DEV Community: Rabbid Raccoon

DON'T PANIC, Part 2: My iPad App Now Runs the PC (a Little)

Rabbid Raccoon — Mon, 26 Oct 2026 20:20:34 +0000

Last week I built The Hitchhiker's Guide: an iPad home-screen app with a big friendly DON'T PANIC on it. Tap it, pick a character, chat. Every answer comes from AIs on my own PC: Claude, free cloud AIs, and local models on my one graphics card. The iPad is just the window.

Today it grew up a little.

Status strip: CPU and GPU temps, memory, when the last backup finished, and whether the internet is up. It goes red only when something is actually wrong. No red, no worries.
PC buttons: exactly four. Back up now. Put the local AIs to sleep (frees the graphics card). Lock the PC. Say something out loud on the PC speakers. That's the whole list. The iPad can't run anything else, on purpose.
Three Claude helpers, talk-only: a planner, a Linux tutor, and a notes keeper. From the iPad they can talk and read. They can't run commands, edit files or browse.
Keeps the PC awake while I'm using it, so the PC doesn't nod off mid-conversation. The screen still locks.
Security pass: I had it reviewed. The big fix: chat text now goes to Claude through the front door (stdin), never on the command line, so a sneaky message can't pretend to be a setting. Plus size checks, stricter browser headers, and one spoken message at a time.

Still private by design: it only answers my own PC and my own Tailscale devices, and only with a key.

One GPU. Zero dollars. Zero excuses.

Code: https://github.com/2db2du-byte/hitchhikers-guide

Everything else I build: https://2db2du-byte.github.io

My PC's Lights Now Dance to the Bass. Cost: $0.

Rabbid Raccoon — Mon, 26 Oct 2026 20:20:22 +0000

Gaming PCs ship with RGB lights that do one thing: rainbow. Forever. Like a screensaver that got a job.

So I made mine pay attention.

rgb-mood has two moods, and it picks on its own:

Music playing: the lights pulse with the bass of whatever comes out of the speakers. Spotify, YouTube, a game. The color drifts slowly around the rainbow while the beat does the punching.
Quiet for a few seconds: the lights glow by the hottest chip in the box. Blue when it's cool, purple when it's working, red when it's cooking. It breathes faster as it heats up.
RAM: stays solid blue, or flip one setting and it turns into a VU meter.

How it works, Tim Urban style: the PC listens to its own speakers (PipeWire), a bit of math (an FFT, via numpy) pulls out the bass 30 times a second, and OpenRGB tells the lights what to do. When it's quiet, it reads the CPU and GPU temperatures instead.

It only runs when I say rgb-mood on. Nothing starts at boot.

Honest status: tested on my RAM and motherboard lights. It hears the music and drives OpenRGB. My case fans aren't wired to the motherboard's light header yet, so the full-case demo is coming.

Free tools only: OpenRGB, PipeWire, Python, numpy. One GPU. Zero dollars. Zero excuses.

Code: https://github.com/2db2du-byte/rgb-mood

Everything else I build lives here: https://2db2du-byte.github.io

DON'T PANIC: Build an iPad AI Chat Hub with Your Own PC

Rabbid Raccoon — Sat, 24 Oct 2026 00:18:46 +0000

DON'T PANIC – An iPad AI Companion

Rabbid just pushed hitchhikers‑guide, an open‑source iPad home‑screen app that lets you chat with a roster of AI personalities—all powered by the PC you already own. No cloud fees, no data leaks; the iPad is just a pretty window.

Character	Brain
Deep Thought	Claude (`claude -p`) – one persistent convo
Eddie	Free cloud AI – casual chat
Trillian	Free cloud AI – deeper thoughts
Slartibartfast	Free cloud AI – Linux, code, how‑to
Marvin	Local Ollama model on your GPU – offline
Ford Prefect	Small local Ollama model – quick offline

The cloud AIs route through a LiteLLM "brain switch" (see the crew repo). Local models call Ollama directly, launching on demand. All chats are saved on the PC, syncing across your iPad and phone.

Private by design – a single stdlib Python file serves on port 4200, only answering your Tailscale network (100.64.0.0/10) and requiring a random key. Open http://<your‑tailscale‑ip>:4200/?key=<key> once, then add to Home Screen. Optimized for an 11" iPad (1180×820) but works on phones too.

MIT licensed, built by Rabbid Raccoon with Claude. Grab it, tweak it, share it.

