<?xml version="1.0" encoding="UTF-8"?>
<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom" xmlns:dc="http://purl.org/dc/elements/1.1/">
  <channel>
    <title>DEV Community: Jonne Riepponen</title>
    <description>The latest articles on DEV Community by Jonne Riepponen (@jronegit).</description>
    <link>https://dev.to/jronegit</link>
    <image>
      <url>https://media2.dev.to/dynamic/image/width=90,height=90,fit=cover,gravity=auto,format=auto/https:%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Fuser%2Fprofile_image%2F4163710%2F26ffbc3d-0094-44d1-9f8b-30737c1f6aa4.png</url>
      <title>DEV Community: Jonne Riepponen</title>
      <link>https://dev.to/jronegit</link>
    </image>
    <atom:link rel="self" type="application/rss+xml" href="https://dev.to/feed/jronegit"/>
    <language>en</language>
    <item>
      <title>I Was the Only IT Person for 3 Years: The Documentation I Wish I'd Written</title>
      <dc:creator>Jonne Riepponen</dc:creator>
      <pubDate>Sat, 14 Nov 2026 15:15:12 +0000</pubDate>
      <link>https://dev.to/jronegit/i-was-the-only-it-person-for-3-years-the-documentation-i-wish-id-written-13bg</link>
      <guid>https://dev.to/jronegit/i-was-the-only-it-person-for-3-years-the-documentation-i-wish-id-written-13bg</guid>
      <description>&lt;h2&gt;
  
  
  The Call I Didn't Want to Get
&lt;/h2&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Ftf530orew881775f4i7p.png" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Ftf530orew881775f4i7p.png" alt="The Call I Didn't Want to Get" width="800" height="457"&gt;&lt;/a&gt;&lt;br&gt;
Three weeks after leaving my job as the only IT person at a mid-sized company, my phone rang. It was Mike, the guy they hired to replace me.&lt;/p&gt;

&lt;p&gt;"Hey, uh, do you remember that backup script you set up? The one that runs on Sundays? It's throwing an error and I can't find where it's configured."&lt;/p&gt;

&lt;p&gt;I helped him. Then he called again the next day about the VPN. Then about the firewall rules. Then about why the CEO's email kept going to spam.&lt;/p&gt;

&lt;p&gt;By the end of month one, Mike had called me 47 times.&lt;/p&gt;

&lt;p&gt;Here's the thing: I thought I had done a good job documenting everything. I had wiki pages. I had a SharePoint folder full of PDFs. I had even written a "handover document" before I left.&lt;/p&gt;

&lt;p&gt;But none of it was useful because I had documented the wrong things in the wrong way.&lt;/p&gt;

&lt;p&gt;The Invisible Knowledge Problem&lt;br&gt;
When you're the only IT person, you become a walking encyclopedia of institutional knowledge that exists nowhere else:&lt;/p&gt;

&lt;p&gt;You know that the server in the closet makes a weird noise every Tuesday at 3 AM (it's the backup tape drive, and it's fine)&lt;br&gt;
You know that the CFO's laptop needs special VPN settings because of some legacy accounting software&lt;br&gt;
You know that the "critical" production database is actually just a reporting copy and can go down for maintenance&lt;br&gt;
You know where the network cables behind the bookshelf go (even though they're not on any diagram)&lt;br&gt;
None of this is in your documentation. It's all in your head.&lt;/p&gt;

&lt;p&gt;And when you leave - whether for a new job, a vacation, or because you got hit by a bus - that knowledge leaves with you.&lt;/p&gt;

&lt;p&gt;The Bus Factor Quiz&lt;br&gt;
Quick test: If you got hit by a bus tomorrow, could someone else:&lt;/p&gt;

&lt;p&gt;Access your password manager and all your accounts?&lt;br&gt;
Find the recovery codes for your two-factor authentication?&lt;br&gt;
Know which servers are critical vs. which can go down?&lt;br&gt;
Understand what that weird cron job does at 2 AM?&lt;br&gt;
Contact your vendors and actually get help?&lt;br&gt;
If you answered "no" or "maybe" to any of these, you have a bus factor problem. And in solo IT, your bus factor is always 1.&lt;/p&gt;

&lt;p&gt;Why Most IT Documentation Fails&lt;br&gt;
I used to think documentation meant writing everything down. So I did:&lt;/p&gt;

&lt;p&gt;47-page network diagram (outdated within a month)&lt;br&gt;
Change management procedures (nobody followed them)&lt;br&gt;
Server inventory spreadsheet (missing half the VMs)&lt;br&gt;
Disaster recovery plan (never tested, probably wouldn't work)&lt;br&gt;
The problem? I was documenting for auditors, not for the poor soul who would eventually need to actually run things.&lt;/p&gt;

&lt;p&gt;What Actually Works&lt;br&gt;
After Mike's 47 calls, I started keeping track of what he actually needed to know. Here's what came up most often:&lt;/p&gt;

&lt;p&gt;The "Why" Behind Decisions&lt;/p&gt;

&lt;p&gt;Why is this server configured this way? (Usually because of some edge case nobody remembers)&lt;br&gt;
Why do we have two different backup systems? (The old one is for legal compliance, the new one is for actual recovery)&lt;br&gt;
Why does the CEO have local admin rights? (Don't ask, just don't take them away)&lt;br&gt;
The Hidden Dependencies&lt;/p&gt;

&lt;p&gt;This server depends on that one, which depends on a service running on a third one&lt;br&gt;
The WiFi in the conference room breaks if you restart the main switch in the wrong order&lt;br&gt;
The CRM integration breaks every time the accounting software updates&lt;br&gt;
The "Normal" Abnormalities&lt;/p&gt;

&lt;p&gt;Yes, that error message appears every day and it's fine&lt;br&gt;
No, that server doesn't need more RAM, it's just badly configured&lt;br&gt;
The warning in the logs is expected behavior, we've been ignoring it for two years&lt;br&gt;
The Documentation Template That Would Have Saved Me 47 Phone Calls&lt;br&gt;
Here's what I should have left behind. This is a template you can copy and fill in today - not when you're about to leave, but right now, while you still remember why you configured things the way you did.&lt;/p&gt;

&lt;p&gt;Section 1: The "If I Get Hit By a Bus" Page&lt;br&gt;
This is a single document that lives somewhere obvious. Not in a wiki that requires login. Not in a folder buried five levels deep. Print it. Put it on the server room door.&lt;/p&gt;

&lt;p&gt;CRITICAL INFORMATION - START HERE&lt;br&gt;
Master Passwords&lt;br&gt;
Password manager master password: [Location of sealed envelope or key escrow]&lt;br&gt;
Domain admin account: [Account name and where credentials are stored]&lt;br&gt;
Cloud console access: [Which email address receives MFA codes]&lt;br&gt;
Critical Contacts&lt;br&gt;
Internet provider support: 1-XXX-XXX-XXXX, Account #XXXXX&lt;br&gt;
Hardware vendor support: 1-XXX-XXX-XXXX, Contract #XXXXX&lt;br&gt;
MSP or consultant (if any): Name, phone, email&lt;br&gt;
CEO's personal cell (for emergencies only): XXX-XXX-XXXX&lt;br&gt;
The "Don't Touch" List&lt;br&gt;
These things will break if you change them:&lt;/p&gt;

&lt;p&gt;[Server/service] - Reason: [Why]&lt;br&gt;
[Configuration] - Reason: [Why]&lt;br&gt;
Known Issues We're Living With&lt;br&gt;
[Issue] - Why we haven't fixed it: [Reason] - Impact: [What happens]&lt;br&gt;
Section 2: The Services Inventory&lt;br&gt;
Not just "what servers do we have" but "what services exist and why do they matter?"&lt;/p&gt;

&lt;p&gt;Services Inventory&lt;br&gt;
Tier 1 - Critical (Down = Business Stopped)&lt;br&gt;
Service Server  What It Does    Who Uses It     Restart Procedure&lt;br&gt;
Main database   PROD-SQL-01     All customer data       Everyone        See Runbook #3&lt;br&gt;
File server     FS-01   All company files       Everyone        Auto-restart, check shares after&lt;br&gt;
Email   Office 365      Email   Everyone        Microsoft handles it&lt;br&gt;
Tier 2 - Important (Down = Some People Can't Work)&lt;br&gt;
Service Server  What It Does    Who Uses It     Restart Procedure&lt;br&gt;
Tier 3 - Nice to Have (Down = Annoying But Not Urgent)&lt;br&gt;
Service Server  What It Does    Who Uses It     Restart Procedure&lt;br&gt;
Tier 4 - We Should Probably Turn This Off&lt;br&gt;
Service Server  Why It Still Exists     Can We Delete It?&lt;br&gt;
Old CRM LEGACY-01       Legal requires 7 years  Delete after 2027&lt;br&gt;
Section 3: The "Why" Document&lt;br&gt;
For every non-obvious configuration decision:&lt;/p&gt;

&lt;p&gt;Configuration Decisions &amp;amp; Rationale&lt;br&gt;
Why Does the VPN Require Two-Factor Auth for Some Users But Not Others?&lt;br&gt;
Reason: Executives complained about the extra step. Yes, this is a security risk. No, I couldn't convince them otherwise. The CFO specifically requested exemption.&lt;/p&gt;

&lt;p&gt;Date of Decision: March 2024Who Approved: CFO, CEOCan This Be Changed? Only with CFO approval&lt;/p&gt;

&lt;p&gt;Why Is the Backup Server Running an Old OS Version?&lt;br&gt;
Reason: The backup software vendor hasn't certified the new OS version yet. We tried upgrading in test and it broke the backup catalog.&lt;/p&gt;

&lt;p&gt;Date of Decision: January 2024Workaround: Applied all security patches that don't affect the backup softwareRevisit Date: Check vendor website monthly&lt;/p&gt;

&lt;p&gt;Why Does the CEO Have Local Admin Rights?&lt;br&gt;
Reason: He installs software for "productivity" and refuses to wait for IT. We tried restricting it. He complained to the board.&lt;/p&gt;

&lt;p&gt;Date of Decision: Before my timeMitigation: His laptop has extra monitoring and we image it monthlyCan This Be Changed? LOL no&lt;/p&gt;

&lt;p&gt;Section 4: The Dependency Map&lt;br&gt;
Service Dependencies&lt;br&gt;
If Server X Goes Down, What Breaks?&lt;br&gt;
PROD-WEB-01 (Main Web Server)├── Requires: PROD-SQL-01 (Database)├── Requires: REDIS-01 (Cache)├── Requires: NFS-01 (File uploads)└── If down: External customers cannot access the website&lt;/p&gt;

&lt;p&gt;PROD-SQL-01 (Main Database)├── Requires: BACKUP-01 (Backup jobs)├── Requires: MON-01 (Monitoring)└── If down: EVERYTHING breaks (see affected services list)&lt;/p&gt;

&lt;p&gt;Startup Order (After Power Outage)&lt;br&gt;
Wait 5 minutes for network switches&lt;br&gt;
Start PROD-DC-01 (Domain Controller)&lt;br&gt;
Start PROD-SQL-01 (Database)&lt;br&gt;
Start REDIS-01 (Cache)&lt;br&gt;
Start PROD-WEB-01 (Web Server)&lt;br&gt;
Verify services with checklist in Runbook #5&lt;br&gt;
Section 5: The Vendor &amp;amp; License Tracker&lt;br&gt;
Vendors &amp;amp; Licenses&lt;br&gt;
Software Licenses&lt;br&gt;
Software        License Type    Renewal Date    Admin Contact   License Key Location&lt;br&gt;
Microsoft 365   50 seats        Annual, March   &lt;a href="mailto:admin@company.com"&gt;admin@company.com&lt;/a&gt;       Admin portal&lt;br&gt;
Backup Software Perpetual       N/A     N/A     License.txt on BACKUP-01&lt;br&gt;
Antivirus       100 seats       Annual, June    &lt;a href="mailto:vendor@support.com"&gt;vendor@support.com&lt;/a&gt;      Email from 2023&lt;br&gt;
Hardware Support&lt;br&gt;
Device  Serial Number   Support Expires Vendor Support #&lt;br&gt;
Server PROD-01  SN123456        2025-12-31      1-800-XXX-XXXX&lt;br&gt;
SAN Storage     SN789012        2024-06-30 (EXPIRING!)  1-800-XXX-XXXX&lt;br&gt;
Cloud Services&lt;br&gt;
Service Account Billing Contact MFA Recovery&lt;br&gt;
AWS     &lt;a href="mailto:root@company.com"&gt;root@company.com&lt;/a&gt;        CFO     Security team email&lt;br&gt;
Azure   &lt;a href="mailto:admin@company.com"&gt;admin@company.com&lt;/a&gt;       IT      Backup codes in safe&lt;br&gt;
How to Actually Take a Vacation&lt;br&gt;
The real test of your documentation isn't whether someone can replace you - it's whether you can disappear for two weeks without your phone ringing.&lt;/p&gt;

&lt;p&gt;The Pre-Vacation Checklist&lt;br&gt;
Two weeks before:&lt;/p&gt;

&lt;p&gt;Identify who will cover while you're gone (even if it's just "call this MSP")&lt;br&gt;
 Walk them through the critical systems&lt;br&gt;
 Test that they can actually access everything they need&lt;br&gt;
 Update the "If I Get Hit By a Bus" document&lt;br&gt;
 Schedule any risky changes for after you return&lt;br&gt;
One week before:&lt;/p&gt;

&lt;p&gt;No new changes. Period. Even "small" ones.&lt;br&gt;
 Verify backups are running and test a restore&lt;br&gt;
 Check disk space on all critical systems&lt;br&gt;
 Review monitoring alerts - are there any that might trigger?&lt;br&gt;
Before you leave:&lt;/p&gt;

&lt;p&gt;Set email auto-reply with contact info for coverage&lt;br&gt;
 Forward critical alerts to your backup person&lt;br&gt;
 Leave your phone number for absolute emergencies only&lt;br&gt;
 Define what counts as an emergency (server down = yes, printer jam = no)&lt;br&gt;
The Coverage Handoff Document&lt;br&gt;
Vacation Coverage Handoff&lt;br&gt;
I Will Be Gone: [Dates]&lt;br&gt;
Who to Contact (In Order)&lt;br&gt;
[Name/Company] - First line of defense&lt;br&gt;
[Name] - For decisions above [First contact]'s level&lt;br&gt;
Me - For absolute emergencies only: [Phone number]&lt;br&gt;
What I've Already Done&lt;br&gt;
 Checked disk space (all OK)&lt;br&gt;
 Verified backups (restored test file successfully)&lt;br&gt;
 Patched critical servers (reboots done)&lt;br&gt;
 No changes scheduled during my absence&lt;br&gt;
Things That Might Break (And What to Do)&lt;br&gt;
If This Happens Do This Call Me If&lt;br&gt;
Website down    Restart PROD-WEB-01, see Runbook #2     Not back up in 15 min&lt;br&gt;
Email not sending       Check Microsoft 365 status page Outage lasts &amp;gt; 1 hour&lt;br&gt;
Can't access files      Check FS-01, restart if needed  Files missing/corrupted&lt;br&gt;
Things That Will Definitely Happen (And Are Fine)&lt;br&gt;
Backup alerts on Sunday night (normal, check logs)&lt;br&gt;
CEO's laptop "slow" (clear browser cache)&lt;br&gt;
Printer jam in accounting (turn it off and on again)&lt;br&gt;
The Quarterly "Knowledge Transfer" Meeting&lt;br&gt;
Even if you're the only IT person, you should have a quarterly meeting with someone - your manager, an MSP, anyone who might need to step in.&lt;/p&gt;

&lt;p&gt;Agenda Template&lt;br&gt;
What's changed since last quarter? (New services, retired services, major config changes)&lt;br&gt;
What's keeping you up at night? (Risks, aging hardware, upcoming expirations)&lt;br&gt;
Review the "Bus Factor" document (Is everything still accurate? Any new passwords?)&lt;br&gt;
Walk through one disaster scenario ("If the main server died right now, what would we do?")&lt;br&gt;
Budget/Project updates (What do you need? What's coming up?)&lt;br&gt;
This meeting does two things: it forces you to keep documentation current, and it ensures someone else has context when things go wrong.&lt;/p&gt;

&lt;p&gt;Start Today, Not When You're Leaving&lt;br&gt;
The best time to write this documentation isn't your last week on the job. It's right now, while the context is fresh in your mind.&lt;/p&gt;

&lt;p&gt;Here's your homework:&lt;/p&gt;

&lt;p&gt;This week: Fill out the "If I Get Hit By a Bus" document. Print it. Put it somewhere accessible.&lt;br&gt;
Next week: Start the Services Inventory. Just do Tier 1 (critical systems).&lt;br&gt;
The week after: Add one entry to the "Why" document every time you make a non-obvious configuration change.&lt;br&gt;
This month: Schedule your first quarterly knowledge transfer meeting.&lt;br&gt;
Your future self will thank you. And the next person in your role? They'll thank you even more.&lt;/p&gt;

&lt;p&gt;Pro tip: If you want to practice documentation before the pressure is on, check out the Incident Report Generator - it turns your raw notes into professional documentation. Because sometimes you need corporate language to get management to actually pay attention.&lt;/p&gt;

&lt;p&gt;&lt;em&gt;Related reads:&lt;/em&gt;&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;&lt;em&gt;&lt;a href="https://pragmaticsysadmin.help/sysadmin/2026-03-26-the-mistakes-i-forgot-to-renew-domain-and-other-crimes-against-it/" rel="noopener noreferrer"&gt;The Mistakes I Made (And Why They Led Me to Build My Own Password Manager)&lt;/a&gt;&lt;/em&gt;&lt;/li&gt;
&lt;li&gt;&lt;em&gt;&lt;a href="https://pragmaticsysadmin.help/sysadmin/stop-doing-things-manually/" rel="noopener noreferrer"&gt;Stop Doing Things Manually: 5 Scripts That'll Make You Look Like a Genius&lt;/a&gt;&lt;/em&gt;&lt;/li&gt;
&lt;li&gt;&lt;em&gt;&lt;a href="https://pragmaticsysadmin.help/sysadmin/2025-12-09-the-5-minute-server-health-check-that-could-save-your-career/" rel="noopener noreferrer"&gt;The 5-Minute Server Health Check That Could Save Your Career&lt;/a&gt;&lt;/em&gt;&lt;/li&gt;
&lt;/ul&gt;

