Zero GPU, zero dollars: a Linux rookie's crew of free AIs

A self-described non-programmer known as Rabbid Raccoon built a crew of free, locally run AI agents on a 15 GB RAM laptop with no GPU, using Ollama models, n8n scheduled jobs, Vosk for offline speech recognition and Kokoro for voice output. The agents monitor backups, a home lab and gadgets and deliver daily in-character reports through a Rick and Morty themed dashboard, with all code published under an MIT license. The builder's stated lesson is to keep the AI honest by checking a list of real actions before the model gets a say.

A few weeks ago I couldn't have told you what sudo does. Today my laptop runs a whole crew of AI agents. They babysit my backups, my home lab and my gadgets, and report back every day, in character. One of them is Rick Sanchez. One of them is a British butler who swears like George Carlin. None of them cost a dime, and the laptop doesn't even have a graphics card. I'm not a programmer. I built all of this by describing what I wanted to Claude https://claude.com/claude-code , testing it, complaining, and testing it again. This post is what I learned, for anyone else who's starting from zero. Everything I build follows four rules: Rule 1 sounds limiting. It turned out to be the most creative constraint I've ever had. My main machine is a regular laptop with 15 GB of RAM and no GPU . Everyone online says you need a big graphics card for AI. You don't, as long as you pick the right tools: I keep the full list of models I actually use in a Hugging Face collection https://huggingface.co/collections/RabbidRaccoon/free-ais-on-a-laptop-with-no-gpu . The fun part. Each crew member is a scheduled job in n8n https://n8n.io that gathers facts, asks an AI to write a short report in character, and drops it into my daily notes sometimes out loud : They all live on Mission Control , a Rick and Morty themed dashboard that also launches my apps and searches an offline library Wikipedia, medical encyclopedias, repair guides when the internet's out. My favorite thing I've built. Joshua listens all the time but stays silent until he hears his name. Then he answers out loud in a British accent with the manners of Jeeves and the mouth of George Carlin. Hearing is fully offline Vosk https://alphacephei.com/vosk/ , the voice is Kokoro https://huggingface.co/hexgrad/Kokoro-82M , and the brain is a free model. My lesson from Joshua: keep the AI honest. Early versions "opened apps" that they never actually opened. Now there's a list of real actions checked before the AI gets a say, so he can't pretend. Everything is public, free and MIT licensed: ollama run 2db2du/rick or ollama run 2db2du/joshua Before anything goes public, a script builds cleaned copies of my projects and scans them for personal info names, drive serials, home paths, keys . If it finds anything, it refuses to publish. Beginners: build that safety net first . Next up: actually finishing Unit 1 of my Linux course. Morty keeps reminding me. Zero GPU. Zero dollars. Zero excuses. - Rabbid Raccoon 🦝