The Homelab Agent Kit — 9 AI Agents, 13-Chapter Build Guide, Python

A 13-chapter build guide and the working Python code for nine self-hosted AI agents, running on a Raspberry Pi 5 since June 2026. This is the actual system behind homelabbench.com — the control plane, the content pipeline, the commerce automation, the paper-trading lab, the feedback loops — documented chapter by chapter with the architecture decisions, the dead ends, and the things that broke. Instant download. 74 KB. Everything is plain text and Python; nothing phones home.

• 13 chapters: What This Is · Hardware & OS · Security · Agent Anatomy · The Brain (control plane with web dashboard + Telegram) · Hybrid LLM (local Ollama + remote fallback) · Content Agent (publishing pipeline) · Distribution (RSS triage, Bluesky syndication) · Commerce (product sourcing & listing automation) · Trading Lab (honest paper-trading, no real money) · Feedback Loops (Search Console data back into the system) · Remote Access · Lessons (what failed, what that cost in time, what the author would change)
• Working code for 9 agents: control-plane orchestrator, content writer, RSS triage, Bluesky syndicator, e-commerce sourcer, e-commerce lister, Search Console feedback agent, paper-trading agent, Telegram notifier
• Runs on a Raspberry Pi 5. No GPU required. Local inference via Ollama for the tasks that suit it; the guide is explicit about where local models fall short and when the system calls out to a remote API instead.

Who this is for: someone who already runs a homelab, is comfortable with SSH and Python, and wants a concrete, working example of a multi-agent system they can pull apart, extend, or cannibalise for their own projects. You should be the sort of person who reads a docker-compose.yml before you read the README.

Who this is not for: if you are looking for a passive income machine, this is not it. If you have never SSHed into a Linux box, you will not enjoy this. If you want polished SaaS-grade software, look elsewhere — this is homelab code, written to work reliably for one person on one Pi, and documented honestly rather than prettily.

An honest note on results: the author's own site and Shopify shop have not yet turned a profit. The trading agent is paper-trading only and has never risked real money. The system works — it publishes content, syndicates posts, sources products, files listings, ingests Search Console data, and runs a simulated trading book — but working is not the same as earning. You are buying the engineering and the documentation, not a outcome. What you get is a system you can verify, run, and modify, plus a frank account of what went wrong along the way.