</description>
      <category>documentation</category>
      <category>sysadmin</category>
      <category>career</category>
      <category>knowledgemanagement</category>
    </item>
    <item>
      <title>I Built My Own J.A.R.V.I.S. — A Local, Privacy-First Multi-Agent AI Assistant</title>
      <dc:creator>Jonne Riepponen</dc:creator>
      <pubDate>Sat, 14 Nov 2026 15:14:37 +0000</pubDate>
      <link>https://dev.to/jronegit/i-built-my-own-jarvis-a-local-privacy-first-multi-agent-ai-assistant-276e</link>
      <guid>https://dev.to/jronegit/i-built-my-own-jarvis-a-local-privacy-first-multi-agent-ai-assistant-276e</guid>
      <description>&lt;p&gt;I wanted a personal AI assistant. Not a chatbot. Not a copilot. Something that actually &lt;em&gt;does things&lt;/em&gt; — controls my desktop, reads my screen, talks to my smart home, answers questions in Finnish or English, and never, ever sends my data to a server I don't control.&lt;/p&gt;

&lt;p&gt;The cloud options are good. They're not &lt;em&gt;mine&lt;/em&gt;.&lt;/p&gt;

&lt;p&gt;So I started building one.&lt;/p&gt;

&lt;p&gt;This post is the architecture document for that project. It's a work in progress — the docker-compose is ready, the Ollama instance is running, and the first agent is next. I'm writing this because the design decisions are interesting, and because someone else might want to build the same thing.&lt;/p&gt;

&lt;h2&gt;
  
  
  What J.A.R.V.I.S. Is (And Isn't)
&lt;/h2&gt;

&lt;p&gt;This isn't a Jarvis like Tony Stark's. It's closer to a privacy-respecting home automation brain — or what you'd get if you crossed a really good CLI assistant with a smart home hub and gave it eyes.&lt;/p&gt;

&lt;p&gt;The core idea: a lightweight orchestrator (Ollama) decides &lt;em&gt;what to do&lt;/em&gt;. Specialized agents handle &lt;em&gt;how to do it&lt;/em&gt;. Each agent runs in its own Docker container. Everything stays on my local network.&lt;/p&gt;

&lt;p&gt;What it does:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;Controls my Linux desktop (volume, brightness, window management)&lt;/li&gt;
&lt;li&gt;Reads my screen when I ask it to&lt;/li&gt;
&lt;li&gt;Talks to Home Assistant (lights, sensors, automations)&lt;/li&gt;
&lt;li&gt;Answers questions, explains code, translates Finnish ↔ English&lt;/li&gt;
&lt;li&gt;Runs entirely locally on consumer hardware&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;What it doesn't do: send my commands, my screen contents, or my voice to OpenAI, Google, or anyone else.&lt;/p&gt;

&lt;h2&gt;
  
  
  The Architecture
&lt;/h2&gt;



&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight plaintext"&gt;&lt;code&gt;                      ┌─────────────────────────────────┐
                      │     Input / Output Layer         │
                      │  (Whisper STT · Piper TTS)     │
                      └──────────────┬──────────────────┘
                                     │
                                     v
                      ┌─────────────────────────────────┐
                      │        OLLAMA (The Brain)       │
                      │   qwen2.5:3b · Function Calling  │
                      │   Response time: &amp;lt; 200ms         │
                      └──────────────┬──────────────────┘
                                     │
         ┌────────────────────────────┼────────────────────────────┐
         │ HTTP/REST                  │ HTTP/REST                  │ HTTP/REST
         v                            v                            v
┌─────────────────────┐  ┌─────────────────────┐  ┌─────────────────────────┐
│   System Agent      │  │   Vision Agent     │  │   Home Assistant Agent  │
│  (Linux / DBus)    │  │  (Screen capture)  │  │  (Smart home control)   │
│ · Volume / Brightness│  │ · maim / grim      │  │ · Lights, sensors       │
│ · Window management │  │ · Moondream2 VLM   │  │ · WebSocket API         │
│ · Process control   │  │ · Screen analysis  │  │ · Automations           │
└─────────────────────┘  └─────────────────────┘  └─────────────────────────┘
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;Each agent is a &lt;strong&gt;FastAPI service&lt;/strong&gt; inside a Docker container. They don't know about each other. They just expose a clean REST interface and wait for instructions from Ollama.&lt;/p&gt;

&lt;h2&gt;
  
  
  Why Ollama as the Brain?
&lt;/h2&gt;

&lt;p&gt;Ollama runs small, capable LLMs locally. The key feature here is &lt;strong&gt;function calling&lt;/strong&gt; (tool calling) — Ollama doesn't execute commands directly. It returns a structured JSON decision: &lt;em&gt;"call the &lt;code&gt;set_volume&lt;/code&gt; function with parameter &lt;code&gt;level: 75&lt;/code&gt;"&lt;/em&gt;.&lt;/p&gt;

&lt;p&gt;This is cleaner than giving the LLM raw bash access. The function list is a strict allowlist — defined in JSON Schema. The model can only call what's explicitly permitted. No &lt;code&gt;rm -rf /&lt;/code&gt;, no matter how creatively phrased.&lt;/p&gt;

&lt;p&gt;The models I'm using are intentionally small:&lt;/p&gt;

&lt;div class="table-wrapper-paragraph"&gt;&lt;table&gt;
&lt;thead&gt;
&lt;tr&gt;
&lt;th&gt;Model&lt;/th&gt;
&lt;th&gt;Size&lt;/th&gt;
&lt;th&gt;Why&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td&gt;&lt;code&gt;qwen2.5:3b&lt;/code&gt;&lt;/td&gt;
&lt;td&gt;~2GB&lt;/td&gt;
&lt;td&gt;Main orchestrator. Fast, good Finnish support&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;&lt;code&gt;llama3.2:3b&lt;/code&gt;&lt;/td&gt;
&lt;td&gt;~2GB&lt;/td&gt;
&lt;td&gt;Fallback / general reasoning&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;&lt;code&gt;phi3:mini&lt;/code&gt;&lt;/td&gt;
&lt;td&gt;~2GB&lt;/td&gt;
&lt;td&gt;Lightweight tasks&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;&lt;em&gt;Moondream2&lt;/em&gt;&lt;/td&gt;
&lt;td&gt;~1GB&lt;/td&gt;
&lt;td&gt;Vision agent — screen capture analysis&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;

&lt;p&gt;All fit comfortably on a mid-range GPU or run on CPU. My Intel i5 desktop handles it fine.&lt;/p&gt;

&lt;h2&gt;
  
  
  The Agents
&lt;/h2&gt;

&lt;h3&gt;
  
  
  System Agent — Linux Desktop Control
&lt;/h3&gt;

&lt;p&gt;Controls the running Linux session via &lt;code&gt;wmctrl&lt;/code&gt;, &lt;code&gt;xdotool&lt;/code&gt;, &lt;code&gt;playerctl&lt;/code&gt;, and DBus. Things it can do:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;Set volume (&lt;code&gt;pamixer&lt;/code&gt;)&lt;/li&gt;
&lt;li&gt;Adjust brightness (&lt;code&gt;xbacklight&lt;/code&gt; or &lt;code&gt;brightnessctl&lt;/code&gt;)&lt;/li&gt;
&lt;li&gt;Launch applications (&lt;code&gt;gtk-launch&lt;/code&gt; or &lt;code&gt;xdg-open&lt;/code&gt;)&lt;/li&gt;
&lt;li&gt;Get active window info&lt;/li&gt;
&lt;li&gt;Control music playback (&lt;code&gt;playerctl&lt;/code&gt;)&lt;/li&gt;
&lt;li&gt;Read system stats (CPU, memory, disk)&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;No raw bash. Every function is explicitly defined.&lt;/p&gt;

&lt;h3&gt;
  
  
  Vision Agent — "The Eyes"
&lt;/h3&gt;

&lt;p&gt;Takes a screenshot (&lt;code&gt;maim&lt;/code&gt; or &lt;code&gt;grim&lt;/code&gt;) or reads a camera stream, then runs the image through a lightweight VLM (Moondream2). Can answer: &lt;em&gt;"What window is in the foreground?"&lt;/em&gt;, &lt;em&gt;"Did that build succeed?"&lt;/em&gt;, &lt;em&gt;"What's on my second monitor right now?"&lt;/em&gt;&lt;/p&gt;

&lt;h3&gt;
  
  
  Home Assistant Agent
&lt;/h3&gt;

&lt;p&gt;Talks to my Home Assistant instance over its WebSocket API. Lights, switches, climate, sensors — all accessible. I can say: &lt;em&gt;"Turn off the office lights, but only if nobody's in there"&lt;/em&gt; and it checks the occupancy sensor before acting.&lt;/p&gt;

&lt;h3&gt;
  
  
  Media &amp;amp; Data Agent
&lt;/h3&gt;

&lt;p&gt;Pulls in external data when needed: weather forecasts, electricity prices (relevant in Finland), InfluxDB logs, or web searches via a headless browser.&lt;/p&gt;

&lt;h2&gt;
  
  
  The Safety Layer
&lt;/h2&gt;

&lt;p&gt;Giving an AI system control over your desktop is a significant trust boundary. I'm implementing &lt;strong&gt;human-in-the-loop&lt;/strong&gt; for sensitive operations:&lt;/p&gt;

&lt;p&gt;For destructive or significant actions, the system doesn't just execute. It:&lt;/p&gt;

&lt;ol&gt;
&lt;li&gt;Shows a desktop notification with the proposed action&lt;/li&gt;
&lt;li&gt;Waits for confirmation (click or hotkey)&lt;/li&gt;
&lt;li&gt;Executes only on explicit approval&lt;/li&gt;
&lt;/ol&gt;

&lt;p&gt;This isn't paranoid — it's sensible. The function allowlist prevents most problems at the model level. The confirmation prompt handles the remaining edge cases.&lt;/p&gt;

&lt;h2&gt;
  
  
  The Docker Compose
&lt;/h2&gt;

&lt;p&gt;This is where it lives:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight yaml"&gt;&lt;code&gt;&lt;span class="na"&gt;version&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="s1"&gt;'3.8'&lt;/span&gt;

&lt;span class="na"&gt;services&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;
  &lt;span class="c1"&gt;# The Brain
&lt;/span&gt;  &lt;span class="na"&gt;ollama&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;
    &lt;span class="na"&gt;image&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="s"&gt;ollama/ollama:latest&lt;/span&gt;
    &lt;span class="na"&gt;container_name&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="s"&gt;jarvis-ollama&lt;/span&gt;
    &lt;span class="na"&gt;restart&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="s"&gt;unless-stopped&lt;/span&gt;
    &lt;span class="na"&gt;ports&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;
      &lt;span class="p"&gt;-&lt;/span&gt; &lt;span class="s2"&gt;"11434:11434"&lt;/span&gt;
    &lt;span class="na"&gt;volumes&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;
      &lt;span class="p"&gt;-&lt;/span&gt; &lt;span class="s"&gt;ollama_data:/root/.ollama&lt;/span&gt;
    &lt;span class="c1"&gt;# GPU passthrough if you have one:
&lt;/span&gt;    &lt;span class="c1"&gt;# deploy:
&lt;/span&gt;    &lt;span class="c1"&gt;#   resources:
&lt;/span&gt;    &lt;span class="c1"&gt;#     reservations:
&lt;/span&gt;    &lt;span class="c1"&gt;#       devices:
&lt;/span&gt;    &lt;span class="c1"&gt;#         - driver: nvidia
&lt;/span&gt;    &lt;span class="c1"&gt;#           count: all
&lt;/span&gt;    &lt;span class="c1"&gt;#           capabilities: [gpu]
&lt;/span&gt;
  &lt;span class="c1"&gt;# The Orchestrator
&lt;/span&gt;  &lt;span class="na"&gt;jarvis-core&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;
    &lt;span class="na"&gt;build&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="s"&gt;./core&lt;/span&gt;
    &lt;span class="na"&gt;container_name&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="s"&gt;jarvis-core&lt;/span&gt;
    &lt;span class="na"&gt;restart&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="s"&gt;unless-stopped&lt;/span&gt;
    &lt;span class="na"&gt;environment&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;
      &lt;span class="p"&gt;-&lt;/span&gt; &lt;span class="s"&gt;OLLAMA_BASE_URL=http://ollama:11434&lt;/span&gt;
      &lt;span class="p"&gt;-&lt;/span&gt; &lt;span class="s"&gt;DEFAULT_MODEL=qwen2.5:3b&lt;/span&gt;
    &lt;span class="na"&gt;ports&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;
      &lt;span class="p"&gt;-&lt;/span&gt; &lt;span class="s2"&gt;"8000:8000"&lt;/span&gt;
    &lt;span class="na"&gt;depends_on&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;
      &lt;span class="p"&gt;-&lt;/span&gt; &lt;span class="s"&gt;ollama&lt;/span&gt;

  &lt;span class="c1"&gt;# Linux System Agent
&lt;/span&gt;  &lt;span class="na"&gt;agent-system&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;
    &lt;span class="na"&gt;build&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="s"&gt;./agents/system&lt;/span&gt;
    &lt;span class="na"&gt;container_name&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="s"&gt;jarvis-agent-system&lt;/span&gt;
    &lt;span class="na"&gt;restart&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="s"&gt;unless-stopped&lt;/span&gt;
    &lt;span class="na"&gt;network_mode&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="s"&gt;host&lt;/span&gt;   &lt;span class="c1"&gt;# Needs access to host D-Bus / display
&lt;/span&gt;
&lt;span class="na"&gt;volumes&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;
  &lt;span class="na"&gt;ollama_data&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;h2&gt;
  
  
  Voice I/O
&lt;/h2&gt;

&lt;p&gt;For voice input, I'm using &lt;strong&gt;Faster-Whisper&lt;/strong&gt; (or &lt;code&gt;whisper.cpp&lt;/code&gt; for CPU-only) — push-to-talk on a hotkey. For output, &lt;strong&gt;Piper TTS&lt;/strong&gt; — fast, local, and surprisingly natural Finnish voices available.&lt;/p&gt;

&lt;p&gt;The pipeline: hotkey → Whisper → Ollama → function call or text → Piper TTS → speaker. Total latency target: under 1 second end-to-end.&lt;/p&gt;

&lt;h2&gt;
  
  
  Why Not Just Use Home Assistant's Built-in Assist?
&lt;/h2&gt;

&lt;p&gt;Home Assistant has an Assist feature withwyoming-satellite and OpenAI-compatible endpoints. That's actually the longer-term plan — integrate as an Assist pipeline so it works with HA's voice push button.&lt;/p&gt;

&lt;p&gt;But building the agents standalone first means they work independently of HA, and can be composed in other ways later. Modular by design.&lt;/p&gt;

&lt;h2&gt;
  
  
  What's Next
&lt;/h2&gt;

&lt;ol&gt;
&lt;li&gt;
&lt;strong&gt;This week:&lt;/strong&gt; Get Ollama running with &lt;code&gt;qwen2.5:3b&lt;/code&gt;, test function calling with a Python script&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Next:&lt;/strong&gt; Build the first agent (System Agent) — volume and brightness control as a proof of concept&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;After that:&lt;/strong&gt; Voice pipeline (Whisper + Piper), then Home Assistant integration&lt;/li&gt;
&lt;/ol&gt;

&lt;p&gt;I'll post updates as the build progresses. The code will be on GitHub when it's worth sharing.&lt;/p&gt;




&lt;p&gt;&lt;em&gt;Building your own tools is the whole point of this site. If you're interested in self-hosted AI, the Ollama docs are a good starting point: &lt;a href="https://ollama.com" rel="noopener noreferrer"&gt;ollama.com&lt;/a&gt;. And if you want something more polished out of the box, Home Assistant's Assist feature is worth exploring.&lt;/em&gt;&lt;/p&gt;

</description>
      <category>ai</category>
      <category>selfhosted</category>
      <category>linux</category>
      <category>homelab</category>
    </item>
    <item>
      <title>How to Actually Reduce Your Cloud Spend Before Year-End 2025</title>
      <dc:creator>Jonne Riepponen</dc:creator>
      <pubDate>Sat, 14 Nov 2026 15:14:01 +0000</pubDate>
      <link>https://dev.to/jronegit/how-to-actually-reduce-your-cloud-spend-before-year-end-2025-5cag</link>
      <guid>https://dev.to/jronegit/how-to-actually-reduce-your-cloud-spend-before-year-end-2025-5cag</guid>
      <description>&lt;h1&gt;
  
  
  How to Actually Reduce Your Cloud Spend Before Year-End 2025
&lt;/h1&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fmi76sgwkr6gzh3ws79u7.png" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fmi76sgwkr6gzh3ws79u7.png" alt="How to Actually Reduce Your Cloud Spend Before Year-End 2025" width="800" height="457"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;{{&amp;lt; aff-disclosure-inline &amp;gt;}}&lt;/p&gt;

&lt;p&gt;I only recommend products and services I genuinely use and believe will help you reduce cloud costs.&lt;/p&gt;

&lt;p&gt;As we approach year-end, many organizations are scrambling to optimize their cloud spend before budget renewals. If you're a sysadmin or DevOps engineer looking to make a real impact, this guide will help you identify and eliminate cloud waste while improving performance.&lt;/p&gt;

&lt;h2&gt;
  
  
  Why Focus on Cloud Costs Now?
&lt;/h2&gt;

&lt;p&gt;Q4 is the perfect time for cloud optimization because:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Budget cycles&lt;/strong&gt;: Finance teams are reviewing annual spending&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Idle resources&lt;/strong&gt;: Holiday traffic patterns expose unused resources&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Renewal season&lt;/strong&gt;: Many cloud contracts come up for negotiation&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Performance pressure&lt;/strong&gt;: Year-end deadlines force prioritization&lt;/li&gt;
&lt;/ul&gt;

&lt;h2&gt;
  
  
  5 Immediate Actions to Reduce Cloud Costs
&lt;/h2&gt;

&lt;h3&gt;
  
  
  1. &lt;strong&gt;Stop the Bleeding: Identify Zombie Resources&lt;/strong&gt;
&lt;/h3&gt;

&lt;p&gt;Start with a comprehensive audit using cloud-native tools:&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;AWS Cost Explorer + Cur&lt;/strong&gt;&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight shell"&gt;&lt;code&gt;&lt;span class="c"&gt;# Install AWS CLI cost management tools&lt;/span&gt;
pip install awscli cost-explorer-cli

