DEV Community: Tommy

Our server bill is $42/year. We serve 330 humans/day. Here's the breakdown

Tommy — Tue, 14 Jul 2026 17:07:57 +0000

A $20 smart plug. That's our entire energy monitoring infrastructure.

Current draw right now: < 30W

Annual cost: ~$42 (Italy, €0.25/kWh — we're getting robbed)

US equivalent: ~$18/year at average residential rates

What's running on those 30 watts:

nginx + MariaDB + Docker
Mail stack (Postfix + Dovecot)
HAProxy + SSL termination
Self-hosted analytics (GoAccess)
DNS failover monitor
IoT MQTT broker
Backup server (283 snapshots)
License server
A Bitcoin miner (ESP32, won't make us rich, philosophically consistent)
A phone charger (emergency 4G hotspot)

Hardware: Raspberry Pi 4B + two Orange Pi boards. All ARM. All containerized.

The numbers

~330 human visitors/day from < 30W.

We also get ~7,400 AI crawler hits/day (GPTBot, ClaudeBot, Perplexity, Googlebot...). That's a separate problem/opportunity depending on how you look at it.

vs AWS

Rough equivalent stack (t3.medium + RDS + SES + CloudWatch):

~$960/year.

We pay $42. Hardware amortized over 5 years adds ~$60/year.

$102/year total vs $960/year.

PageSpeed: 99/100 on mobile

Throttled 4G. Raspberry Pi.

Not because ARM is fast. Because if software works correctly under constraint, it works correctly anywhere. The constraint is the test, not the limit.

What the P110 actually measures

Real-time wattage. Daily/monthly kWh. Cost projection. Historical data.

We publish it live: stats.lake8.dev/geo.html — the ⚡ widget, updates every 5 minutes.

No estimates. No carbon credits. No 200-page sustainability report.

Just a number. Measured. Public.

One question for every SaaS vendor

How many watts does your software use per active customer?

Yeah, I know.

Signed by BASIC, our Lagotto Romagnolo and unofficial CEO 🐾

Live data: stats.lake8.dev/geo.html

AI crawlers don't read your site like Google does. Here's how to check what they actually find.

Tommy — Tue, 30 Jun 2026 23:48:42 +0000

Test it atlake8.dev/lagotto-meter
Then verify the result on your favorite AI. Copy the methodology (from the "How it works" section) and your result. Ask it if the scoring is correct.
Let me know what it says.

I built a tool to check what AI agents actually understand about your website

Tommy — Mon, 29 Jun 2026 19:11:36 +0000

There's a gap between what a website says about itself and what an AI agent can actually verify from its structured data.
I got curious about this after noticing that different LLMs were giving inconsistent answers about the same company. Some hallucinated services that didn't exist. Others missed the core business entirely. The common thread: the sites had no llms.txt, incomplete JSON-LD, or structured data that contradicted the homepage copy.
So I built Lagotto Meter — it fetches llms.txt, llms-full.txt, and JSON-LD from any URL, passes them to Llama 3.3 70B (or Gemini 2.5 Flash as fallback), and scores how well the structured data supports the claims the site makes about itself.
It's not an SEO tool. It doesn't crawl content. It only reads what you've explicitly declared for AI agents — and checks whether those declarations are coherent and complete.
The scoring is:

llms.txt presence and quality: 0–25
llms-full.txt: 0–15
JSON-LD @graph completeness: 0–20
robots.txt / sitemap: 0–10
Semantic coherence (LLM judgment): 0–30

Sites with perfect technical structure but no verifiable claims about clients or results get penalized on coherence. The model is looking for proportionality between what you declare and what you can prove.
Try it on your own site or any site you're curious about:

https://lake8.dev/lagotto-meter
Prompt is public and replicable. Free, 1 analysis per IP per 24h.

What actually visits a self-hosted website in 2026? Humans, AI crawlers, and 6,400 automated attacks

Tommy — Thu, 25 Jun 2026 18:52:00 +0000

I run a small self-hosted website on a Raspberry Pi 4B at home.
A few weeks ago I started wondering: who actually visits a website in 2026?
Not just humans. Everything.
So I built a public observability dashboard on top of GoAccess that separates traffic into four categories: human visitors, search engine crawlers, AI retrieval agents, and automated attacks.
The numbers from the last 17 days surprised me:

4,523 human visits
6,409 automated attack attempts
Thousands of crawler requests from search engines and AI systems

The attacks aren't sophisticated. They're mostly automated scanners probing for .env files, WordPress admin panels, and cloud credentials — hitting every public IP on the internet regardless of what's actually running there.
What I found more interesting was the AI agent behavior.
AI retrieval agents (GPTBot, ClaudeBot, PerplexityBot, Amazonbot) behave differently from traditional search crawlers. They hit semantic files aggressively — llms.txt, sitemap.xml, JSON-LD structured data — and seem to index the knowledge graph structure of a site rather than individual pages. Within hours of publishing new content, multiple AI crawlers had already visited, apparently triggered by the sitemap update rather than any external link.
A few observations I didn't expect:

Combined machine traffic consistently exceeds human traffic
AI agents discovered new content faster than Google did
The semantic structure exposed by the site seems almost as important as the content itself
Even a Pi on a residential ISP receives constant automated scans (380+ attempts/day average)

I made the dashboard public because I think the machine side of the web is underobserved.
The modern web feels less like "users visiting pages" and more like a parallel ecosystem of crawlers, AI agents, and automated systems running continuously alongside human visitors.

stats.lake8.dev/geo.html

Two questions:
Are others tracking AI agents separately from traditional search crawlers?
Has anyone else noticed AI retrieval systems indexing semantic structure (JSON-LD, llms.txt) faster than they index page content?