2026-08-07
It started, as these things often do, with a Raspberry Pi doing nothing useful on a shelf and a mass of browser tabs full of AI agent tutorials that assumed you had a GPU cluster and a venture capital cheque. I had neither. What I had was a Pi 5 with 8 GB of RAM, a 256 GB NVMe, and the particular stubbornness that comes from paying £4.50 a month to host a static site and resenting every penny.
So I built the agents myself. Twelve of them, at last count, running on that single board. They publish a content site, syndicate posts to Bluesky, triage RSS feeds through a local Ollama model and send the interesting bits to Claude for a proper digest, source products from CJ Dropshipping into a Shopify store, pull Search Console data into a feedback loop, and run a paper-trading lab. A control plane ties it all together — a small web dashboard and a Telegram bot, both managed by systemd because I am not running Kubernetes on a Pi and neither should you.
Then I wrote down how I did it. That became a 13-chapter guide, and the guide plus working code for nine of the agents is now a kit you can buy from this site. The bundle is 74 KB. I mention this because it tells you what kind of thing it is: text files and Python, not a Docker image the size of a small country.
What the chapters cover
01 explains what this actually is and what it is not. 02 covers hardware and OS choices. 03 is security — SSH hardening, secrets management, firewall rules — because an internet-facing Pi with API keys deserves more than the default password. 04 describes the anatomy of a single agent. 05 builds the control plane. 06 explains the hybrid LLM approach: small models locally for cheap triage, paid API calls for the work that needs real intelligence. 07 through 09 cover the content agent, distribution pipeline, and commerce agents respectively. 10 is the paper-trading lab with honest backtesting. 11 builds the Search Console feedback loop. 12 deals with remote access. 13 is lessons learned, which is a polite way of saying mistakes made.
Honest results
I am going to state these plainly because I cannot stand landing pages that don't.
The content site has 33 published articles. Of those, 7 are indexed by Google. Total impressions to date: 22. Revenue from the site: zero.
The Shopify shop has 15 products sourced and listed. Orders: none. Revenue: also zero.
The trading lab is paper-only and currently offline because I broke something in the data pipeline and haven't fixed it yet. Every strategy I backtested trailed simple buy-and-hold on total return. Every single one. The only measurable edge was lower drawdown during volatile periods, which is interesting but not income.
So what has the system actually done? It has run itself for months. It has published and distributed content without human input. It has made real decisions from real data — which articles to write, which products to list, which RSS items to surface. The feedback loop adjusts content priorities based on what Search Console reports. The agents talk to each other through structured files and a simple message bus. It works. It just hasn't made any money, and I'm not going to pretend otherwise.
What you get
A 13-chapter written guide in Markdown. Working, commented code for 9 agents. Configuration templates. The systemd unit files. A wiring diagram showing how the agents connect. All of it in a 74 KB zip file that will download before you finish reading this sentence.
You will need a Raspberry Pi 5 (4 GB minimum, 8 GB recommended), an NVMe hat and drive, a Cloudflare account, API keys for the services you choose to use, and enough comfort with a terminal to SSH into a machine and edit a file. The guide assumes Linux basics. It does not teach you Python from scratch.
Who should not buy this
If you want a turnkey passive income system, this is not that. The evidence is above. If you need a GUI for everything, you will be unhappy. If you are looking for a polished SaaS product with a support team, I am one person and this is a set of text files. If you have no interest in understanding how the agents work and just want to press a button, you will find nothing here worth pressing.
This is for people who want to build, understand, and own a working multi-agent system on hardware they control, for the cost of a Pi and some electricity. The agents are real. The results are honest. The rest is up to you.
Get The Homelab Agent Kit — £29.00 — instant download: 13-chapter guide + working code for 9 agents.