&lt;span class="c"&gt;# Run cost analysis&lt;/span&gt;
aws ce get-cost-and-usage \
  --time-period &lt;span class="nv"&gt;Start&lt;/span&gt;&lt;span class="o"&gt;=&lt;/span&gt;2025-01-01,End=2025-12-31 \
  --granularity MONTHLY \
  --metrics BlendedCost \
  --group-by &lt;span class="nv"&gt;Type&lt;/span&gt;&lt;span class="o"&gt;=&lt;/span&gt;DIMENSION,Key=SERVICE \
  --filter file://zombie-resources.json
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;&lt;strong&gt;Quick Wins Checklist:&lt;/strong&gt;&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;✅ Delete unattached volumes (saves $10-50/month per volume)&lt;/li&gt;
&lt;li&gt;✅ Stop unused EC2 instances (save $20-200/month per instance)&lt;/li&gt;
&lt;li&gt;✅ Remove unused elastic IPs (save $4.50/month per IP)&lt;/li&gt;
&lt;li&gt;✅ Delete unused load balancers (save $20-25/month)&lt;/li&gt;
&lt;li&gt;✅ Stop unused RDS instances (save $100-500/month per instance)&lt;/li&gt;
&lt;/ul&gt;

&lt;h3&gt;
  
  
  2. &lt;strong&gt;Right-Size Your Infrastructure&lt;/strong&gt;
&lt;/h3&gt;

&lt;p&gt;Most organizations run instances 2-3x larger than needed.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Rightsizing Analysis:&lt;/strong&gt;&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight shell"&gt;&lt;code&gt;&lt;span class="c"&gt;# CloudWatch CPU analysis&lt;/span&gt;
aws cloudwatch get-metric-statistics \
  --namespace AWS/EC2 \
  --metric-name CPUUtilization \
  --dimensions &lt;span class="nv"&gt;Name&lt;/span&gt;&lt;span class="o"&gt;=&lt;/span&gt;InstanceId,Value=i-1234567890abcdef0 \
  --statistics Maximum \
  --period 86400 \
  --start-time 2025-10-01T00:00:00Z \
  --end-time 2025-11-01T00:00:00Z
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;&lt;strong&gt;Rule of thumb&lt;/strong&gt;: If CPU usage consistently below 30%, downsize. If consistently above 80%, you need vertical scaling.&lt;/p&gt;

&lt;h3&gt;
  
  
  3. &lt;strong&gt;Implement Auto-Scaling (Properly)&lt;/strong&gt;
&lt;/h3&gt;

&lt;p&gt;Many teams set up auto-scaling but configure it wrong.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Good Auto-Scaling Configuration:&lt;/strong&gt;&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight yaml"&gt;&lt;code&gt;&lt;span class="c1"&gt;# Kubernetes HPA example
&lt;/span&gt;&lt;span class="na"&gt;apiVersion&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="s"&gt;autoscaling/v2&lt;/span&gt;
&lt;span class="na"&gt;kind&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="s"&gt;HorizontalPodAutoscaler&lt;/span&gt;
&lt;span class="na"&gt;metadata&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;
  &lt;span class="na"&gt;name&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="s"&gt;app-hpa&lt;/span&gt;
&lt;span class="na"&gt;spec&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;
  &lt;span class="na"&gt;scaleTargetRef&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;
    &lt;span class="na"&gt;apiVersion&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="s"&gt;apps/v1&lt;/span&gt;
    &lt;span class="na"&gt;kind&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="s"&gt;Deployment&lt;/span&gt;
    &lt;span class="na"&gt;name&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="s"&gt;web-app&lt;/span&gt;
  &lt;span class="na"&gt;minReplicas&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="mi"&gt;2&lt;/span&gt;
  &lt;span class="na"&gt;maxReplicas&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="mi"&gt;10&lt;/span&gt;
  &lt;span class="na"&gt;metrics&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;
  &lt;span class="p"&gt;-&lt;/span&gt; &lt;span class="na"&gt;type&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="s"&gt;Resource&lt;/span&gt;
    &lt;span class="na"&gt;resource&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;
      &lt;span class="na"&gt;name&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="s"&gt;cpu&lt;/span&gt;
      &lt;span class="na"&gt;target&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;
        &lt;span class="na"&gt;type&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="s"&gt;Utilization&lt;/span&gt;
        &lt;span class="na"&gt;averageUtilization&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="mi"&gt;70&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;&lt;strong&gt;Common Auto-Scaling Mistakes:&lt;/strong&gt;&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;❌ Setting min replicas too high&lt;/li&gt;
&lt;li&gt;❌ Using inappropriate metrics (memory instead of CPU)&lt;/li&gt;
&lt;li&gt;❌ Not warming up instances (causes slow scaling)&lt;/li&gt;
&lt;/ul&gt;

&lt;h3&gt;
  
  
  4. &lt;strong&gt;Optimize Storage Costs&lt;/strong&gt;
&lt;/h3&gt;

&lt;p&gt;Storage is where most cloud costs hide.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Storage Optimization Strategies:&lt;/strong&gt;&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;S3 lifecycle policies&lt;/strong&gt;: Move old data to cheaper storage classes&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Compression&lt;/strong&gt;: Use tools like {{&amp;lt; aff merchant="newegg" text="GNU Parallel" url="&lt;a href="https://www.newegg.com/p/pl?d=gnu+parallel+book" rel="noopener noreferrer"&gt;https://www.newegg.com/p/pl?d=gnu+parallel+book&lt;/a&gt;" &amp;gt;}} for batch compression&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Deduplication&lt;/strong&gt;: Remove duplicate files before uploading&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Tiering&lt;/strong&gt;: Move infrequently accessed data to Glacier&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;&lt;strong&gt;S3 Cost Optimization Script:&lt;/strong&gt;&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight shell"&gt;&lt;code&gt;&lt;span class="c"&gt;#!/bin/bash&lt;/span&gt;
&lt;span class="c"&gt;# S3 storage optimization&lt;/span&gt;
aws s3api list-objects-v2 \
  --bucket my-important-bucket \
  --query &lt;span class="s1"&gt;'Contents[?LastModified&amp;lt;=`2025-08-01`].Key'&lt;/span&gt; \
  --output text &lt;span class="o"&gt;|&lt;/span&gt; &lt;span class="k"&gt;while&lt;/span&gt; &lt;span class="nb"&gt;read&lt;/span&gt; key&lt;span class="o"&gt;;&lt;/span&gt; &lt;span class="k"&gt;do&lt;/span&gt;
  aws s3api put-object-lifecycle-configuration \
    --bucket my-important-bucket \
    --lifecycle-configuration file://lifecycle.json
&lt;span class="k"&gt;done&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;h3&gt;
  
  
  5. &lt;strong&gt;Negotiate Better Rates&lt;/strong&gt;
&lt;/h3&gt;

&lt;p&gt;End of year is prime time for contract negotiations.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Negotiation Leverage Points:&lt;/strong&gt;&lt;/p&gt;

&lt;ol&gt;
&lt;li&gt;
&lt;strong&gt;Multi-year commitments&lt;/strong&gt;: 20-30% discounts&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Reserved instances&lt;/strong&gt;: 40-75% savings&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Volume discounts&lt;/strong&gt;: Tier pricing at thresholds&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Private offers&lt;/strong&gt;: Custom pricing for high-volume users&lt;/li&gt;
&lt;/ol&gt;

&lt;h2&gt;
  
  
  Tools I Use for Cloud Cost Analysis
&lt;/h2&gt;

&lt;p&gt;&lt;strong&gt;Free Tools:&lt;/strong&gt;&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;{{&amp;lt; aff merchant="amazon" text="AWS Cost Explorer" url="&lt;a href="https://console.aws.amazon.com/cost-reports/" rel="noopener noreferrer"&gt;https://console.aws.amazon.com/cost-reports/&lt;/a&gt;" &amp;gt;}} - Built-in cost analysis&lt;/li&gt;
&lt;li&gt;
&lt;a href="https://cloud.google.com/billing/docs/how-to/export-data" rel="noopener noreferrer"&gt;Google Cloud Billing Export&lt;/a&gt; - BigQuery integration&lt;/li&gt;
&lt;li&gt;
&lt;a href="https://portal.azure.com/#blade/Microsoft_Azure_Billing/CostManagementBVMVBlade/overview" rel="noopener noreferrer"&gt;Azure Cost Management&lt;/a&gt; - Automated alerts&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;&lt;strong&gt;Paid Tools Worth the Investment:&lt;/strong&gt;&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;a href="https://kubecost.com/" rel="noopener noreferrer"&gt;Kubecost&lt;/a&gt; - Kubernetes cost visibility ($199/month)&lt;/li&gt;
&lt;li&gt;
&lt;a href="https://www.vmware.com/products/cloudhealth.html" rel="noopener noreferrer"&gt;CloudHealth&lt;/a&gt; - Multi-cloud optimization&lt;/li&gt;
&lt;li&gt;
&lt;a href="https://www.cloudcheckr.com/" rel="noopener noreferrer"&gt;CloudCheckr&lt;/a&gt; - AI-powered cost optimization&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;&lt;em&gt;[Prices may be higher for you on some links, but I only recommend tools that save more than they cost.]&lt;/em&gt;&lt;/p&gt;

&lt;h2&gt;
  
  
  Case Study: How I Saved $2,400/month in One Week
&lt;/h2&gt;

&lt;p&gt;&lt;strong&gt;Problem&lt;/strong&gt;: Client's AWS bill jumped from $8,000 to $15,000 in 3 months.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Actions Taken:&lt;/strong&gt;&lt;/p&gt;

&lt;ol&gt;
&lt;li&gt;Identified 23 unused EC2 instances running for 90+ days&lt;/li&gt;
&lt;li&gt;Right-sized 15 instances (saved 40% per instance)&lt;/li&gt;
&lt;li&gt;Implemented S3 lifecycle policies (saved 60% on storage)&lt;/li&gt;
&lt;li&gt;Set up proper auto-scaling (reduced peak instances by 30%)&lt;/li&gt;
&lt;/ol&gt;

&lt;p&gt;&lt;strong&gt;Results&lt;/strong&gt;:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Week 1&lt;/strong&gt;: Saved $1,200/month (immediate wins)&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Month 1&lt;/strong&gt;: Saved $2,400/month (after optimizations)&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Year 1&lt;/strong&gt;: Projected savings of $28,800&lt;/li&gt;
&lt;/ul&gt;

&lt;h2&gt;
  
  
  Essential Cloud Cost Monitoring Setup
&lt;/h2&gt;

&lt;p&gt;&lt;strong&gt;Dashboard to Set Up This Week:&lt;/strong&gt;&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight shell"&gt;&lt;code&gt;&lt;span class="c"&gt;# CloudWatch custom metrics for cost tracking&lt;/span&gt;
aws cloudwatch put-metric-data \
  --metric-name DailyCost \
  --namespace CloudCosts \
  --value 123.45 \
  --timestamp 2025-11-01T00:00:00Z

&lt;span class="c"&gt;# Alert when daily costs exceed thresholds&lt;/span&gt;
aws cloudwatch put-metric-alarm \
  --alarm-name HighDailyCosts \
  --metric-name DailyCost \
  --namespace CloudCosts \
  --statistic Average \
  --period 86400 \
  --evaluation-periods 1 \
  --threshold 200.00 \
  --comparison-operator GreaterThanThreshold
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;h2&gt;
  
  
  Year-End Optimization Checklist
&lt;/h2&gt;

&lt;p&gt;&lt;strong&gt;Before December 1st:&lt;/strong&gt;&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;[ ] Complete infrastructure audit&lt;/li&gt;
&lt;li&gt;[ ] Set up cost monitoring alerts&lt;/li&gt;
&lt;li&gt;[ ] Implement right-sizing recommendations&lt;/li&gt;
&lt;li&gt;[ ] Start vendor negotiations&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;&lt;strong&gt;Before January 1st:&lt;/strong&gt;&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;[ ] Execute optimization changes&lt;/li&gt;
&lt;li&gt;[ ] Document savings for next year's budget&lt;/li&gt;
&lt;li&gt;[ ] Set up automated scaling policies&lt;/li&gt;
&lt;li&gt;[ ] Negotiate multi-year contracts&lt;/li&gt;
&lt;/ul&gt;

&lt;h2&gt;
  
  
  Getting Buy-In from Management
&lt;/h2&gt;

&lt;p&gt;&lt;strong&gt;Frame cost optimization in business terms:&lt;/strong&gt;&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;"We're reducing technical debt"&lt;/strong&gt; - Better infrastructure = less downtime&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;"Improving performance"&lt;/strong&gt; - Right-sized resources = better user experience
&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;"Scaling efficiently"&lt;/strong&gt; - Auto-scaling = handle traffic spikes without waste&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;"Budget predictability"&lt;/strong&gt; - Better monitoring = fewer surprises&lt;/li&gt;
&lt;/ul&gt;

&lt;h2&gt;
  
  
  Next Steps for This Month
&lt;/h2&gt;

&lt;ol&gt;
&lt;li&gt;
&lt;strong&gt;Week 1&lt;/strong&gt;: Audit current spending using cloud provider tools&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Week 2&lt;/strong&gt;: Implement quick wins (delete zombie resources)&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Week 3&lt;/strong&gt;: Set up monitoring and alerts&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Week 4&lt;/strong&gt;: Plan and negotiate vendor contracts for next year&lt;/li&gt;
&lt;/ol&gt;

&lt;p&gt;&lt;strong&gt;Need help with your cloud optimization?&lt;/strong&gt; Start with the free tools and focus on the quick wins first. The goal isn't perfection—it's steady improvement that compounds over time.&lt;/p&gt;




&lt;p&gt;&lt;strong&gt;Ready to dive deeper?&lt;/strong&gt; Check out my guide on &lt;a href="https://pragmaticsysadmin.help/sysadmin/kubernetes-without-jargon-pods-processes-services/" rel="noopener noreferrer"&gt;Kubernetes Without Jargon&lt;/a&gt; for container optimization strategies that also reduce costs.&lt;/p&gt;

&lt;p&gt;&lt;em&gt;What cloud cost challenges are you facing? Share your experiences in the comments below.&lt;/em&gt;&lt;/p&gt;

</description>
      <category>cloudoptimization</category>
      <category>finops</category>
      <category>costreduction</category>
      <category>devops</category>
    </item>
    <item>
      <title>How I Built This Blog: Hugo + Firebase + Buttondown in 30 Minutes</title>
      <dc:creator>Jonne Riepponen</dc:creator>
      <pubDate>Sat, 14 Nov 2026 15:13:26 +0000</pubDate>
      <link>https://dev.to/jronegit/how-i-built-this-blog-hugo-firebase-buttondown-in-30-minutes-1flc</link>
      <guid>https://dev.to/jronegit/how-i-built-this-blog-hugo-firebase-buttondown-in-30-minutes-1flc</guid>
      <description>&lt;h2&gt;
  
  
  Why This Stack?
&lt;/h2&gt;

&lt;p&gt;As a sysadmin, I wanted a blog that's:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Fast&lt;/strong&gt;: Static sites load instantly&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Cheap&lt;/strong&gt;: Firebase free tier handles tons of traffic
&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Simple&lt;/strong&gt;: No databases, no server maintenance&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;SEO-ready&lt;/strong&gt;: Hugo generates clean, optimized HTML&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;Here's exactly how I built this site in under 30 minutes.&lt;/p&gt;

&lt;h2&gt;
  
  
  The Stack
&lt;/h2&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Hugo&lt;/strong&gt;: Static site generator (like Jekyll, but faster)&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;PaperMod&lt;/strong&gt;: Clean, minimal Hugo theme&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Firebase Hosting&lt;/strong&gt;: Google's CDN with free SSL&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Buttondown&lt;/strong&gt;: Simple, elegant newsletter platform&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Google Analytics&lt;/strong&gt;: Track what people read&lt;/li&gt;
&lt;/ul&gt;

&lt;h2&gt;
  
  
  Step-by-Step Build Process
&lt;/h2&gt;

&lt;h3&gt;
  
  
  1. Install the Tools
&lt;/h3&gt;

&lt;p&gt;On Windows with Chocolatey:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight shell"&gt;&lt;code&gt;choco install hugo-extended
npm install -g firebase-tools
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;h3&gt;
  
  
  2. Create the Site
&lt;/h3&gt;



&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight shell"&gt;&lt;code&gt;hugo new site pragmatic-sysadmin
&lt;span class="nb"&gt;cd&lt;/span&gt; pragmatic-sysadmin
git init
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;h3&gt;
  
  
  3. Add PaperMod Theme
&lt;/h3&gt;



&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight shell"&gt;&lt;code&gt;git submodule add https://github.com/adityatelange/hugo-PaperMod themes/PaperMod
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;h3&gt;
  
  
  4. Configure Hugo
&lt;/h3&gt;

&lt;p&gt;Updated &lt;code&gt;hugo.toml&lt;/code&gt; with SEO basics:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight toml"&gt;&lt;code&gt;&lt;span class="na"&gt;baseURL&lt;/span&gt; &lt;span class="o"&gt;=&lt;/span&gt; &lt;span class="s2"&gt;"https://pragmaticsysadmin.help/"&lt;/span&gt;
&lt;span class="na"&gt;title&lt;/span&gt; &lt;span class="o"&gt;=&lt;/span&gt; &lt;span class="s2"&gt;"Pragmatic Sysadmin"&lt;/span&gt;
&lt;span class="na"&gt;theme&lt;/span&gt; &lt;span class="o"&gt;=&lt;/span&gt; &lt;span class="s2"&gt;"PaperMod"&lt;/span&gt;
&lt;span class="na"&gt;enableRobotsTXT&lt;/span&gt; &lt;span class="o"&gt;=&lt;/span&gt; &lt;span class="kc"&gt;true&lt;/span&gt;

&lt;span class="k"&gt;[pagination]&lt;/span&gt;
  &lt;span class="na"&gt;pagerSize&lt;/span&gt; &lt;span class="o"&gt;=&lt;/span&gt; &lt;span class="mi"&gt;10&lt;/span&gt;

&lt;span class="k"&gt;[markup]&lt;/span&gt;
  &lt;span class="k"&gt;[markup.goldmark]&lt;/span&gt;
    &lt;span class="k"&gt;[markup.goldmark.renderer]&lt;/span&gt;
      &lt;span class="na"&gt;unsafe&lt;/span&gt; &lt;span class="o"&gt;=&lt;/span&gt; &lt;span class="kc"&gt;true&lt;/span&gt;  &lt;span class="c1"&gt;# Allows HTML in markdown&lt;/span&gt;

&lt;span class="k"&gt;[params]&lt;/span&gt;
  &lt;span class="na"&gt;description&lt;/span&gt; &lt;span class="o"&gt;=&lt;/span&gt; &lt;span class="s2"&gt;"Funny, practical home networking and de-mystified cloud."&lt;/span&gt;
  &lt;span class="na"&gt;defaultTheme&lt;/span&gt; &lt;span class="o"&gt;=&lt;/span&gt; &lt;span class="s2"&gt;"auto"&lt;/span&gt;
  &lt;span class="na"&gt;ShowReadingTime&lt;/span&gt; &lt;span class="o"&gt;=&lt;/span&gt; &lt;span class="kc"&gt;true&lt;/span&gt;
  &lt;span class="na"&gt;ShowPostNavLinks&lt;/span&gt; &lt;span class="o"&gt;=&lt;/span&gt; &lt;span class="kc"&gt;true&lt;/span&gt;
  &lt;span class="na"&gt;ShowShareButtons&lt;/span&gt; &lt;span class="o"&gt;=&lt;/span&gt; &lt;span class="kc"&gt;true&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;h3&gt;
  
  
  5. Add Google Analytics
&lt;/h3&gt;

&lt;p&gt;Created &lt;code&gt;layouts/partials/head-end.html&lt;/code&gt; with GA4 tracking:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight html"&gt;&lt;code&gt;&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;h3&gt;
  
  
  6. Deploy to Firebase
&lt;/h3&gt;



&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight shell"&gt;&lt;code&gt;hugo --minify          &lt;span class="c"&gt;# Build static files&lt;/span&gt;
firebase login         &lt;span class="c"&gt;# Authenticate&lt;/span&gt;
firebase init hosting  &lt;span class="c"&gt;# Configure hosting&lt;/span&gt;
firebase deploy        &lt;span class="c"&gt;# Go live!&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;Firebase setup answers:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;Public directory: &lt;code&gt;public&lt;/code&gt;
&lt;/li&gt;
&lt;li&gt;Single-page app: &lt;code&gt;No&lt;/code&gt;
&lt;/li&gt;
&lt;li&gt;GitHub deploys: &lt;code&gt;No&lt;/code&gt;
&lt;/li&gt;
&lt;/ul&gt;

&lt;h3&gt;
  
  
  7. Add Buttondown Newsletter
&lt;/h3&gt;

&lt;p&gt;Created newsletter page with embedded form:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight markdown"&gt;&lt;code&gt;&lt;span class="gu"&gt;## Join the newsletter
&lt;/span&gt;Short, practical tips. No spam, just sysadmin sanity.