Find more of my tinkering at my homepage: https://2db2du-byte.github.io

Bridge: Turn Your PC Into a Star Trek Command Deck

Rabbid Raccoon — Sat, 24 Oct 2026 00:18:32 +0000

Bridge – A Star Trek LCARS Control Center for Your Home PC

If you’ve ever wanted to feel like the captain of the Enterprise while you code, Rabbid’s new Bridge repo delivers exactly that. It’s a single‑page, full‑screen UI that mimics the LCARS panels of Picard’s era (with a button to flip back to Kirk’s classic look). The centerpiece is a WebGL‑rendered star field that doubles as a viewscreen – you can even project your webcam onto it, and watch YouTube on the right half via a personal SearXNG search.

What’s on Deck?

Engineering: Real‑time CPU/GPU/RAM usage, temps, and a pulse‑like warp core visual.
Ops: Kopia backup status with a one‑click "back up now" button.
Science: Offline library, maps, and course progress tracking.
Tactical: Firewall status displayed as shield strength.
Helm: Launch whitelisted games/apps directly from the UI.
Comms: Quick links to your public accounts.

Red alerts only fire when something truly needs attention – a stale backup, a full drive, or an overheating GPU. The "Deep Space Nine" side panel opens Claude Code agents in separate tmux sessions, keeping conversations alive.

Tech Stack

Tiny Python stdlib server on 127.0.0.1:1701
WebAudio‑synthesized sounds, local TTS (say) for the computer voice
Claude Code terminal via ttyd on a private socket

The project is MIT‑licensed, built by Rabbid Raccoon with Claude. Grab it, tweak it, and make your desk the bridge of a starship.

Find more of my projects at https://2db2du-byte.github.io

I turned a dead 4 GB laptop into a Matrix cyber-lab

Rabbid Raccoon — Fri, 09 Oct 2026 02:13:32 +0000

Update, Oct 2026: I have since moved up to my own Ubuntu rig with an RTX 5060. Everything in this post still works with no graphics card at all. New motto: One GPU. Zero dollars. Zero excuses. 🦝

I had an old laptop gathering dust. An 11-inch thing with 4 GB of RAM, a weak little CPU, and no graphics card. The kind of machine the internet tells you is e-waste.

I made it my cyber-lab instead. Here's how it looks and what it actually does, because "old and slow" turned out to mean "perfect for learning without fear of breaking anything."

One GPU. Zero dollars. Zero excuses.

Make it look the part

If you're going to learn Linux and cybersecurity, the machine might as well feel like it. So I gave it the full green-on-black treatment, all generated with Python, all free:

A raccoon cursor. A normal arrow with a tiny raccoon riding along behind it. It's a real X cursor theme, so the whole desktop uses it.
A live Matrix wallpaper. Falling green code that actually moves. The trick is making it loop seamlessly so it plays forever with no jump, and keeping it light enough that a 4 GB laptop doesn't choke. One glowing line in the middle: "Wake up, Rabbid..."
A Matrix theme for the whole desktop, so the bar, the terminals, and the menus all match.

The cursor and wallpaper scripts are here if you want them: github.com/2db2du-byte/matrix-lab. Change the colors and the message and they're yours.

The honest part: a 4 GB laptop can't think

Here's where people get it wrong. You cannot run a big local AI model on 4 GB of RAM. It won't even load. I tried; physics said no.

But here's the thing nobody tells beginners: the free cloud AIs don't run on your memory. They run on someone else's servers. So I gave the laptop a simple chat screen — looks like a chat app, pure green-on-black — that talks to the free AI services I already use. The laptop stays light; the thinking happens elsewhere; and it costs nothing because the free tiers are plenty for one person asking questions.

Ask it anything. "What does chmod do?" "Explain this error." "Why is the sky blue?" It answers, and the old laptop barely breaks a sweat, because it's just passing the question along.

Why bother?

Because a dedicated lab machine means I can mess around, break things, and practice, without touching the laptop that runs my real life. It's the whole point of a lab: a place where mistakes are free.

And because I'm a few weeks into Linux. I'm not an expert. I built this by describing what I wanted, testing it, swearing at it, and testing it again. If a rookie can turn a dead laptop into something this fun, so can you.

Old hardware isn't dead. It's just waiting for someone to give it a weirder job.

🦝 Rabbid Raccoon — homepage · GitHub · Bluesky

Two more free AI brains for my no-GPU laptop: NVIDIA NIM and Ollama Cloud

Rabbid Raccoon — Thu, 08 Oct 2026 14:55:49 +0000

Update, Oct 2026: I have since moved up to my own Ubuntu rig with an RTX 5060. Everything in this post still works with no graphics card at all. New motto: One GPU. Zero dollars. Zero excuses. 🦝

Quick follow-up to Zero GPU, zero dollars.

My AI crew talks to one address on my laptop: a LiteLLM "brain switch". Every job (chat, think hard, code, look at pictures, read long documents) has a list of free AI services to try in order. If one is out of free uses, the next one answers. If the internet dies, the laptop's own Ollama models take over.

