but it used 1 GB+ of memory, bankrupt alongside all update, and they kept changing functionality in ways that bankrupt my existing workflows. Zeroclaw feels akin a respiration of caller air. It uses ~3 MB RSS whenever idle, covers all the functionality that I need, and the safety defaults are sensible and uncomplicated adequate to disable. The next type volition (hopefully!) have affluent communication assistance for Telegram, implemented by yours truly. The PR can be tracked here!

I’ve stated Zeroclaw admission to the OpenStick (but nothing beyond that). Effectively it’s a base shell, but stated that the blast radius is limited, I figured it would be acceptable. I mostly use this to detect the OpenStick, alongside questions specified as “anything in the logs that requires my attention?” or “how is the disk space?”. The biggest advantage is that it communicates through Telegram, so if I’m in the mountains alongside constricted signal, I can motionless ask it how my small server is doing :)

I looked at alternatives specified as Picoclaw and Nullclaw, and no doubt they’re fine as well. However, Zeroclaw seemed to have the finest records and the most energetic community, so I went alongside that :)

GitHub Zeroclaw Documentation

Zeroclaw dashboard showing uptime, RAM usage, disbursal overview and the energetic Telegram channel

Cloudflare Tunnel (~8-20 MB)

Cloudflared makes it imaginable to disclose all these services securely to the big, bad internet, for free, without exposing any ports. The downside power be that the NSA power be capable to peruse everything you do, but for a homelab I regard that to be an satisfactory trade-off.

Cloudflare Tunnel

Beszel (20 MB)

Beszel shows a beautiful chart of memory, CPU usage, and so on, and alerts you if item looks suspicious, all during using extremely small memory.

I ran Grafana and the entire stack on my Kubernetes collection for a while, and it’s powerful, but not extremely person friendly, and it felt akin overkill.

Beszel is a really notable solution: it looks beautiful and requires ~20 MB, 10 MB for the delegate and 10 MB for the server. It has built-in alerting for a entire bunch of protocols, and I let it dispatch messages to Telegram whenever item requires my attention.

GitHub Beszel

Beszel scheme leaf for the OpenStick showing CPU, Podman CPU, recollection and Podman recollection use complete 24 hours

Dozzle

Dozzle is a beautiful interface for managing Podman containers. It has a ample overlap alongside Beszel, so if I always need to preserve 10 or 20 MB, I’ll get rid of one of them. Dozzle has a way improved interface for checking Podman logs, but it doesn’t recommendation the identical flat of stats as Beszel does.

GitHub Dozzle

Dozzle receptacle catalog showing eight operating containers on openstick-debian alongside CPU and recollection per container

Gatus (10 MB)

Beszel monitors your server(s). Gatus monitors your services. In another words, if you would akin to get a notification whenever your webserver or repository has crashed, you’ll desire Gatus. It additionally creates a really beautiful web interface!

GitHub Gatus - 10 MB

Gatus position leaf alongside uptime bars per endpoint for AdGuard Home, Beszel, Dozzle, the Cloudflare tunnel and more

RIP Google Reader. I tried Feedly for a while, but 7 USD for the Pro type was fair a small bit additional than I felt comfortable with. Besides, why pay whenever you can have hours of fun environment up your own alternative?

I tried a entire bunch of RSS readers and eventually settled on Fusion. It does all I need, looks good, feels responsive, and uses 22 MB of RAM.

GitHub Fusion - 22 MB

Fusion's three-pane audience showing 63 unread articles in the all-articles list

Shiori (~10–20 MB)

Browser bookmarks are great, nevertheless a lot of my older bookmarks have stopped operating since the sites disappeared, the URLs changed, or for a myriad of another reasons. Shiori fixes this by offering a awesome bookmark director that additionally automatically archives the website. It does so in ~10 MB. You can anticipate several spikes whenever archiving, but nothing that our mighty OpenStick cannot handle!

Shiori - 10 MB

Diun (9 MB)

Just a small assistance that sends a communication to my Telegram whenever one of my containers has an update available.

Diun - 9 MB

Other Article Hacker News
↑
Close Right Ads
Close Left Ads