&amp;lt;form action="https://buttondown.com/api/emails/subscribe-jonne" method="post"&amp;gt;
  &amp;lt;input type="email" name="email" placeholder="you@example.com" required&amp;gt;
  &amp;lt;button type="submit"&amp;gt;Subscribe&amp;lt;/button&amp;gt;
&amp;lt;/form&amp;gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;h3&gt;
  
  
  8. Connect Custom Domain
&lt;/h3&gt;

&lt;p&gt;In Firebase Console → Hosting → Add custom domain, then updated DNS:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;A record: &lt;code&gt;pragmaticsysadmin.help&lt;/code&gt; → &lt;code&gt;199.36.158.100&lt;/code&gt;
&lt;/li&gt;
&lt;li&gt;TXT record: &lt;code&gt;pragmaticsysadmin.help&lt;/code&gt; → &lt;code&gt;hosting-site=johnnysblogman&lt;/code&gt;
&lt;/li&gt;
&lt;/ul&gt;

&lt;h2&gt;
  
  
  The Results
&lt;/h2&gt;

&lt;p&gt;&lt;strong&gt;Live site&lt;/strong&gt;: &lt;a href="https://pragmaticsysadmin.help" rel="noopener noreferrer"&gt;https://pragmaticsysadmin.help&lt;/a&gt; (once DNS propagates)&lt;br&gt;&lt;br&gt;
&lt;strong&gt;Backup URL&lt;/strong&gt;: &lt;a href="https://johnnysblogman.web.app" rel="noopener noreferrer"&gt;https://johnnysblogman.web.app&lt;/a&gt;&lt;br&gt;&lt;br&gt;
&lt;strong&gt;Cost&lt;/strong&gt;: $0/month for reasonable traffic&lt;br&gt;&lt;br&gt;
&lt;strong&gt;Build time&lt;/strong&gt;: ~2 minutes&lt;br&gt;&lt;br&gt;
&lt;strong&gt;Deploy time&lt;/strong&gt;: ~30 seconds&lt;/p&gt;
&lt;h2&gt;
  
  
  Quick Commands
&lt;/h2&gt;


&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight shell"&gt;&lt;code&gt;&lt;span class="c"&gt;# Local development&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;


&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fq4q1mtar07y18bvg4xz7.png" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fq4q1mtar07y18bvg4xz7.png" alt="Local development" width="800" height="457"&gt;&lt;/a&gt;&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight shell"&gt;&lt;code&gt;hugo server -D

&lt;span class="c"&gt;# Build and deploy&lt;/span&gt;
hugo --minify
firebase deploy
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;h2&gt;
  
  
  Why This Works
&lt;/h2&gt;

&lt;p&gt;This stack scales from zero to thousands of visitors without touching server configs. Hugo pre-builds everything, Firebase's CDN serves it globally, and analytics + newsletter capture runs automatically.&lt;/p&gt;

&lt;p&gt;Perfect for technical blogs, documentation sites, or any content that doesn't need real-time data.&lt;/p&gt;

&lt;p&gt;Next up: I'll be writing about secure home networking, because apparently everyone's router is still using admin/admin. 🤦‍♂️&lt;/p&gt;

&lt;p&gt;&lt;em&gt;Related reads:&lt;/em&gt;&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;&lt;em&gt;&lt;a href="https://pragmaticsysadmin.help/senior-tech/ultimate-secure-fast-home-network-2025/" rel="noopener noreferrer"&gt;The Ultimate Guide to a Secure &amp;amp; Fast Home Network (2025)&lt;/a&gt;&lt;/em&gt;&lt;/li&gt;
&lt;li&gt;&lt;em&gt;&lt;a href="https://pragmaticsysadmin.help/sysadmin/2025-10-29-setting-up-a-home-lab-a-beginner-s-guide/" rel="noopener noreferrer"&gt;Setting Up a Home Lab: A Beginner's Guide&lt;/a&gt;&lt;/em&gt;&lt;/li&gt;
&lt;/ul&gt;

</description>
      <category>hugo</category>
      <category>firebase</category>
      <category>blogging</category>
      <category>staticsites</category>
    </item>
    <item>
      <title>I Built Homelab Architect — Plan Your Stack Before You Build It</title>
      <dc:creator>Jonne Riepponen</dc:creator>
      <pubDate>Sat, 14 Nov 2026 15:12:51 +0000</pubDate>
      <link>https://dev.to/jronegit/i-built-homelab-architect-plan-your-stack-before-you-build-it-ofi</link>
      <guid>https://dev.to/jronegit/i-built-homelab-architect-plan-your-stack-before-you-build-it-ofi</guid>
      <description>&lt;p&gt;Most homelab beginners make the same mistake: they grab a Docker Compose template, paste it in, and then discover three months later that their "simple setup" is a sprawling mess of undocumented dependencies, a database with no backup, and ports wide open to the internet.&lt;/p&gt;

&lt;p&gt;I've been there. I still end up there sometimes.&lt;/p&gt;

&lt;p&gt;So I built a tool to catch it before it happens.&lt;/p&gt;

&lt;h2&gt;
  
  
  What Homelab Architect Does
&lt;/h2&gt;

&lt;p&gt;&lt;a href="https://pragmaticsysadmin.help/tools/homelab-architect.html" rel="noopener noreferrer"&gt;Homelab Architect&lt;/a&gt; takes your docker-compose.yml — or lets you build your stack manually from a curated database of 45+ common homelab services — and generates:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Resource dashboard&lt;/strong&gt; — total CPU and RAM estimates across your entire stack&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Service cards&lt;/strong&gt; — each service with its resource footprint, exposed ports, and restart policy&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Security audit&lt;/strong&gt; — flags &lt;code&gt;:latest&lt;/code&gt; tags, exposed ports without firewall rules, missing restart policies&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Single points of failure&lt;/strong&gt; — detects services with no backup plan that others depend on&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Dependency map&lt;/strong&gt; — see what talks to what, so you know what to restart in what order&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Backup plan&lt;/strong&gt; — tells you exactly what to back up and how, based on your actual services&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Export&lt;/strong&gt; — save as JSON to share or reload later, or export a Mermaid diagram for documentation&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;It's essentially the "before you build" companion to &lt;a href="https://pragmaticsysadmin.help/tools/prism-engine.html" rel="noopener noreferrer"&gt;Prism Engine&lt;/a&gt;'s "after you've built" visualization.&lt;/p&gt;

&lt;h2&gt;
  
  
  How It Works
&lt;/h2&gt;

&lt;p&gt;&lt;strong&gt;Import tab:&lt;/strong&gt; Paste your docker-compose.yml. The parser extracts services, ports, volumes, &lt;code&gt;depends_on&lt;/code&gt; relationships, and restart policies. It maps image names to a database of 45+ known services (PostgreSQL, Nginx, Plex, Home Assistant, Pi-hole, and more) to fill in resource estimates and security flags automatically.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Builder tab:&lt;/strong&gt; Don't have a compose file yet? Browse services by category (Infrastructure, Media, Automation, Security, Network) or search by name. Add them one by one, specify dependencies, and build your plan from scratch.&lt;/p&gt;

&lt;h2&gt;
  
  
  What It Catches
&lt;/h2&gt;

&lt;p&gt;A few examples from my own docker-compose files:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;&lt;code&gt;nginx:latest&lt;/code&gt;&lt;/strong&gt; → flagged as a security issue. Pin to &lt;code&gt;:1.27-alpine&lt;/code&gt; or similar&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Database without &lt;code&gt;restart: unless-stopped&lt;/code&gt;&lt;/strong&gt; → marked as a single point of failure. A crash means manual intervention&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Service with no backup recommendation&lt;/strong&gt; → added to the backup plan with its data directory path&lt;/li&gt;
&lt;/ul&gt;

&lt;h2&gt;
  
  
  The Service Database
&lt;/h2&gt;

&lt;p&gt;The builder includes 45+ services with realistic resource estimates:&lt;/p&gt;

&lt;div class="table-wrapper-paragraph"&gt;&lt;table&gt;
&lt;thead&gt;
&lt;tr&gt;
&lt;th&gt;Category&lt;/th&gt;
&lt;th&gt;Services&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td&gt;&lt;strong&gt;Infra&lt;/strong&gt;&lt;/td&gt;
&lt;td&gt;Nginx, Traefik, Caddy, PostgreSQL, MySQL, MariaDB, Redis, Portainer, Prometheus, Grafana, InfluxDB, MinIO, Watchtower, Docker&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;&lt;strong&gt;Media&lt;/strong&gt;&lt;/td&gt;
&lt;td&gt;Plex, Jellyfin, Emby, Sonarr, Radarr, Prowlarr, SABnzbd, NZBGet, Bazarr, Calibre-Web, PhotoPrism, Immich&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;&lt;strong&gt;Automation&lt;/strong&gt;&lt;/td&gt;
&lt;td&gt;Home Assistant, Nextcloud, Gitea, Node-RED, Actual Budget, MQTT, Zigbee2MQTT&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;&lt;strong&gt;Security&lt;/strong&gt;&lt;/td&gt;
&lt;td&gt;WireGuard, Vaultwarden, Bitwarden, Authelia, Authentik, Duplicati, Vault, Frigate, Mailcow&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;&lt;strong&gt;Network&lt;/strong&gt;&lt;/td&gt;
&lt;td&gt;Pi-hole, AdGuard Home, UniFi Controller, Uptime Kuma, Statping&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;

&lt;p&gt;Each entry has CPU/RAM/disk estimates, default ports, known security flags, and backup recommendations.&lt;/p&gt;

&lt;h2&gt;
  
  
  No Account, No Upload
&lt;/h2&gt;

&lt;p&gt;Everything runs in your browser. Your docker-compose.yml is never sent anywhere — the YAML parsing, security analysis, and rendering all happen client-side. Works offline after first load.&lt;/p&gt;

&lt;p&gt;Try it: &lt;a href="https://pragmaticsysadmin.help/tools/homelab-architect.html" rel="noopener noreferrer"&gt;Homelab Architect&lt;/a&gt;&lt;/p&gt;

</description>
      <category>homelab</category>
      <category>docker</category>
      <category>selfhosted</category>
      <category>tools</category>
    </item>
    <item>
      <title>Probabilistic Code Needs Human Ground Truth</title>
      <dc:creator>Jonne Riepponen</dc:creator>
      <pubDate>Thu, 12 Nov 2026 16:57:10 +0000</pubDate>
      <link>https://dev.to/jronegit/probabilistic-code-needs-human-ground-truth-1ima</link>
      <guid>https://dev.to/jronegit/probabilistic-code-needs-human-ground-truth-1ima</guid>
      <description>&lt;p&gt;Last month I watched a support team demo their new AI deflection wall with pride. A customer asked where their invoice was, and the bot answered — confidently, politely, instantly — with a ticket number that didn't exist. Nobody caught it for three days. The bot wasn't lying. It was doing exactly what probabilistic models do: producing the most likely next tokens. The team had just mistaken fluency for truth, and their customers paid for the confusion.&lt;/p&gt;

&lt;p&gt;I've seen the same mistake in my own terminal. You'll pull a generated script that passes CI, then watch it break in production because the logic missed an edge case the model never saw. That gap between plausible output and reliable behavior is where the real work lives — and it's work only a human can do.&lt;/p&gt;

&lt;h2&gt;
  
  
  The deflection wall: fluency is not truth
&lt;/h2&gt;

&lt;p&gt;A customer service bot that answers instantly feels like progress. Your ticket queue shrinks, your response-time dashboard turns green, and management declares victory. But a language model doesn't check your CRM — it predicts what a helpful answer &lt;em&gt;sounds like&lt;/em&gt;. When it doesn't know the invoice status, it doesn't say "I don't know." It invents a ticket number with total confidence.&lt;/p&gt;

&lt;p&gt;That's not a bug you patch with a better prompt. It's the nature of the tool. A model has no notion of your system state, your security policy, or the particular quirks of your hardware. It might suggest a command that works in a container but assumes a different PATH on your bare-metal host. It'll introduce off-by-one loops, missing error handling, or wrong assumptions about return values — and every one of them will look perfectly reasonable until it runs.&lt;/p&gt;

&lt;h2&gt;
  
  
  Faked evidence is the same failure with higher stakes
&lt;/h2&gt;

&lt;p&gt;Now push that failure one step further. A generated photo, a cloned voice note, a "screenshot" of a conversation that never happened — probabilistic output presented as ground truth. I've watched people forward AI-generated images as proof of things that never occurred, not out of malice but because the output &lt;em&gt;looked&lt;/em&gt; authoritative.&lt;/p&gt;

&lt;p&gt;This is why I don't trust model output as evidence of anything. A model can't swear an affidavit. It can't be fired for getting it wrong. When faked evidence enters an incident report, a support thread, or a compliance audit, the only thing standing between you and a bad decision is a human who checks the primary source — the logs, the database, the actual system state.&lt;/p&gt;

&lt;h2&gt;
  
  
  What the model can't give you: accountability and empathy
&lt;/h2&gt;

&lt;p&gt;Here's the part the tooling discourse skips. Even if the model were always right, two things would still be missing.&lt;/p&gt;

&lt;p&gt;First, accountability. When a generated migration script drops a table at 3 AM, "the model suggested it" isn't an explanation your boss accepts. You ran it. You own it. The model can't take the pager, can't write the postmortem, can't look a customer in the eye. Responsibility doesn't transfer to a probability distribution — it stays with the human who pressed enter.&lt;/p&gt;

&lt;p&gt;Second, empathy. A support agent reading "we're looking into it" for the fifth time isn't facing a technical problem — they're facing a trust problem. A human can hear frustration, adjust tone, and say "I understand this is blocking your payroll run, here's exactly what happens next." A deflection wall says the words without the understanding, and customers can tell. Every time.&lt;/p&gt;

&lt;h2&gt;
  
  
  Where probabilistic output genuinely helps
&lt;/h2&gt;

&lt;p&gt;None of that means the tool is useless — it means the tool has a lane. For boilerplate, repetitive scaffolding, or translating a clear comment into a function signature, the model saves real typing. A well-prompted model produces a decent first draft of a cron wrapper, a systemd unit file, or a simple parsing loop. I've used &lt;a href="https://pragmaticsysadmin.help/sysadmin/ai-for-it-troubleshooting-2026/" rel="noopener noreferrer"&gt;AI for IT troubleshooting&lt;/a&gt; the same way: fast first pass, human verdict at the end.&lt;/p&gt;

&lt;p&gt;In those cases your effort shifts from writing to reviewing. That's still valuable work — arguably the more skilled half. Don't let anyone tell you reviewing is "just" checking.&lt;/p&gt;

&lt;h2&gt;
  
  
  Ground truth means running it on the real system
&lt;/h2&gt;

&lt;p&gt;Ground truth comes from observing the actual outcome on the target system, not from trusting a confidence score. You verify by running the code, checking logs, and confirming the intended state change really happened. That step can't be outsourced to the model because it requires touching the real environment, which the model never sees.&lt;/p&gt;

&lt;p&gt;My rule is simple: nothing generated runs on production without a staging run first. A &lt;a href="https://pragmaticsysadmin.help/sysadmin/2026-09-26-a-sandbox-for-the-linux-commands-youre-afraid-to-run/" rel="noopener noreferrer"&gt;sandbox where you can watch destructive commands fail safely&lt;/a&gt; builds the same reflex — you learn what "verify first" feels like before the stakes are real. And when output needs explaining rather than executing, I reach for &lt;a href="https://pragmaticsysadmin.help/meta/plain-english-explain-any-command-error-concept/" rel="noopener noreferrer"&gt;plain-English explanations of commands and errors&lt;/a&gt; instead of trusting the first plausible-sounding answer.&lt;/p&gt;

