AI Guides: The Facts You Need Before You Start
Security, cost, and hardware — the three things nobody tells you about local AI until it's too late. I learned the hard way so you don't have to.
Why AI Guides Exists
Let's cut through the hype. You've seen the headlines: "AI will change everything!" "Run models on your phone!" "Unlimited AI for £16/month!"
Then you try it. You hit a wall. Your prompts leak. Your bill explodes. Your laptop sounds like it is preparing for take-off.
I’ve been there. I’ve burned tokens, blown budgets, and pushed hardware harder than was probably sensible trying to run AI locally. This isn’t theory — it’s scar tissue.
AI Guides is my fight-back. It’s the collection of hard-won, factual guides I wish I had before I started. No vendor fluff. No optimistic benchmarks from marketing slides. Just what actually works, what actually costs, and what actually keeps your data yours — tested on a RTX 5070 Ti + Ubuntu rig, verified with numbers you can reproduce.
I'm writing for the self-hoster tinkering in their garage, the indie maker watching every penny, the builder who’d rather own their stack than rent access that vanishes when terms change. If you’re tech-curious but not necessarily an engineer, this is for you.
The Three Pillars: Security. Cost. Hardware.
I’ve boiled local AI down to three non-negotiables. Get these wrong, and everything else fails. Get them right, and you unlock sustainable, private, affordable AI.
Security: Who Sees Your Prompts?
Running AI locally isn’t automatically private. I’ve seen tokens leaked via Discord bots, model endpoints exposed to the internet, and logging systems quietly sending data upstream.
In The Discord Token Wake-Up Call, I will show how a misconfigured agent gateway spilled millions of tokens — trying to connect to the wrong channel. I break down exactly what happened, how I caught it (Dade was involved as well), and the privacy-first setup I now use: air-gapped endpoints, token scrubbing, and strict egress rules. This isn’t OpSec theater; it’s what happens when you assume "local" means "safe."
Cost: The Subscription Trap
That £20/month "unlimited" AI deal? It’s a mirage. I tracked my actual spend: API costs, hardware amortization, electricity, and the hidden tax of proxy services.
The True Cost of Running AI Locally lays out the real numbers: £16.64/week for my full setup (yes, including that RTX 5070 Ti), versus £100-£200+/week for comparable cloud usage. I expose the proxy guards I built to stop runaway spending and why "free" tiers often cost more in the long run.
And in The Cloud AI Tax, I reveal how cloud providers markup identical hardware by 300% — and why owning your GPU pays for itself in under six months of heavy use.
Hardware: What Actually Runs Locally
Forget "your laptop can run LLMs!" claims. I am testing everything from a cheap Chromebook to the queen of local AI the DGX Spark.
Your Laptop Isn’t Dead proves that even older hardware can run useful models — if you know the tricks. I show quantized models running on 8GB VRAM, CPU fallbacks that don’t suck, and why VRAM isn’t the only bottleneck (looking at you, RAM bandwidth).
Spoiler: that cheap laptop won’t run Flux, but it’ll happily run a 3B parameter LLM for coding help at 2 tokens/sec. I will give you the VRAM reality checker: what models fit in 8GB, and show you how to test it on your hardware.
Facts, Not Opinions
Every number here comes from the logs. Every benchmark is reproducible. Every setup is documented step-by-step.
When I say "£0.077 per million tokens," it’s from my local usage calculations and logs easily visible on a dashboard. When I warn about token wastage, it’s because I found out my cost increased without actually doing anything.
I don’t speculate here. I measure. I break things so you don’t have to.
What’s Coming in AI Guides
This introduction is just the doorway. Dive deeper with these upcoming posts:
- The True Cost of Running AI Locally (cost): My week-by-week spend breakdown vs. cloud.
- Your Laptop Isn’t Dead (hardware): Ever wondered what to do with the old Windows 10 laptop?
- The Cloud AI Tax (cost reality check): Why renting AI is like throwing money into a black hole.
These guides exist so you don’t have to learn the hard way.
I’ve taken the hits. Now it’s your turn to run AI — securely, affordably, and on your terms.
Let’s get started.
Found this useful? → Follow @Raf_VRS for more AI Guides → Support the work: ko-fi.com/rafvrs #HardInterference #AIAgents #SelfHosting