Until today that list had 5 services: Groq, Gemini, Mistral, OpenRouter and Cloudflare Workers AI. Now it has 7.

NVIDIA NIM

build.nvidia.com gives you a free API key for a big menu of open models. The limit is about 40 requests a minute for the whole account, which is plenty for one person.

The star for me is Kimi K3. It now leads my "think hard" job:

answers in about 3 seconds
handles tool calls properly (a lot of free models fumble these)
saves Gemini Pro's small free daily allowance for when I really need it

GLM-5.3 from the same key backs up chat, thinking and coding.

Ollama Cloud

If you already use Ollama locally, the free cloud plan is the same app with bigger models running on their servers. The catch: the free plan only offers a few models (gpt-oss, gemma4:31b, nemotron-3-super), runs one request at a time, and has 5-hour and weekly limits they don't publish.

So it's a backup, not a main brain. It sits at the bottom of the chat list, and Gemma 4 31B is the second choice for looking at pictures.

The order now

Job	Tries first	Then
Think hard	Kimi K3 (NVIDIA)	Gemini Pro, GLM-5.3, Nemotron, gpt-oss
Chat	gpt-oss on Groq	Gemini Flash, Llama 3.3 on Cloudflare, Gemma 4, Mistral, GLM-5.3, Ollama Cloud
Code	Codestral (Mistral)	GLM-5.3, Kimi K3, Groq, Gemini, OpenRouter, Cloudflare
Pictures	Gemini Flash	Gemma 4 on Ollama Cloud, OpenRouter
Long docs	Gemini Flash	Kimi K3, Nemotron

Still no graphics card, still no credit card. The whole config is in my crew repo (bring your own keys, none are included).

One GPU. Zero dollars. Zero excuses. 🦝

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

Rabbid Raccoon — Wed, 07 Oct 2026 20:09:12 +0000

Update, Oct 2026: I have since moved up to my own Ubuntu rig with an RTX 5060. Everything in this post still works with no graphics card at all. New motto: One GPU. Zero dollars. Zero excuses. 🦝

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, testing it, complaining, and testing it again. This post is what I learned, for anyone else who's starting from zero.

The house rules

Everything I build follows four rules:

If it ain't free, skip it. Free and open source first, every time.
Local beats cloud. Private, self-hosted, works when the internet doesn't.
Prototype first. Test wild ideas on cheap, spare hardware before building them for real.
Nothing starts at boot unless I say so. My laptop, my rules.

Rule 1 sounds limiting. It turned out to be the most creative constraint I've ever had.

The setup: no graphics card

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:

Omarchy (Arch Linux + Hyprland) as the operating system.
Ollama to run small AI models on the CPU: Qwen3, Llama 3.2, Gemma 3, Mistral. The 3-8 billion parameter ones are the sweet spot - slow-ish, but they work.
A "brain switch" (LiteLLM) that sends requests to the free tiers of Groq, Gemini, Mistral, OpenRouter and Cloudflare when I'm online, and falls back to the local models when I'm not. One address, many brains. It even worked out which free service is fast for which job: my coding agent went from 1-3 minutes per step to about 4 seconds once I put Mistral's Codestral first.

I keep the full list of models I actually use in a Hugging Face collection.

The crew

The fun part. Each crew member is a scheduled job in n8n that gathers facts, asks an AI to write a short report in character, and drops it into my daily notes (sometimes out loud):

Rick gives the morning briefing (PG - kids use this computer).
Birdperson guards the backups.
Beth checks the home lab apps.
Morty reminds me about my Linux lessons, which is fitting.
Mr. Poopybutthole keeps my GitHub and Hugging Face looking good. Ooo-wee.

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.

Joshua, the voice butler

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), the voice is Kokoro, 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.

Sharing it

Everything is public, free and MIT licensed:

🧪 Try it right now: Ask Rick - a PG Rick whose AI brain runs inside your browser. No server, no account.
🦙 Run them yourself: ollama run 2db2du/rick or ollama run 2db2du/joshua (on Ollama) - no GPU needed.
🐙 The code: github.com/2db2du-byte
🤗 The whole setup + every model I run: Rabbid's Lab and the collection
🌐 Homepage: 2db2du-byte.github.io
🦋 Say hi: @rabbid-raccoon.bsky.social

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.

What I'd tell another beginner

You don't need a GPU to start. Small models on a CPU are slow but real.
Describe what you want in plain words. AI coding tools are good at turning "I want a butler that only talks when I say his name" into working code.
Test everything. "It should work" and "it works" are different sentences.
Make it fun. I learn more from a burping Rick than from any tutorial.

Next up: actually finishing Unit 1 of my Linux course. Morty keeps reminding me.

One GPU. Zero dollars. Zero excuses.

- Rabbid Raccoon 🦝