&lt;p&gt;Here's the smallest version of the workflow — a wrapper I actually use before letting generated scripts near anything I care about:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight shell"&gt;&lt;code&gt;&lt;span class="c"&gt;#!/bin/bash&lt;/span&gt;
&lt;span class="c"&gt;# verify-first.sh — run a generated script against staging, diff the&lt;/span&gt;
&lt;span class="c"&gt;# result against the known-good state, and refuse to proceed on mismatch.&lt;/span&gt;
&lt;span class="c"&gt;# Why a wrapper: it makes "did it change what I expected?" a habit&lt;/span&gt;
&lt;span class="c"&gt;# instead of a judgment call at 3 AM.&lt;/span&gt;
&lt;span class="nb"&gt;set&lt;/span&gt; -euo pipefail
&lt;span class="nv"&gt;STAGING_HOST&lt;/span&gt;&lt;span class="o"&gt;=&lt;/span&gt;&lt;span class="s2"&gt;"${1:?usage: verify-first.sh &amp;lt;staging-host&amp;gt; &amp;lt;script&amp;gt;}"&lt;/span&gt;
&lt;span class="nv"&gt;SCRIPT&lt;/span&gt;&lt;span class="o"&gt;=&lt;/span&gt;&lt;span class="s2"&gt;"${2:?usage: verify-first.sh &amp;lt;staging-host&amp;gt; &amp;lt;script&amp;gt;}"&lt;/span&gt;
&lt;span class="nv"&gt;BEFORE&lt;/span&gt;&lt;span class="o"&gt;=&lt;/span&gt;&lt;span class="nv"&gt;$(&lt;/span&gt;ssh &lt;span class="s2"&gt;"$STAGING_HOST"&lt;/span&gt; &lt;span class="s1"&gt;'systemctl list-units --state=failed --no-legend | wc -l'&lt;/span&gt;&lt;span class="o"&gt;)&lt;/span&gt;
scp &lt;span class="s2"&gt;"$SCRIPT"&lt;/span&gt; &lt;span class="s2"&gt;"$STAGING_HOST:/tmp/candidate.sh"&lt;/span&gt;
ssh &lt;span class="s2"&gt;"$STAGING_HOST"&lt;/span&gt; &lt;span class="s1"&gt;'bash -n /tmp/candidate.sh &amp;amp;&amp;amp; bash /tmp/candidate.sh --dry-run'&lt;/span&gt;
&lt;span class="nv"&gt;AFTER&lt;/span&gt;&lt;span class="o"&gt;=&lt;/span&gt;&lt;span class="nv"&gt;$(&lt;/span&gt;ssh &lt;span class="s2"&gt;"$STAGING_HOST"&lt;/span&gt; &lt;span class="s1"&gt;'systemctl list-units --state=failed --no-legend | wc -l'&lt;/span&gt;&lt;span class="o"&gt;)&lt;/span&gt;
&lt;span class="o"&gt;[&lt;/span&gt; &lt;span class="s2"&gt;"$BEFORE"&lt;/span&gt; &lt;span class="o"&gt;=&lt;/span&gt; &lt;span class="s2"&gt;"$AFTER"&lt;/span&gt; &lt;span class="o"&gt;]&lt;/span&gt; &lt;span class="o"&gt;||&lt;/span&gt; &lt;span class="o"&gt;{&lt;/span&gt; &lt;span class="nb"&gt;echo&lt;/span&gt; &lt;span class="s2"&gt;"❌ staging state changed: investigate before prod"&lt;/span&gt;&lt;span class="o"&gt;;&lt;/span&gt; &lt;span class="nb"&gt;exit&lt;/span&gt; 1&lt;span class="o"&gt;;&lt;/span&gt; &lt;span class="o"&gt;}&lt;/span&gt;
&lt;span class="nb"&gt;echo&lt;/span&gt; &lt;span class="s2"&gt;"✅ staging clean — review the dry-run output, then decide like a human"&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;h2&gt;
  
  
  A short checklist for reviewing generated code
&lt;/h2&gt;

&lt;ul&gt;
&lt;li&gt;Run it in staging that mirrors production — never trust the model's word for it&lt;/li&gt;
&lt;li&gt;Check error paths: missing files, bad input, non-zero exit codes&lt;/li&gt;
&lt;li&gt;Compare output against a known-good reference or a manual calculation&lt;/li&gt;
&lt;li&gt;Look for hard-coded values that should be parameters (paths, ports, credentials)&lt;/li&gt;
&lt;li&gt;Verify privileged operations sit inside the right sudo or capability boundaries&lt;/li&gt;
&lt;/ul&gt;

&lt;h2&gt;
  
  
  Keep the human in the loop
&lt;/h2&gt;

&lt;p&gt;The fix isn't less AI — it's AI in the right seat. A high-bandwidth copilot that drafts, suggests, and explains, with a human who verifies, decides, and answers for the outcome. The model gives you speed. You supply the truth.&lt;/p&gt;

&lt;p&gt;&lt;em&gt;Next time a generated answer looks perfect, ask yourself one question: what would prove it wrong? Then go check.&lt;/em&gt;&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;&lt;em&gt;&lt;a href="https://pragmaticsysadmin.help/sysadmin/ai-for-it-troubleshooting-2026/" rel="noopener noreferrer"&gt;AI for IT Troubleshooting: Real-World Use Cases&lt;/a&gt;&lt;/em&gt;&lt;/li&gt;
&lt;li&gt;&lt;em&gt;&lt;a href="https://pragmaticsysadmin.help/sysadmin/2026-09-26-a-sandbox-for-the-linux-commands-youre-afraid-to-run/" rel="noopener noreferrer"&gt;A Sandbox for the Linux Commands You're Afraid to Run&lt;/a&gt;&lt;/em&gt;&lt;/li&gt;
&lt;li&gt;&lt;em&gt;&lt;a href="https://pragmaticsysadmin.help/meta/plain-english-explain-any-command-error-concept/" rel="noopener noreferrer"&gt;I Built a Tool That Explains Any Command or Error in Plain Language&lt;/a&gt;&lt;/em&gt;&lt;/li&gt;
&lt;/ul&gt;

</description>
      <category>ai</category>
      <category>automation</category>
      <category>trust</category>
      <category>troubleshooting</category>
    </item>
    <item>
      <title>Replace Google Photos with Immich (The Honest 2026 Guide)</title>
      <dc:creator>Jonne Riepponen</dc:creator>
      <pubDate>Thu, 12 Nov 2026 16:56:35 +0000</pubDate>
      <link>https://dev.to/jronegit/replace-google-photos-with-immich-the-honest-2026-guide-4fo4</link>
      <guid>https://dev.to/jronegit/replace-google-photos-with-immich-the-honest-2026-guide-4fo4</guid>
      <description>&lt;p&gt;The moment I moved my parents off Google Photos, the free 15 GB of Google One became 200 GB of headroom, and I got back the photos Google had been helping itself to. That part was easy.&lt;/p&gt;

&lt;p&gt;The part nobody mentions is that Immich is a real server, not a web app, and it has requirements Google quietly pays for on your behalf. Six gigabytes of RAM, a disk you will back up separately, and a public endpoint on your home connection — which is where most of the sharp edges live.&lt;/p&gt;

&lt;p&gt;Here is what actually happened for me when I replaced Google Photos with Immich, checked against the official documentation on 10 November 2026. I've run it on a Pi 5 and on an N100 box, and both taught me something different.&lt;/p&gt;

&lt;h2&gt;
  
  
  What you actually get
&lt;/h2&gt;

&lt;p&gt;Immich is an open-source photo and video backup platform, AGPL-3.0, with apps for Android and iOS that upload new shots to your server automatically. On top of the backup it gives you the things people actually miss when they leave Google Photos:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;A timeline and albums&lt;/strong&gt; that mirror your phone's structure, with the option to exclude an album from backup.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Face recognition and natural-language search.&lt;/strong&gt; Type "beach 2024" or "grandchildren" and it finds them, because the machine learning runs on your hardware.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Original quality.&lt;/strong&gt; No compression, no re-encoding. EXIF and GPS stay intact.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Shared albums&lt;/strong&gt; and public links, so your family can drop photos in without touching your admin account.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Free Up Space&lt;/strong&gt; on the phone, which removes the backed-up copy from the device once the server has it.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;The version numbers matter here. Immich 3.3.6 shipped on 4 November 2026, and it carries a fix for CVE-2026-111497 — a critical buffer overflow in the Brotli decompression library. If you read a guide from 2025, it will predate this and it will tell you to run something with a known critical CVE in it.&lt;/p&gt;

&lt;h2&gt;
  
  
  The hardware it actually needs
&lt;/h2&gt;

&lt;p&gt;This is where most guides either lie or skip. Here's what the official requirements actually say:&lt;/p&gt;

&lt;div class="table-wrapper-paragraph"&gt;&lt;table&gt;
&lt;thead&gt;
&lt;tr&gt;
&lt;th&gt;Resource&lt;/th&gt;
&lt;th&gt;Minimum&lt;/th&gt;
&lt;th&gt;Recommended&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td&gt;RAM&lt;/td&gt;
&lt;td&gt;6 GB&lt;/td&gt;
&lt;td&gt;8 GB&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;CPU&lt;/td&gt;
&lt;td&gt;2 cores&lt;/td&gt;
&lt;td&gt;4 cores&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Disk&lt;/td&gt;
&lt;td&gt;Whatever your library needs&lt;/td&gt;
&lt;td&gt;SSD for the database&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Storage overhead&lt;/td&gt;
&lt;td&gt;—&lt;/td&gt;
&lt;td&gt;10 to 20% on top of your originals&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;

&lt;p&gt;The 10 to 20% is for thumbnails and transcoded video, and it surprised me too. A 100 GB library becomes 120 GB on the disk.&lt;/p&gt;

&lt;p&gt;Two more constraints that aren't obvious:&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;The database can't live on a network share.&lt;/strong&gt; The Postgres files have to sit on a local filesystem with proper user and group ownership — EXT4, ZFS, APFS, BTRFS, XFS all work. NTFS, exFAT and network mounts don't. The database itself is typically 1 to 3 GB, so give it local SSD space and let the photos live wherever you like.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Since version 3, the machine learning container on x86-64 needs at least the x86-64-v2 instruction set.&lt;/strong&gt; That's most CPUs built since around 2012. On an older machine, the machine learning features either fail to start or you run the ARM build.&lt;/p&gt;

&lt;h2&gt;
  
  
  Installing it
&lt;/h2&gt;

&lt;p&gt;The official quick start is a Docker Compose file and an environment file, and it takes about five minutes:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight shell"&gt;&lt;code&gt;mkdir -p ~/immich-app &lt;span class="o"&gt;&amp;amp;&amp;amp;&lt;/span&gt; &lt;span class="nb"&gt;cd&lt;/span&gt; ~/immich-app

&lt;span class="c"&gt;# The official compose file and environment template&lt;/span&gt;
wget -O docker-compose.yml \
  https://github.com/immich-app/immich/releases/latest/download/docker-compose.yml
wget -O .env \
  https://github.com/immich-app/immich/releases/latest/download/example.env

&lt;span class="c"&gt;# Set your upload location and a database password before starting&lt;/span&gt;
sed -i &lt;span class="s1"&gt;'s|^UPLOAD_LOCATION=.*|UPLOAD_LOCATION=./library|'&lt;/span&gt; .env
sed -i &lt;span class="s1"&gt;'s|^DB_PASSWORD=.*|DB_PASSWORD='&lt;/span&gt;&lt;span class="s2"&gt;"$(pwgen 24 1 -A)"&lt;/span&gt;&lt;span class="o"&gt;|&lt;/span&gt;&lt;span class="s1"&gt;' .env

docker compose up -d
&lt;/span&gt;&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;Then open &lt;code&gt;http://&amp;lt;your-server-ip&amp;gt;:2283&lt;/code&gt; and register. &lt;strong&gt;The first account you create becomes the administrator&lt;/strong&gt;, and Immich locks admin sign-up automatically after that first registration. Nobody else can register until you deliberately open it up again. If you expose this to the internet before you have claimed the admin role, you are one unlucky scan away from handing your server to a stranger — so keep port 2283 behind a firewall until the admin account exists.&lt;/p&gt;

&lt;h2&gt;
  
  
  The phone setup is where the friction is
&lt;/h2&gt;

&lt;p&gt;The apps are good. The permissions dance is not. Both platforms need one specific thing turned on, and if you skip it the backup silently does nothing:&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;On iOS&lt;/strong&gt;, enable &lt;strong&gt;Background App Refresh&lt;/strong&gt; — Settings, General, Background App Refresh, then Immich on. iOS decides for itself when to run the upload, so the more often your parent opens the app, the more often the backup happens.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;On Android&lt;/strong&gt;, you've got to exempt Immich from battery optimisation. Android's aggressive power management will kill the background worker and the upload stops with no error message. The community guide for this is literally called "Don't kill my app". If the phone is on a manufacturer's skin, expect a different path through the settings.&lt;/p&gt;

&lt;p&gt;If you have the phone connected over USB, you can check it and fix it from the shell:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight shell"&gt;&lt;code&gt;&lt;span class="c"&gt;# Is the app on the battery-optimisation ignore list?&lt;/span&gt;
adb shell dumpsys deviceidle whitelist &lt;span class="o"&gt;|&lt;/span&gt; grep -i immich

&lt;span class="c"&gt;# If it isn't, add it&lt;/span&gt;
adb shell dumpsys deviceidle whitelist +app.immich.android

&lt;span class="c"&gt;# Confirm the background worker is actually alive&lt;/span&gt;
adb shell dumpsys jobscheduler &lt;span class="o"&gt;|&lt;/span&gt; grep -A3 immich
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;Backups default to Wi-Fi only, which is the right default. Your parent's phone will not thank you for a 4K video uploading over mobile data.&lt;/p&gt;

&lt;h2&gt;
  
  
  The honest downsides
&lt;/h2&gt;

&lt;p&gt;&lt;strong&gt;It only works while it's reachable.&lt;/strong&gt; Google Photos works from a hotel in Spain. Immich on your home connection works from your kitchen. You need a real domain, a certificate, and a reverse proxy, or a tunnel. That is the single biggest quality-of-life difference between the two, and it is the reason I would not hand Immich to a parent who travels.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Pulling from iCloud eats mobile data.&lt;/strong&gt; If you select an iCloud album for backup, Immich pulls those assets from iCloud into a cache folder on the phone, then uploads them. On a metered connection, that is your data plan disappearing.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;The built-in backup only covers the database.&lt;/strong&gt; The database holds metadata and user records. The actual photos and videos in your upload location need their own backup, and a second one at that, ideally off-site. People discover this on the bad day.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;In my experience it's a lot of surface area for a photo album.&lt;/strong&gt; Anyone running a server for a hobby has learned that the interesting part is never the application. It is keeping the thing patched, the disk monitored and the certificate renewed.&lt;/p&gt;

&lt;h2&gt;
  
  
  When it is worth it
&lt;/h2&gt;

&lt;p&gt;The honest summary:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Worth it&lt;/strong&gt; if you have more photos than Google One's cheapest plan holds, if you want face recognition that Google doesn't hand you for free, or if the privacy matters to you.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Not worth it&lt;/strong&gt; if your library fits comfortably in the free 15 GB and the idea of running a public endpoint makes your stomach turn.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Not worth it yet&lt;/strong&gt; if the person using the phone travels or isn't technical — they will call you the first time the backup stops, and it will be because iOS didn't feel like running a background task.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;If you want the whole stack rather than just the photos, the &lt;a href="https://pragmaticsysadmin.help/meta/build-your-own-free-cloud-home-server-guide/" rel="noopener noreferrer"&gt;Build Your Own Cloud guide&lt;/a&gt; has the order to install it in, and the &lt;a href="https://pragmaticsysadmin.help/sysadmin/2026-11-10-cloud-for-cheap-the-honest-cost-math/" rel="noopener noreferrer"&gt;honest cost post&lt;/a&gt; works out what the disk and the electricity actually come to. A machine that can run Immich comfortably is not a machine you should buy on impulse — the RAM floor alone rules out most of the cheap boxes.&lt;/p&gt;

&lt;h2&gt;
  
  
  Before you start
&lt;/h2&gt;

&lt;p&gt;Three things to sort out first, in this order:&lt;/p&gt;

&lt;ol&gt;
&lt;li&gt;
&lt;strong&gt;How much data is this?&lt;/strong&gt; Google Takeout will export your whole library, and the download is slower and more painful than you expect. Do it before you commit.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Where does it live?&lt;/strong&gt; The upload location needs to be on a real filesystem with real space. Decide that before you install anything.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Who else touches it?&lt;/strong&gt; Every extra person with an account is another invitation to forget a password.&lt;/li&gt;
&lt;/ol&gt;

&lt;p&gt;Then install it, put the admin account in place, and only then open it to the family.&lt;/p&gt;

&lt;p&gt;&lt;em&gt;Self-hosting your photos is not saving money. It is deciding that the photos are yours, and then accepting what that costs.&lt;/em&gt;&lt;/p&gt;

</description>
      <category>selfhosted</category>
      <category>immich</category>
      <category>photos</category>
      <category>privacy</category>
    </item>
    <item>
      <title>Plex vs Jellyfin in 2026: The Paywall and the Hardware</title>
      <dc:creator>Jonne Riepponen</dc:creator>
      <pubDate>Thu, 12 Nov 2026 16:55:59 +0000</pubDate>
      <link>https://dev.to/jronegit/plex-vs-jellyfin-in-2026-the-paywall-and-the-hardware-4o87</link>
      <guid>https://dev.to/jronegit/plex-vs-jellyfin-in-2026-the-paywall-and-the-hardware-4o87</guid>
      <description>&lt;p&gt;My old Plex server stopped working for my parents the moment they left the house. Not broken — &lt;em&gt;working exactly as designed&lt;/em&gt;. Remote streaming had become a subscription, the person watching from abroad needed their own pass, and I had become the family's IT department for a media server I'd already paid for.&lt;/p&gt;

&lt;p&gt;That is the whole story of Plex in 2026, and it is why the answer for most people is Jellyfin. But the interesting part is what the paywall does and doesn't cover, and how much hardware a free media server actually needs. I have moved two households off Plex and the second was much easier, because I'd already made the mistakes on the first one.&lt;/p&gt;

&lt;h2&gt;
  
  
  What changed, and when
&lt;/h2&gt;

&lt;p&gt;Two dates matter here, and both are recent.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;April 2025&lt;/strong&gt; is when Plex removed free remote playback. Before that, streaming your own library from anywhere was included in the free server. After it, remote streaming needs either a Plex Pass on the account that owns the server, or a separate Remote Watch Pass on each viewer's account. Plex's own support documentation is blunt about the second one: it "only enables remote streaming of personal content and does not provide access to any other premium features."&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;1 July 2026&lt;/strong&gt; is when the price of a new Lifetime Plex Pass went from $249.99 to $749.99. Plex says existing lifetime holders keep everything, which is fair, but it means the "buy it once" option now costs three times what it did.&lt;/p&gt;

&lt;p&gt;Jellyfin, meanwhile, was forked from Emby in 2018 and has never had a paid tier. No subscription, no account, no premium unlock. The server, every official client, hardware transcoding, live TV and recording are all free.&lt;/p&gt;

&lt;h2&gt;
  
  
  The price list
&lt;/h2&gt;

&lt;p&gt;Checked against Plex's own plans page on 10 November 2026:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Jellyfin&lt;/strong&gt; — $0.00. No paid tier, no account, no telemetry, no feature behind a paywall.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Plex Pass&lt;/strong&gt; — $6.99 a month, $69.99 a year, or $749.99 for the lifetime.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Remote Watch Pass&lt;/strong&gt; — $2.99 a month or $29.99 a year, remote streaming only, for someone watching a server they don't own.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Plex free&lt;/strong&gt; — still free, but local streaming only. Your phone on the same Wi-Fi, and nothing else.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;So the honest comparison for somebody whose parents watch from another country: Plex costs $6.99 a month forever, or $2.99 a month per viewer with the cheaper pass, or $749.99 once. Jellyfin costs nothing and your time, which for me was a weekend and then never again.&lt;/p&gt;

&lt;h2&gt;
  
  
  What Plex Pass actually unlocks
&lt;/h2&gt;

&lt;p&gt;This is the part worth reading twice, because it's longer than most people assume:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;Remote streaming, for everyone on the server&lt;/li&gt;
&lt;li&gt;Hardware transcoding — offloading the codec work to the GPU&lt;/li&gt;
&lt;li&gt;Live TV and DVR recording, with a tuner&lt;/li&gt;
&lt;li&gt;Offline sync, downloads for offline viewing&lt;/li&gt;
&lt;li&gt;Skip Intro and Skip Credits&lt;/li&gt;
&lt;li&gt;Parental controls and the Plex Dash monitoring app&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;Against that, Jellyfin gives you all of it, free, except that you have to set up the network path yourself.&lt;/p&gt;

&lt;p&gt;That last part is the actual trade. Plex runs a relay service so you don't have to open a port on your router. Jellyfin doesn't run anything for you, so remote access means port forwarding, a reverse proxy with a real certificate, or a VPN back to your house. It's maybe half an hour of work and it's yours forever. I did it with a reverse proxy and a Let's Encrypt certificate, and it's been running for two years without a single interruption.&lt;/p&gt;

&lt;h2&gt;
  
  
  The hardware question
&lt;/h2&gt;

&lt;p&gt;People assume a media server needs a big machine. It doesn't, unless you want to transcode.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Direct play&lt;/strong&gt; means the phone or TV plays the file exactly as it is stored. No processing, no GPU, almost no power. A Raspberry Pi 4 or 5 handles direct play for a normal family's library without breaking a sweat, and mine has been doing exactly that for years.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Transcoding&lt;/strong&gt; happens when the client can't handle the codec, the bitrate is too high for the connection, or subtitles have to be burned in. That's when you want a GPU.&lt;/p&gt;

&lt;p&gt;The practical guidance from Jellyfin's own hardware notes, and the reason I like this project:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;An &lt;strong&gt;Intel N100 mini PC&lt;/strong&gt; handles 8 to 12 simultaneous 1080p hardware transcodes. The chip's integrated GPU does Intel Quick Sync with H.264, HEVC and AV1 decode and encode.&lt;/li&gt;
&lt;li&gt;A &lt;strong&gt;Pi 4 or 5&lt;/strong&gt; is fine for direct play only.&lt;/li&gt;
&lt;li&gt;Jellyfin's guide recommends against AMD integrated graphics, and suggests an Intel Core i5-11400, a Pentium Gold G7400, an N100, or Apple silicon if you want to go further.&lt;/li&gt;
&lt;li&gt;Jokes aside, the guide also wants 8 GB of RAM and a 100 GB SSD for the config and metadata.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;Jellyfin supports Intel QSV, NVIDIA NVENC, AMD AMF, VA-API on Linux, Apple VideoToolink on macOS and Rockchip RKMPP. On an N100 box that's free hardware acceleration. On Plex, hardware transcoding is behind Plex Pass — so the same box costs you $70 a year, or $249.99 for the five-year pass, to do the identical job.&lt;/p&gt;

&lt;h2&gt;
  
  
  The privacy difference
&lt;/h2&gt;

&lt;p&gt;Plex requires a plex.tv account, and authentication routes through Plex's own cloud servers even for local-only playback. That's not a bug — it's how a hosted service works — but it does mean the operator of your media server is talking to Plex on every playback.&lt;/p&gt;

&lt;p&gt;Jellyfin runs no servers for its users. There is no account to create, no telemetry to opt out of, and nothing phoned home. The only traffic is between your devices and your own server.&lt;/p&gt;

&lt;p&gt;Neither project will ever get into your files. But one of them gets a log of when you watch things.&lt;/p&gt;

&lt;h2&gt;
  
  
  The thing Jellyfin does not replace
&lt;/h2&gt;

&lt;p&gt;Let me be blunt, because this is where "replace Netflix with Jellyfin" advice goes wrong.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Jellyfin does not replace Netflix.&lt;/strong&gt; It streams the media you already own. If you don't own the films, Jellyfin has nothing to play, and neither does any other self-hosted server on earth. A 4 TB disk of your own library is worth more than a catalogue you rent.&lt;/p&gt;

&lt;p&gt;What Jellyfin replaces is the &lt;strong&gt;software&lt;/strong&gt;. The Plex paywall, the account, the client apps and the "your server is offline" messages. That's a real saving, and for the price of a weekend it's an enormous one.&lt;/p&gt;

&lt;p&gt;The same logic applies to the rest of this site. If you're weighing up replacing a subscription with a self-hosted box, the honest question is whether you were going to use it anyway. Sometimes the answer is yes and the saving is enormous. Sometimes the answer is that you never used it and the saving is zero — in which case, cancel.&lt;/p&gt;

&lt;h2&gt;
  
  
  Running it
&lt;/h2&gt;

&lt;p&gt;If you want the transcoding, this is the whole thing — the &lt;code&gt;/dev/dri&lt;/code&gt; mount is what lets Jellyfin reach your Intel GPU for Quick Sync:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight shell"&gt;&lt;code&gt;docker run -d --name jellyfin \
  --restart unless-stopped \
  -v /opt/jellyfin/config:/config \
  -v /mnt/media/movies:/media/movies:ro \
  -v /mnt/media/tv:/media/tv:ro \
  -p 8096:8096 \
  --device /dev/dri:/dev/dri \
  --group-add &lt;span class="s2"&gt;"$(getent group render | cut -d: -f3)"&lt;/span&gt; \
  jellyfin/jellyfin:latest
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;Then open &lt;code&gt;http://&amp;lt;server-ip&amp;gt;:8096&lt;/code&gt;, create the admin account, and turn on the hardware transcoder under Dashboard, Playback, Transcoding. I skipped that step on my first install and spent an evening wondering why a 4K film stuttered on the TV while the same file played fine on the laptop.&lt;/p&gt;

&lt;h2&gt;
  
  
  Migrating from Plex
&lt;/h2&gt;

&lt;p&gt;This part is genuinely easy, which surprises people who've spent years inside Plex's ecosystem.&lt;/p&gt;

&lt;ol&gt;
&lt;li&gt;
&lt;strong&gt;Point Jellyfin at the same folders.&lt;/strong&gt; Plex's library structure carries over. Your movies, your show metadata, your artwork — all of it.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Let Jellyfin fetch its own metadata.&lt;/strong&gt; It scans the same files and pulls from the same sources. Give it an evening.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Point the clients at the new server.&lt;/strong&gt; Jellyfin has official apps for the platforms Plex covers.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Sort out the remote path.&lt;/strong&gt; Port forward 8096, or put a reverse proxy with a certificate in front of it.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Keep Plex running for a fortnight&lt;/strong&gt; before you cancel anything. Not because it will be missed, but because you will find something.&lt;/li&gt;
&lt;/ol&gt;

&lt;p&gt;The clients are the only real difference, and it's a small one. Jellyfin's smart TV apps are good rather than beautiful, and its interface takes an afternoon to make sense of. Plex is a bit more polished. Neither is a reason to pay a subscription forever.&lt;/p&gt;

&lt;h2&gt;
  
  
  What to do next
&lt;/h2&gt;

&lt;p&gt;If you want the whole home-lab picture rather than just the media server, the &lt;a href="https://pragmaticsysadmin.help/meta/build-your-own-free-cloud-home-server-guide/" rel="noopener noreferrer"&gt;Build Your Own Cloud guide&lt;/a&gt; has the stack and the order to install it in, and the &lt;a href="https://pragmaticsysadmin.help/sysadmin/2026-11-10-cloud-for-cheap-the-honest-cost-math/" rel="noopener noreferrer"&gt;honest cost post&lt;/a&gt; works out what the box and the electricity come to.&lt;/p&gt;

&lt;p&gt;If you want the short version of the hardware decision:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Direct play only&lt;/strong&gt; — a Pi, an old laptop, whatever is in the cupboard.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Transcoding&lt;/strong&gt; — an N100 mini PC and Jellyfin, and that's the whole bill.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;Check your library before you commit. Run the hardware through Jellyfin's own transcoding guide with a few of your actual files, not a demo clip. The format that breaks everything is the one you own.&lt;/p&gt;

&lt;p&gt;&lt;em&gt;Free software and a free subscription are different things. One of them stops working when the company changes its mind.&lt;/em&gt;&lt;/p&gt;

</description>
      <category>selfhosted</category>
      <category>streaming</category>
      <category>mediaserver</category>
      <category>homelab</category>
    </item>
    <item>
      <title>Pi-hole vs AdGuard Home: Which One Should You Self-Host?</title>
      <dc:creator>Jonne Riepponen</dc:creator>
      <pubDate>Thu, 12 Nov 2026 16:55:24 +0000</pubDate>
      <link>https://dev.to/jronegit/pi-hole-vs-adguard-home-which-one-should-you-self-host-3pdn</link>
      <guid>https://dev.to/jronegit/pi-hole-vs-adguard-home-which-one-should-you-self-host-3pdn</guid>
      <description>&lt;p&gt;I ran both on the same network for a month in 2025, switched one to the other, and then switched back. Neither is a trap. One of them is just a much better fit for a house with a child's tablet in it, and the difference comes down to a single checkbox that one project ships with and the other one makes you build.&lt;/p&gt;

&lt;p&gt;Here's the honest comparison, checked against both projects' own releases and documentation on 10 November 2026.&lt;/p&gt;

&lt;h2&gt;
  
  
  What both of them actually do
&lt;/h2&gt;

&lt;p&gt;A DNS sinkhole. Your devices ask it instead of asking your ISP, it answers from a local cache, and it refuses to resolve the domains on a blocklist. Ads and trackers never even get a connection, which is why it works on the smart TV and the Android apps where browser extensions do nothing.&lt;/p&gt;

&lt;p&gt;You point your router's DHCP at it and every device on the network is covered. No client software, no per-device setup, no app to install on your parent's phone.&lt;/p&gt;

&lt;p&gt;And here is the ceiling, which I want to be upfront about because every advert for these tools ignores it: &lt;strong&gt;neither of them blocks in-app ads.&lt;/strong&gt; The ads inside YouTube, Twitch and Facebook are served from the same domains as the content itself, so no DNS filter can separate them. What you actually get is third-party ad domains, trackers, telemetry and malware domains blocked. That is a real saving, and it is not "no more ads".&lt;/p&gt;

&lt;h2&gt;
  
  
  The difference that matters: encrypted DNS
&lt;/h2&gt;

&lt;p&gt;This is where the choice gets decided.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Pi-hole speaks plain DNS on port 53, full stop.&lt;/strong&gt; It cannot answer clients over DoH, DoT or DoQ. If you want your upstream queries encrypted, you add a sidecar — &lt;code&gt;unbound&lt;/code&gt; for recursive resolution, or &lt;code&gt;cloudflared&lt;/code&gt; for a DoH proxy. Both are well documented and both work. Pi-hole's own cloudflared guide now opens with a warning that Cloudflare deprecated the proxy-dns feature it relies on, and that new installs by that route aren't recommended. So you are adding a second service to deploy, monitor and patch, and in one case the recommended route is deprecated.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;AdGuard Home ships encrypted DNS in both directions, natively.&lt;/strong&gt; DoH, DoT, DoQ and DNSCrypt, outbound to your upstream and inbound for your own clients, all from the first-run wizard. The default upstream is Quad9 over DoH, so your queries leave the house encrypted before you've touched anything. It can also act as an encrypted DNS server for clients outside your network once you've attached a certificate.&lt;/p&gt;

&lt;p&gt;That difference is the whole argument for a new installation, and it's the reason I moved my own house across. A plain Pi-hole on a home LAN still hands your ISP a list of every domain my household looked up. Adding &lt;code&gt;unbound&lt;/code&gt; fixes it, and you should, but it's twenty minutes of extra work plus a service that can break on its own.&lt;/p&gt;

&lt;h2&gt;
  
  
  Per-device control, the feature you'll actually use
&lt;/h2&gt;

&lt;p&gt;Pi-hole has groups: lists, allowlists, denylists and regex rules belong to groups, and clients belong to groups. That's a proper structure and it's more flexible than it sounds. It has no SafeSearch enforcement, no parental-control presets and no scheduled service blocking.&lt;/p&gt;

&lt;p&gt;AdGuard Home treats clients as first-class objects. You identify a client by IP, CIDR range, MAC address or ClientID, and each one gets its own upstream resolvers, blocklists, forced SafeSearch on Google, YouTube, Bing and DuckDuckGo, the adult-content filter, and scheduled service blocking.&lt;/p&gt;

&lt;p&gt;So "block TikTok and YouTube on the kids' tablet on school nights" is five clicks and a schedule in AdGuard Home. In Pi-hole it's a hand-written regex, or a local DNS record and a cron job. Both are doable. Only one of them is something I'll actually do at 21:00 on a Tuesday.&lt;/p&gt;

&lt;h2&gt;
  
  
  The install, and what it costs you
&lt;/h2&gt;

&lt;p&gt;Neither is heavy. DNS blocking is a light workload and both are happy on the smallest box you can find.&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Pi-hole v6.5&lt;/strong&gt; — core released 17 October 2026, licensed EUPL-1.2, written in C around the &lt;code&gt;pihole-FTL&lt;/code&gt; resolver. Documented floor of 512 MB of RAM and 2 GB of disk, with 4 GB recommended. Around 50 to 100 MB of RAM at idle. Container is the officially supported Docker route, and the installer is the famous one-liner.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;AdGuard Home v0.107.79&lt;/strong&gt; — released 18 August 2026, licensed GPL-3.0, written in Go as a single self-contained binary. No stated minimum, and it uses roughly 30 to 80 MB of RAM at idle. There's a first-run wizard on port 3000, and there's genuinely nothing else to wire together.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Both&lt;/strong&gt; ship x86 and ARM builds, so a Raspberry Pi is fine and so is a spare x86 box.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Both&lt;/strong&gt; can serve DHCP themselves, though leaving DHCP to your router and only handing out the DNS address is usually less painful.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;I ran the Pi-hole on a Pi 5 for years and it never once used more than a rounding error of the machine. The box you already have will do.&lt;/p&gt;

&lt;h2&gt;
  
  
  The community and the blocklists
&lt;/h2&gt;

&lt;p&gt;This is Pi-hole's real strength and it isn't a small one. It has roughly 48,000 GitHub stars against AdGuard Home's 24,000, a decade of guides, and the deepest blocklist ecosystem of the two. There are community lists for German telcos, for privacy, for specific apps, for specific people's annoyances. If the thing you want to block exists as a Pi-hole list, it probably exists.&lt;/p&gt;

&lt;p&gt;AdGuard Home has a smaller but coherent ecosystem, and it uses AdGuard's own filtering syntax with modifiers that a Pi-hole regex can only approximate.&lt;/p&gt;

&lt;p&gt;Log retention differs too: Pi-hole keeps a year of SQLite query history, AdGuard Home keeps 90 days by default.&lt;/p&gt;

&lt;h2&gt;
  
  
  Which one to pick
&lt;/h2&gt;

&lt;p&gt;Here's what I'd tell a friend choosing today.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Choose AdGuard Home if&lt;/strong&gt; you're starting fresh, you want encrypted DNS without a sidecar project, or you have children and want per-device control with schedules. In my experience that's most people, and it's why I'd install it on a new network without hesitating.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Choose Pi-hole if&lt;/strong&gt; you want the bigger community and the deeper list ecosystem, you already know it, or you're running it on a Raspberry Pi inside your own LAN where plain DNS on port 53 is fine and you want the decade of accumulated knowledge. It is not a worse product. It's a different bet.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Choose either if&lt;/strong&gt; you want a fifth of your DNS traffic to stop leaving the house. That's the part that actually works, and it's worth doing.&lt;/p&gt;

&lt;h2&gt;
  
  
  Setting it up on your network
&lt;/h2&gt;

&lt;p&gt;Whichever you pick, the mechanics are the same. On AdGuard Home the setup is genuinely one command and a browser:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight shell"&gt;&lt;code&gt;&lt;span class="c"&gt;# AdGuard Home: single binary, no dependencies&lt;/span&gt;
docker run --name adguardhome \
  --restart unless-stopped \
  -v /opt/adguardhome/work:/opt/adguardhome/work \
  -v /opt/adguardhome/conf:/opt/adguardhome/conf \
  -p 3000:3000 \
  -p 53:53/tcp -p 53:53/udp \
  --dns=0.0.0.0 \
  adguard/adguardhome:latest
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;Then open &lt;code&gt;http://&amp;lt;server-ip&amp;gt;:3000&lt;/code&gt;, set the upstream resolvers, and point your router's DHCP at the box.&lt;/p&gt;

&lt;p&gt;If you go the Pi-hole way with encrypted upstreams, the extra piece is &lt;code&gt;unbound&lt;/code&gt; sitting in front:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight shell"&gt;&lt;code&gt;&lt;span class="c"&gt;# Pi-hole with a local recursive resolver in front of it&lt;/span&gt;
docker run -d --name unbound --restart unless-stopped \
  -v /opt/unbound:/opt/unbound \
  -p 53:53/tcp -p 53:53/udp \
  --cap-add=NET_ADMIN \
  ghcr.io/kowith/pihole-unbound:latest
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;h2&gt;
  
  
  The honest gotchas
&lt;/h2&gt;

&lt;p&gt;Six things I'd tell my past self, all of which cost me an afternoon each:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Give the server a stable address.&lt;/strong&gt; A DHCP reservation in your router, not a static IP you invented.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Don't run two of them at once.&lt;/strong&gt; The second one silently wins the port 53 race.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Learn to read the query log.&lt;/strong&gt; It is the difference between a two-minute fix and an evening.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Expect a site to break.&lt;/strong&gt; Blocking an authentication domain gives you a login loop that looks like a broken router.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Keep a way in.&lt;/strong&gt; Know how to comment out a rule from the console before you need it.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Update them.&lt;/strong&gt; Both are maintained projects, and an old sinkhole is a sinkhole with a known bug in it.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;&lt;strong&gt;Both need a stable address.&lt;/strong&gt; Give the server a DHCP reservation in your router, because a sinkhole that changes IP is a house full of devices quietly falling back to your ISP's DNS — which looks exactly like the software not working.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Don't run two of them.&lt;/strong&gt; You will, at least once, and the second one will silently win the port 53 race and half your network will bypass the first.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;A DNS filter can take a site down.&lt;/strong&gt; If you block an authentication domain, you get a login loop that looks like a broken router. Know how to comment out a rule from the console before you need it.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Check the query log before you trust it.&lt;/strong&gt; Both keep a record of what was blocked. When something doesn't load, the log tells you which rule did it, which is the difference between a two-minute fix and an evening.&lt;/p&gt;

&lt;h2&gt;
  
  
  What to do next
&lt;/h2&gt;

&lt;p&gt;If you want the whole home-lab picture rather than just the DNS piece, the &lt;a href="https://pragmaticsysadmin.help/meta/build-your-own-free-cloud-home-server-guide/" rel="noopener noreferrer"&gt;Build Your Own Cloud guide&lt;/a&gt; has the stack and the order to install it in, and the &lt;a href="https://pragmaticsysadmin.help/sysadmin/2026-11-10-cloud-for-cheap-the-honest-cost-math/" rel="noopener noreferrer"&gt;honest cost post&lt;/a&gt; works out what the box and the electricity come to. A DNS sinkhole is a good first service, because it's cheap, it's visible on every device, and it breaks in ways that teach you how your network actually works.&lt;/p&gt;

&lt;p&gt;And when you set it up, point a phone at it for an afternoon before you point the router at it. Watching a phone go from a dozen trackers to two is the fastest way to understand what the thing is actually doing.&lt;/p&gt;

&lt;p&gt;&lt;em&gt;Either one is a good decision. The wrong one is the one you didn't install because you couldn't tell them apart.&lt;/em&gt;&lt;/p&gt;

</description>
      <category>selfhosted</category>
      <category>privacy</category>
      <category>networking</category>
      <category>homelab</category>
    </item>
    <item>
      <title>Cloud for Cheap: What It Actually Costs to Run Your Own</title>
      <dc:creator>Jonne Riepponen</dc:creator>
      <pubDate>Thu, 12 Nov 2026 16:54:49 +0000</pubDate>
      <link>https://dev.to/jronegit/cloud-for-cheap-what-it-actually-costs-to-run-your-own-3ko4</link>
      <guid>https://dev.to/jronegit/cloud-for-cheap-what-it-actually-costs-to-run-your-own-3ko4</guid>
      <description>&lt;p&gt;I did the arithmetic on a Google One bill last month, then did the same arithmetic on the server sitting under my desk. The cloud won. Not narrowly — by enough that I closed the tab and went to make coffee.&lt;/p&gt;

&lt;p&gt;That's the answer most self-hosting guides won't give you, because "run your own cloud" is a hobby that sells a feeling, and the feeling is free while the electricity isn't.&lt;/p&gt;

&lt;p&gt;Here's the real arithmetic, with every price checked on 10 November 2026: what Google One actually charges, what a cheap VPS actually costs, what a home server burns in electricity, and the one rule that tells you which of the three to pick.&lt;/p&gt;

&lt;h2&gt;
  
  
  Every option, priced
&lt;/h2&gt;

&lt;p&gt;One table, because the whole argument is in it. The Google One and Hetzner prices were checked against the vendors' own pages on 10 November 2026; the electricity figures are calculated at the German household rate of 36.18 euro cents per kWh, which is what Eurostat recorded for the first half of 2026.&lt;/p&gt;

&lt;div class="table-wrapper-paragraph"&gt;&lt;table&gt;
&lt;thead&gt;
&lt;tr&gt;
&lt;th&gt;Option&lt;/th&gt;
&lt;th&gt;What you get&lt;/th&gt;
&lt;th&gt;Monthly&lt;/th&gt;
&lt;th&gt;Yearly&lt;/th&gt;
&lt;th&gt;Over five years&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td&gt;Google One Free&lt;/td&gt;
&lt;td&gt;15 GB, shared&lt;/td&gt;
&lt;td&gt;$0.00&lt;/td&gt;
&lt;td&gt;$0.00&lt;/td&gt;
&lt;td&gt;$0.00&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Google One Basic&lt;/td&gt;
&lt;td&gt;100 GB, shared&lt;/td&gt;
&lt;td&gt;$1.99&lt;/td&gt;
&lt;td&gt;$19.99&lt;/td&gt;
&lt;td&gt;$99.95&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Google One AI Plus&lt;/td&gt;
&lt;td&gt;2 TB, shared&lt;/td&gt;
&lt;td&gt;$9.99&lt;/td&gt;
&lt;td&gt;$99.99&lt;/td&gt;
&lt;td&gt;$499.95&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Google One AI Pro&lt;/td&gt;
&lt;td&gt;5 TB, shared&lt;/td&gt;
&lt;td&gt;$19.99&lt;/td&gt;
&lt;td&gt;$199.99&lt;/td&gt;
&lt;td&gt;$999.95&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Hetzner CX23&lt;/td&gt;
&lt;td&gt;2 vCPU, 4 GB, 40 GB NVMe&lt;/td&gt;
&lt;td&gt;€5.99&lt;/td&gt;
&lt;td&gt;€71.88&lt;/td&gt;
&lt;td&gt;€359.40&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Hetzner CX33&lt;/td&gt;
&lt;td&gt;4 vCPU, 8 GB, 80 GB NVMe&lt;/td&gt;
&lt;td&gt;€8.99&lt;/td&gt;
&lt;td&gt;€107.88&lt;/td&gt;
&lt;td&gt;€539.40&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Pi 5 and an 8 TB disk, home&lt;/td&gt;
&lt;td&gt;3.0 W idle, electricity only&lt;/td&gt;
&lt;td&gt;—&lt;/td&gt;
&lt;td&gt;€9.51&lt;/td&gt;
&lt;td&gt;€47.55&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;N100 mini PC, home&lt;/td&gt;
&lt;td&gt;10 W idle, electricity only&lt;/td&gt;
&lt;td&gt;—&lt;/td&gt;
&lt;td&gt;€31.70&lt;/td&gt;
&lt;td&gt;€158.50&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Old desktop tower, home&lt;/td&gt;
&lt;td&gt;50 W idle, electricity only&lt;/td&gt;
&lt;td&gt;—&lt;/td&gt;
&lt;td&gt;€158.46&lt;/td&gt;
&lt;td&gt;€792.30&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;

&lt;p&gt;Read the last three rows next to the first four and the argument is already over for most people. A desktop tower left running all day costs more over five years than Google One 2TB does. That is not a self-hosting setup, that is an expensive way to store photographs.&lt;/p&gt;

&lt;h2&gt;
  
  
  Where the electricity comes from
&lt;/h2&gt;

&lt;p&gt;The bottom three rows are the ones nobody puts in a spreadsheet, so here is the calculation. German households paid 36.18 euro cents per kWh in the first half of 2026 — the second most expensive rate in the EU, which makes the maths harsher than most guides admit.&lt;/p&gt;

&lt;p&gt;A Raspberry Pi 5 idles at about 3.0 W headless, or 3.6 W with Ethernet and a USB SSD attached. A cheap N100 mini PC idles around three times that. A refurbished office box from 2020 can sit at 18 W doing nothing at all.&lt;/p&gt;

&lt;p&gt;Here is the whole calculation in one line: &lt;strong&gt;at 36.18 cents per kWh, every watt of continuous draw costs €3.17 a year.&lt;/strong&gt; A machine that idles at 50 W is spending more on electricity than a cheap VPS costs for the whole month.&lt;/p&gt;

&lt;p&gt;Measure your own machine instead of trusting someone else's table. On a Pi, the power supply reports real numbers:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight shell"&gt;&lt;code&gt;&lt;span class="c"&gt;# Instantaneous watts on a Raspberry Pi 5&lt;/span&gt;
cat /sys/class/power_supply/*/power/now &lt;span class="o"&gt;|&lt;/span&gt; awk &lt;span class="s1"&gt;'{printf "%.2f W\n", $1/1000000}'&lt;/span&gt;

&lt;span class="c"&gt;# On a desktop with a supported kernel: total and per-process draw&lt;/span&gt;
sudo powertop --interval=2
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;And if you want the annual figure for your own hardware instead of mine:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight shell"&gt;&lt;code&gt;&lt;span class="c"&gt;#!/usr/bin/env bash&lt;/span&gt;
&lt;span class="c"&gt;# Annual electricity cost of a machine that draws $1 W continuously.&lt;/span&gt;
&lt;span class="c"&gt;# Usage: ./power-cost.sh 3.0          # watts, 36.18 cents/kWh&lt;/span&gt;
&lt;span class="c"&gt;#        ./power-cost.sh 18.0 0.30    # watts, your own price per kWh&lt;/span&gt;
awk -v &lt;span class="nv"&gt;w&lt;/span&gt;&lt;span class="o"&gt;=&lt;/span&gt;&lt;span class="s2"&gt;"${1:-3.0}"&lt;/span&gt; -v &lt;span class="nv"&gt;p&lt;/span&gt;&lt;span class="o"&gt;=&lt;/span&gt;&lt;span class="s2"&gt;"${2:-0.3618}"&lt;/span&gt; &lt;span class="s1"&gt;'BEGIN {
  kwh = w * 8760 / 1000
  printf "%.1f W -&amp;gt; %.1f kWh/year -&amp;gt; EUR %.2f/year\n", w, kwh, kwh * p
}'&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;h2&gt;
  
  
  What a cheap VPS actually costs
&lt;/h2&gt;

&lt;p&gt;Hetzner raised its cloud prices twice in 2026. On 1 April most lines went up about a third, and on 15 June every line went up again for new orders and rescales. The cost-optimised CX23 went from €2.99 to €3.99 to €5.49 a month.&lt;/p&gt;

&lt;p&gt;Compare that with DigitalOcean's comparable Droplet at $24 a month and the gap is embarrassing in Hetzner's favour.&lt;/p&gt;

&lt;p&gt;But note the catch: the cheap lines are capacity-limited. Since September 2026 the cost-optimised plans are frequently marked unavailable and sell out within minutes of a restock. Budget for the possibility that your €5.99 server isn't there when you want it.&lt;/p&gt;

&lt;h2&gt;
  
  
  The three ways to run your own cloud
&lt;/h2&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Home server&lt;/strong&gt; — a machine you own, in your house. Cheapest in electricity, most expensive in your time.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Cheap VPS&lt;/strong&gt; — a rented box in a datacentre. One flat monthly price, nobody else's hardware in your hallway.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Nothing at all&lt;/strong&gt; — sync between your own devices, no server involved. Costs nothing at all.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;The third option is the one nobody sells you, and for a lot of people it's the correct one. Syncthing is free, open source, and it syncs folders directly between your laptop, your phone and your parent's laptop without a server in the middle. If what you need is "my photos are on my phone and I want them on my laptop", you don't need a cloud. You need a peer-to-peer sync tool and a free afternoon.&lt;/p&gt;

&lt;h2&gt;
  
  
  Where self-hosting genuinely wins
&lt;/h2&gt;

&lt;p&gt;It wins when you're billed per seat and your files are measured in gigabytes, not terabytes.&lt;/p&gt;

&lt;p&gt;A ten-person team with a shared document library doesn't need a storage plan, it needs a box. And the box costs the same whether ten people use it or fifty. Google Workspace Business Standard is roughly $14 per user per month on an annual plan — about $140 a month for that team, and it climbs with every hire. One CX33 covers the same library and the price doesn't move.&lt;/p&gt;

&lt;p&gt;It also wins when privacy is the point and price isn't. A password manager, a photo library, a calendar — those aren't storage problems, they're trust problems, and no amount of money buys the trust you get from a box in your hallway.&lt;/p&gt;

&lt;h2&gt;
  
  
  Where it never wins
&lt;/h2&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Cheap bulk storage.&lt;/strong&gt; Google One 2TB is $9.99. You aren't beating that with a VPS.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Zero administration.&lt;/strong&gt; You're buying an unpaid job. Every disk that fills, every certificate that expires, every backup you never tested is now yours.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Anything that needs a phone number.&lt;/strong&gt; A mail server on a residential connection will get you blocked, and the block will outlast your patience.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;When the apps already work.&lt;/strong&gt; The most underrated saving in the whole self-hosting world is the subscription you didn't cancel.&lt;/li&gt;
&lt;/ul&gt;

&lt;h2&gt;
  
  
  The costs nobody puts in the spreadsheet
&lt;/h2&gt;

&lt;p&gt;Electricity is the smallest of the three real costs. The other two are bigger and they're both paid in the same currency: your time.&lt;/p&gt;

&lt;p&gt;The 3am restore. Every self-hosted system has a bad day, and the difference between a bad day and a lost weekend is whether you tested the backup. A backup you've never restored is a hypothesis, not a backup.&lt;/p&gt;

&lt;p&gt;The update treadmill. Six services means six things to patch, and the one you don't patch is the one with the advisory on it. This is the cost that makes a modest cloud bill look cheap.&lt;/p&gt;

&lt;p&gt;The migration. Moving a photo library onto a new server takes an evening, and the evening after that you find the folder structure doesn't match what you expected and you start again.&lt;/p&gt;

&lt;h2&gt;
  
  
  What to do next
&lt;/h2&gt;

&lt;p&gt;If the goal is &lt;strong&gt;saving money&lt;/strong&gt;, start with the free option: &lt;a href="https://pragmaticsysadmin.help/sysadmin/self-hosting-isnt-free-honest-cost-running-own-server/" rel="noopener noreferrer"&gt;Syncthing between your own devices&lt;/a&gt; costs nothing and needs no server. Then look at what you actually pay for, and whether that subscription is per-seat or per-terabyte.&lt;/p&gt;

&lt;p&gt;If the goal is &lt;strong&gt;owning your data&lt;/strong&gt;, the &lt;a href="https://pragmaticsysadmin.help/meta/build-your-own-free-cloud-home-server-guide/" rel="noopener noreferrer"&gt;Build Your Own Cloud guide&lt;/a&gt; has the stack and the order to install it in. Budget for a backup disk, and for the evening.&lt;/p&gt;

&lt;p&gt;If the goal is &lt;strong&gt;learning&lt;/strong&gt;, the &lt;a href="https://pragmaticsysadmin.help/sysadmin/2025-10-29-setting-up-a-home-lab-a-beginner-s-guide/" rel="noopener noreferrer"&gt;home lab beginner's guide&lt;/a&gt; is the gentler version of the same question.&lt;/p&gt;

&lt;p&gt;Don't buy the hardware until you've done the arithmetic above for your own machine. The number that decides it is how many watts it draws while idle, and that's a number you can read in five minutes with a €12 plug meter.&lt;/p&gt;

&lt;p&gt;&lt;em&gt;Self-hosting isn't a cheaper cloud. It's a different kind of bill — sometimes smaller, sometimes much larger, and always paid in your own time.&lt;/em&gt;&lt;/p&gt;

</description>
      <category>selfhosted</category>
      <category>cloud</category>
      <category>cost</category>
      <category>homelab</category>
    </item>
    <item>
      <title>AI Is a Fast Helper, Not the Truth</title>
      <dc:creator>Jonne Riepponen</dc:creator>
      <pubDate>Tue, 10 Nov 2026 16:31:34 +0000</pubDate>
      <link>https://dev.to/jronegit/ai-is-a-fast-helper-not-the-truth-2ob4</link>
      <guid>https://dev.to/jronegit/ai-is-a-fast-helper-not-the-truth-2ob4</guid>
      <description>&lt;p&gt;You see a message from your mom: “The bank says I owe $500, here’s the proof.” She attached a picture that looks like a bank statement, but it was made by an AI. You feel a pang of worry.&lt;/p&gt;

&lt;p&gt;You’ve been there before — an AI chatbot gave a support ticket number that turned out to be fake, and nobody noticed for 3 days. The confidence in the answer made it feel real.&lt;/p&gt;

&lt;p&gt;That’s the problem: AI sounds sure even when it’s wrong. Confidence is not proof. If you act on it without checking, you could move money, change a medication, or miss a real issue.&lt;/p&gt;

&lt;p&gt;The simple test is to ask “how do you know that?” If the AI can’t point to a source, treat it as a guess.&lt;/p&gt;

&lt;h2&gt;
  
  
  Treat every AI answer as a draft, not a final decision
&lt;/h2&gt;

&lt;p&gt;AI is good at pulling together words fast. It can give you an outline for a letter, a rough idea for a gift, or a summary of a long article. That’s useful. What it isn’t good at is guaranteeing that every fact it states is true. Think of the output as a first draft that needs a human editor. You wouldn’t send a rough draft to your boss without reading it, and you shouldn’t act on an AI answer without checking it.&lt;/p&gt;

&lt;h2&gt;
  
  
  The ticket-number trick: why fake details feel real
&lt;/h2&gt;

&lt;p&gt;A chatbot once gave a customer a ticket number that did not exist. Nobody noticed for 3 days because the number looked plausible and the chatbot sounded certain. When you see a specific number — ticket, reference, account — your brain treats it as evidence. AI knows how to generate those details, but it doesn’t have a database to pull them from. It’s making them up on the spot, and the confidence tricks you into believing they’re real.&lt;/p&gt;

&lt;h2&gt;
  
  
  Photos lie: AI-generated images are not evidence
&lt;/h2&gt;

&lt;p&gt;People forward AI-made photos as if they were real evidence. A picture made by AI proves nothing about the world outside the model. If your parent shows you an AI‑generated “bank statement” or a “doctor’s note,” you need to treat it like a rumor, not a document. The only way to know if it’s real is to go to the source — log in to the bank’s website, call the clinic, or ask for a paper copy.&lt;/p&gt;

&lt;h2&gt;
  
  
  When speed feels like certainty: the instant answer trap
&lt;/h2&gt;

&lt;p&gt;If an AI answers instantly with no source, treat it as a guess, not an answer. Speed is not a sign of accuracy; it’s a sign that the model is pulling from patterns it has seen before. Real verification takes time — opening a website, reading a bill, speaking to a person. When the AI skips that step, you should too.&lt;/p&gt;

&lt;h2&gt;
  
  
  Your verification checklist: bill, account, real person
&lt;/h2&gt;

&lt;p&gt;My rule: for anything important (money, health, work), I check the original source myself — the bill, the account, the real person. That means:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Money&lt;/strong&gt;: log in to the bank or credit‑card site, look at the transaction, or call the number on the back of the card.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Health&lt;/strong&gt;: check the prescription label, call the pharmacy, or ask the doctor’s office directly.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Work&lt;/strong&gt;: open the ticketing system, look at the email thread, or speak to the colleague who sent the request.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;If the AI’s answer doesn’t match what you see at the source, ignore the AI and follow the verified information.&lt;/p&gt;

&lt;h2&gt;
  
  
  The “how do you know that?” test in practice
&lt;/h2&gt;

&lt;p&gt;Next time an AI gives you a fact, ask out loud (or in your head) “how do you know that?” Listen for a source: a website, a document, a person’s name. If the answer is “I read it somewhere” or “it’s common knowledge,” treat it as unverified. If it can point to a specific, checkable source, go verify that source before you act. This habit takes 10 seconds and saves a lot of trouble later.&lt;/p&gt;

&lt;h2&gt;
  
  
  Keep yourself the checkpoint, not the police
&lt;/h2&gt;

&lt;p&gt;You don't need to audit every message your parent receives. Give them one simple rule instead: if something feels urgent or asks for money, pause and show it to you. You can then run the verification steps above together. If your parents have already discovered ChatGPT, it's worth reading &lt;a href="https://pragmaticsysadmin.help/senior-tech/2026-07-27-is-chatgpt-reading-your-parents-data/" rel="noopener noreferrer"&gt;how their prompts can become a privacy problem&lt;/a&gt; — pasting bank statements into a chatbot is exactly the kind of thing the pause-and-show rule prevents. Over time they'll learn to spot the signs — urgent language, unexpected attachments, too-perfect photos — and you'll both be less likely to act on a false alarm.&lt;/p&gt;

&lt;blockquote&gt;
&lt;p&gt;Treat AI like a helpful intern: fast, eager, but always needing your sign-off before anything goes live.&lt;/p&gt;
&lt;/blockquote&gt;




&lt;p&gt;&lt;em&gt;What's the one thing you make them show you before they click? Tell me and I'll add it to the checklist.&lt;/em&gt;&lt;/p&gt;

&lt;p&gt;&lt;em&gt;Related reads:&lt;/em&gt;&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;&lt;em&gt;&lt;a href="https://pragmaticsysadmin.help/senior-tech/2026-07-27-is-chatgpt-reading-your-parents-data/" rel="noopener noreferrer"&gt;Is ChatGPT Reading Your Parents' Data?&lt;/a&gt;&lt;/em&gt;&lt;/li&gt;
&lt;li&gt;&lt;em&gt;&lt;a href="https://pragmaticsysadmin.help/senior-tech/2026-07-27-password-manager-for-elderly-parents/" rel="noopener noreferrer"&gt;Password Manager for Elderly Parents: 10-Minute Setup&lt;/a&gt;&lt;/em&gt;&lt;/li&gt;
&lt;li&gt;&lt;em&gt;&lt;a href="https://pragmaticsysadmin.help/senior-tech/2026-06-27-5-conversations-aging-parent-online-safety/" rel="noopener noreferrer"&gt;5 Conversations to Have With Your Aging Parent About Online Safety&lt;/a&gt;&lt;/em&gt;&lt;/li&gt;
&lt;/ul&gt;

</description>
      <category>ai</category>
      <category>criticalthinking</category>
      <category>seniortech</category>
      <category>familyit</category>
    </item>
    <item>
      <title>When Code Meets Life: Applying the Sysadmin Mindset to Family Chaos &amp; Home Repairs</title>
      <dc:creator>Jonne Riepponen</dc:creator>
      <pubDate>Tue, 10 Nov 2026 16:30:59 +0000</pubDate>
      <link>https://dev.to/jronegit/when-code-meets-life-applying-the-sysadmin-mindset-to-family-chaos-home-repairs-2fhi</link>
      <guid>https://dev.to/jronegit/when-code-meets-life-applying-the-sysadmin-mindset-to-family-chaos-home-repairs-2fhi</guid>
      <description>&lt;p&gt;Last Tuesday the kitchen tap started dripping.&lt;/p&gt;

&lt;p&gt;Not a lot. Just one drip every four seconds or so. Annoying, not alarming. The kind of thing you notice, mentally file under "weekend," and then forget about for three weeks until it's dripping onto the floor and you've got a puddle forming near the dishwasher.&lt;/p&gt;

&lt;p&gt;I did what any sysadmin does. I didn't fix it. I &lt;strong&gt;investigated&lt;/strong&gt; it.&lt;/p&gt;

&lt;h2&gt;
  
  
  Step 1: Is This Actually a Tap Problem?
&lt;/h2&gt;

&lt;p&gt;Twenty minutes of observation established the following:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;The drip continued with the tap fully closed ✅ (real leak, not a dripping faucet)&lt;/li&gt;
&lt;li&gt;The drip rate was &lt;strong&gt;exactly one drip every 4.1 seconds&lt;/strong&gt; — metronomic, suspiciously regular&lt;/li&gt;
&lt;li&gt;The drip was coming from the spout outlet, not the handle or the joint&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;I turned off the stopcock under the sink. Drip stopped. Turned it back on. Drip resumed.&lt;/p&gt;

&lt;p&gt;Root cause confirmed: &lt;strong&gt;the ceramic cartridge was worn&lt;/strong&gt;. Not the washer, not the supply line, not a loose fitting. A €6 part.&lt;/p&gt;

&lt;p&gt;Ninety seconds of work. Five minutes of diagnosis. €6 and a YouTube video.&lt;/p&gt;

&lt;p&gt;I thought about how I'd actually spent that same Tuesday at work: 40 minutes of a service outage, and the eventual conclusion was "the connection pool was exhausted because someone left a connection open." Same thing. Look at the actual symptom, don't guess at the cause, check the cheapest explanation first.&lt;/p&gt;

&lt;h2&gt;
  
  
  A Family Is a Distributed System
&lt;/h2&gt;

&lt;p&gt;Here's the thing I realized somewhere between the second drip and the cartridge replacement:&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;The troubleshooting methodology I use on servers is identical to the one that works in my house.&lt;/strong&gt; Not similar. Identical.&lt;/p&gt;

&lt;p&gt;And once I saw that, I couldn't stop seeing it everywhere.&lt;/p&gt;

&lt;div class="table-wrapper-paragraph"&gt;&lt;table&gt;
&lt;thead&gt;
&lt;tr&gt;
&lt;th&gt;Sysadmin&lt;/th&gt;
&lt;th&gt;Home / Family&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td&gt;Read the logs before you touch anything&lt;/td&gt;
&lt;td&gt;Notice &lt;em&gt;when&lt;/em&gt; it started, not just &lt;em&gt;that&lt;/em&gt; it started&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Isolate one variable at a time&lt;/td&gt;
&lt;td&gt;Change one thing, wait a week, see if it helped&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;The obvious cause is usually right&lt;/td&gt;
&lt;td&gt;Is it plugged in? Is there a battery?&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Reproduce the bug before you fix it&lt;/td&gt;
&lt;td&gt;Can you make it happen on purpose?&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Working backwards from the symptom&lt;/td&gt;
&lt;td&gt;Which end of the pipe is wet?&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Root cause vs. surface symptom&lt;/td&gt;
&lt;td&gt;The drip is the symptom — the cartridge is the cause&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Escalation path&lt;/td&gt;
&lt;td&gt;When do I call a plumber vs. a YouTube video?&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Document everything&lt;/td&gt;
&lt;td&gt;The house manual, receipts, warranty dates&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Capacity planning&lt;/td&gt;
&lt;td&gt;When does this thing need replacing?&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Patch management&lt;/td&gt;
&lt;td&gt;Annual boiler service, smoke alarm batteries&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Single point of failure&lt;/td&gt;
&lt;td&gt;The one person who knows where everything is&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Redundancy&lt;/td&gt;
&lt;td&gt;Two spare light bulbs, not one&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Technical debt&lt;/td&gt;
&lt;td&gt;The shelf held up with card and tape&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;On-call rotation&lt;/td&gt;
&lt;td&gt;Tag team parenting&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Incident response&lt;/td&gt;
&lt;td&gt;It's 2 AM and the toilet is flooding&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Change management&lt;/td&gt;
&lt;td&gt;Nap schedule, new school, moving house&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Blast radius&lt;/td&gt;
&lt;td&gt;One person's bad mood affects the entire household&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Cache&lt;/td&gt;
&lt;td&gt;The freezer meal stash, the spare key&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Legacy system&lt;/td&gt;
&lt;td&gt;That appliance that works — don't touch it&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Monitoring &amp;amp; alerting&lt;/td&gt;
&lt;td&gt;The leak sensor, the doorbell camera, the fall detector&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Backup&lt;/td&gt;
&lt;td&gt;Photos of receipts, serial numbers, warranty PDFs&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Rollback&lt;/td&gt;
&lt;td&gt;Knowing how to undo what you changed&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;

&lt;p&gt;I keep going back to that last one. &lt;strong&gt;Rollback.&lt;/strong&gt; When I put the smart plug on the boiler's smart control, I wrote down how to put it back to manual. When we changed the DNS settings, I screenshotted the old ones. When the kid's bedtime moved from 7 to 7:30, I didn't just change it — I kept the old routine for a week in case reverting was needed.&lt;/p&gt;

&lt;p&gt;Nobody does this for a family. But it works exactly the same.&lt;/p&gt;




&lt;h2&gt;
  
  
  The House Log
&lt;/h2&gt;

&lt;p&gt;The single most useful thing I've done for my household is keep a &lt;strong&gt;house log&lt;/strong&gt;. It's a note on my phone. That's it.&lt;/p&gt;

&lt;p&gt;When something breaks, I don't start from scratch. I check the log first:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight plaintext"&gt;&lt;code&gt;2026-10-28  Kitchen tap dripping every 4 sec. Closed stopcock → stopped.
            Root cause: ceramic cartridge (€6, took 90 min).
            Fixed. Water pressure in the kitchen is lower than the bathroom.

2026-09-14  Boiler making a clicking noise when it heats. Not a fault.
            Was the same noise last winter. Ignore.

2026-08-02  Fridge freezer stopped freezing. Defrosted, thermostat was
            set to -15 instead of -18. Not the compressor.

2026-06-19  Hallway light flickers. Bulb was fine. The switch was loose.
            Tightened the terminal. Fixed.
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;Why this works: it stops you re-diagnosing solved problems. The boiler clicking noise? I already know. It's the same as last winter. Ignore it. That alone has saved me hours and probably a pointless £150 engineer call.&lt;/p&gt;

&lt;p&gt;It also captures &lt;strong&gt;the pattern&lt;/strong&gt;. Six months after the tap, when the bath tap starts dripping too, I don't start from zero — I go straight to "cartridge."&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;The house log is the single highest-ROI thing on this list.&lt;/strong&gt; It takes 30 seconds to write and it's the difference between "I don't know what's wrong" and "I think I know what's wrong."&lt;/p&gt;

&lt;p&gt;I keep mine in a phone note because I always have my phone. If you prefer, a physical notebook by the boiler works too — old school, zero apps, nobody can break it.&lt;/p&gt;




&lt;h2&gt;
  
  
  Escalation: When to Stop
&lt;/h2&gt;

&lt;p&gt;The most important sysadmin skill — and the one I was worst at learning — is knowing when to &lt;strong&gt;stop&lt;/strong&gt; and call for help.&lt;/p&gt;

&lt;p&gt;My personal escalation criteria, in order:&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;1. It involves electricity and water in the same job.&lt;/strong&gt; Stop. Call someone. No exceptions, ever.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;2. The instructions say "if you are unsure, contact a qualified professional."&lt;/strong&gt; Stop. That's the manufacturer telling you the job is beyond a YouTube video.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;3. You need to cut into the wall, drill through tiles, or move a gas line.&lt;/strong&gt; Stop. These aren't DIY tasks, no matter how many videos you watch.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;4. The fix involves a part you'd have to order anyway.&lt;/strong&gt; If the part costs €60 and you need an electrician to install it, use an electrician. Don't buy the part and spend two hours learning a skill you'll use once.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;5. The symptom is "it worked yesterday and now it doesn't."&lt;/strong&gt; That's the hardest category. Sometimes it's genuinely simple. Sometimes it's a failing component that would have failed anyway. If you've spent 30 minutes and you're guessing, escalate.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Everything else is fair game.&lt;/strong&gt; Dripping taps, running toilets, blocked sinks, squeaky hinges, slow drains, flickering lights, doors that don't latch. These are the sysadmin equivalent of a restart: low risk, clear scope, you can undo it if it goes wrong.&lt;/p&gt;

&lt;p&gt;The cost asymmetry helps: a bad DIY job on a tap costs €6 in parts. A bad DIY job on an electrical circuit costs a house.&lt;/p&gt;




&lt;h2&gt;
  
  
  Parenting as Incident Response
&lt;/h2&gt;

&lt;p&gt;I wasn't expecting the parenting parallels to be this obvious, but they are.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Debugging a tantrum&lt;/strong&gt; is variable isolation. Four-year-olds have a small number of recurring triggers, and if you write them down you'll find them fast:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;Tired&lt;/li&gt;
&lt;li&gt;Hungry&lt;/li&gt;
&lt;li&gt;Overstimulated&lt;/li&gt;
&lt;li&gt;Transition given too late&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;My daughter has a tantrum roughly 20 minutes before any transition — leaving the park, ending a TV show, going to bed. Same input, same output, every time. Once I spotted the 20-minute pattern, I could pre-empt it: "We have 15 more minutes, then we go home." No tantrum. Because I stopped treating each one as a new incident and started treating it as a &lt;strong&gt;known reproducer&lt;/strong&gt;.&lt;/p&gt;

&lt;p&gt;That's just bug reproduction. You can't fix what you can't reproduce.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;The nap log is a metrics dashboard.&lt;/strong&gt; Sleep times, wake times, total sleep, mood rating. After two weeks of actual data I stopped guessing about whether she needed a longer nap and started just... knowing. I'd lost weeks to "should she nap?" debates that a two-column note would have settled.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Load balancing is capacity planning.&lt;/strong&gt; Two parents, one demanding toddler, one work deadline. The failure mode isn't that one person works too hard — it's that both parents independently take on half of &lt;em&gt;everything&lt;/em&gt;, which is more than half. You need explicit division: one person owns the morning, the other owns bedtime. Not both halves of both.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Blast radius matters.&lt;/strong&gt; One parent's bad mood in the morning sets the tone for the whole day, for everyone. I know this from production systems — one unhealthy dependency poisons everything downstream. A ten-minute reset before the day starts ("I'm having a rough morning, here's my plan") does more than you'd think.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Change management for a toddler is brutal.&lt;/strong&gt; We moved her bedtime by 30 minutes and it took eleven days. Then we moved it back and it took four. Never have I seen a rollback take so long or fail so spectacularly.&lt;/p&gt;




&lt;h2&gt;
  
  
  The One I Didn't Expect: Technical Debt in the Home
&lt;/h2&gt;

&lt;p&gt;Every house I've been in has a shelf somewhere that is technically held together by a piece of card and masking tape. Every house has a door that only closes if you lift it slightly. Every house has one electrical socket that's slightly loose.&lt;/p&gt;

&lt;p&gt;These are &lt;strong&gt;technical debt&lt;/strong&gt;. They work right now, so nobody fixes them. Then one day the shelf falls, or the socket fails, and the fix takes twice as long as it should because it's built on a hack nobody documented.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;The house log is how you pay down technical debt.&lt;/strong&gt; When you finally fix the shelf properly, write it down: what it was, how you fixed it, that the tape is gone. Future you doesn't have to reverse-engineer a 2019 hack.&lt;/p&gt;

&lt;p&gt;I also started a simple rule: &lt;strong&gt;one fix a quarter.&lt;/strong&gt; Not a renovation. One thing that was held together with tape gets fixed properly. Rotating a small list of debt, three or four times a year, and the house gets steadily less held together with tape.&lt;/p&gt;




&lt;h2&gt;
  
  
  On-call Is a Role, Not a Personality Trait
&lt;/h2&gt;

&lt;p&gt;One thing I got wrong for years: I thought being "the tech person in the family" was a personality trait. It's a &lt;strong&gt;role&lt;/strong&gt;, and it needs boundaries.&lt;/p&gt;

&lt;p&gt;Being on call for the family tech means:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;Being reachable for real emergencies (no Wi-Fi, no heating in February)&lt;/li&gt;
&lt;li&gt;Being &lt;strong&gt;explicitly not&lt;/strong&gt; on call for "the printer is acting weird" at 9 PM on a Sunday&lt;/li&gt;
&lt;li&gt;Having a backup (a sibling, a partner, a family friend who's comfortable with technology)&lt;/li&gt;
&lt;li&gt;Rotating it — the person who was on call this month isn't automatically next month&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;The difference between "on call" and "always on" is the difference between a resilient system and a burnout. I learned this one the expensive way.&lt;/p&gt;

&lt;p&gt;The whole point of automating things — the smart plug, the leak sensor, the &lt;a href="https://pragmaticsysadmin.help/buddy/" rel="noopener noreferrer"&gt;Buddy app&lt;/a&gt; on my dad's phone — is to reduce the number of pages that reach a human. Every alert that fires at 3 AM erodes the willingness to respond to the alert that actually matters.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Automation is a force multiplier for attention, not a replacement for it.&lt;/strong&gt; The smart plug that turns off the iron means one fewer thing to think about, which means more attention for the stuff that genuinely needs me.&lt;/p&gt;




&lt;h2&gt;
  
  
  The Actual Point
&lt;/h2&gt;

&lt;p&gt;None of this is a clever analogy. I know how it sounds — "everything is distributed systems" is a cliché, and I'm not claiming the metaphor is illuminating.&lt;/p&gt;

&lt;p&gt;But the &lt;em&gt;method&lt;/em&gt; transfers. And the method is:&lt;/p&gt;

&lt;ol&gt;
&lt;li&gt;&lt;strong&gt;Look at the actual symptom before theorizing about the cause&lt;/strong&gt;&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;Write down what you learn&lt;/strong&gt;&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;Change one thing at a time&lt;/strong&gt;&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;Know the cheapest explanation, and check it first&lt;/strong&gt;&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;Know when you're out of your depth, and escalate without shame&lt;/strong&gt;&lt;/li&gt;
&lt;/ol&gt;

&lt;p&gt;That's it. That's the whole method. It works on a leaking tap, a frustrated four-year-old, and a production database at 2 AM, and the reason it works is the same in all three cases: the person who wrote the code doesn't know what the bug is until they look.&lt;/p&gt;




&lt;p&gt;&lt;em&gt;What's the house log entry you wrote down this year that you still remember? I'd genuinely like to know — send me yours at &lt;a href="mailto:pragmatic@pragmaticsysadmin.help"&gt;pragmatic@pragmaticsysadmin.help&lt;/a&gt;.&lt;/em&gt;&lt;/p&gt;

</description>
      <category>sysadmin</category>
      <category>family</category>
      <category>homerepairs</category>
      <category>philosophy</category>
    </item>
  </channel>
</rss>
