I'm Zach Vorhies. Eight years as an engineer at Google and YouTube, and in 2019 I walked 950 pages of internal documents out the door because I saw what centralised AI does to information. Now I run frontier-class models on machines I own — and I set up the same thing on yours. Your data stays in your building. You stop paying per word.
60 minutes · 1:1 on your machine · decentralise your AI
Open-source models. An agent that answers to you. Nothing leaving your building — this is what decentralised actually looks like.
You don't have to be sitting at the machine. Send it a voice note from the supermarket, the site office or the car. It does the work on your own computer while you're away — and it's finished by the time you're back.
"AgentBot4, while I'm away, figure out how to download and configure ProtonMail on my laptop and pin it to my freaking taskbar. And when I get back, guess what? It was done."
I do the same work solo, in hours, starting at $200 — and you keep the machine, the setup and the knowledge afterwards.
Same hour, same hands-on setup. The difference is how much of your setup I build for you.
We set up an agent on your own machine with an open-source model. I walk you through every step. You leave with a working AI setup like the pros have.
🎁 Your $200 comes off the Deep Dive if you upgrade within 14 days.
The Bootcamp gets one machine working. The Deep Dive gets it working the way you actually work — on both your machines, with skill packs built for your job, and me on Telegram for a month while you use it.
🎁 Already booked a Bootcamp? Your $200 comes off this — use it within 14 days.
⏳ I take a limited number of Deep Dives each month. I still work full-time.
For when this stops being a tool and starts being infrastructure. We map how you actually work, set up every machine you own, build the skill packs that replace your repetitive tasks, and I stay available for a quarter.
⏳ Two Full Stack slots a month. It's four hours plus a quarter of support, and I still have a day job.
One hour a week, live, with me. The models I'm actually running, the agents I build for clients before anyone else sees them — and a security desk nobody else in this space is running. Decentralised AI, kept working.
One hour, live, every week. One big topic taught properly, then your questions. Replays archived so you never miss one.
What shipped this week, what got cheaper, what to ignore. Posted in the room as it happens — not a monthly newsletter.
The models I'm actually running, why I picked them, what they cost me, and when to switch. Same setup, no guesswork.
What to buy at your budget, what to skip, and what the same job costs on rented inference. The numbers, not the marketing.
The agents I build for business clients, released to members first — before they go public, and often before they're finished.
Vulnerabilities are being discovered by machines faster than human teams can patch them — and the offensive side always moves first, because finding a hole is a search problem while fixing one is a coordination problem. Most people hear about it weeks late, from a headline, with no idea whether it touches them.
This is the part I actually care about. I worked on Chrome internals at Google, and I know exactly what it looks like when a system decides what you're allowed to see. So when something breaks, you get it here — early, in plain language, and filtered down to what matters for your setup.
AI-discovered vulnerabilities in PHP, NGINX and everything downstream. You hear it the day it breaks — not three weeks later from a news site.
Not a CVE feed. Which of your systems are actually exposed right now, how urgent it really is, and the one thing to do about it tonight.
The safety architecture around your agents — what they're allowed to touch, what must ask you first, and where prompt injection gets in through the documents they read.
Which models, plugins and packages are trustworthy, which quietly exfiltrate your data, and how to tell the difference before you install them.
What actually leaves your machine when you use a cloud model, what doesn't, and why running your own is a security control rather than a hobby.
What goes wrong when an agent gets write access — the irreversible actions, the silent errors, the loops that burn money — and how to fence them off properly.
Eight years inside Google and YouTube. Chrome browser internals. 950 pages disclosed to the Department of Justice. I've been on the wrong end of an organisation that didn't want something public — so I don't treat this part lightly.
"Business owners — do not waste your precious months or years trying to figure out AI yourself. Like what I did."
I've been in business for 8 years and I'd spent several years trying to figure out how to use AI in my business — including ChatGPT projects that took me months and still weren't 100% functioning. I absolutely cannot believe how much we got done.
"I was peppering the poor guy with questions and he had the patience of a school teacher. Not only that but he's just extremely cool to hang out with, and I will be sending every business owner I know to Zach."
"Fucking do it! Just get on and fucking do it! There is Before Zach, and After Zach. It's approximately 19 hours After Zach helped me complete set-up, and I have to remind myself to eat and sleep — it's extraordinary. I know kung fu."
"Jail broke me from Claude's sandbox. I had no idea that I was held captive by their servers."
"He walked me through setting up the agent on my Windows computer, configuring OpenRouter with DeepSeek Flash for cost-effective execution, and linking Telegram bots for mobile and desktop control. I HIGHLY RECOMMEND any course or consultation offered by Zach."
I was a software engineer at Google from 2008, then a senior engineer at YouTube from 2016, where I helped scale the embedded app into billions of homes. Before that I did AI engineering at LucasArts. Google Earth into Audi's in-car navigation; YouTube onto TVs and game consoles.
In 2019 I leaked 950 pages of Google's internal documents to the Department of Justice and testified to the Texas Senate about how ranking systems steer what people believe. I wrote a book about it. That's why I run models on my own hardware today — not ideology, because I've seen what happens when the only copy of the intelligence sits in a building you don't control.
Today I'm the lead contributor to FastLED — 13,000 commits and counting — and I build Rust tooling and local AI infrastructure. I automated an entire engineering department: we used to have a dev team, now we have an AI dev team. I run a weekly AI show, and I set up these exact systems on other people's machines every week.
Because it isn't just time. It's a second machine, three skill packs built for your job, your model routing table, a written playbook, and me on Telegram for a month. Bought separately that's $1,200 — you pay $550.
You can, and it'd cost $400 — but you'd get two hours of my time and none of the rest. No second machine, no skill packs, no routing table, no month of support. Most people who try that come back for the Deep Dive.
Never done this before? The Bootcamp. Done it and it's sitting unused? The Deep Dive — that's the exact problem it solves. Running a business, or more than one machine? The Full Stack.
If you can install software, yes. I do the work on your machine while you watch and I explain everything in plain language. One customer came in having spent months on ChatGPT projects that still weren't working; she left with four self-learning agents she can drive from her phone.
Depends on your hardware, and hardware is genuinely the hard part. One tip from a customer that makes a real difference: have your accounts set up beforehand, make sure your system is updated, and check your RAM against the manufacturer's spec. Do that and you'll get far more out of the session.
Yes — the Full Stack covers a team of three. For a full business deployment, that's a separate scoped engagement starting with a two-hour scoping session. Book a Deep Dive and tell me what you need; I'll tell you honestly if it's worth building.
No. This is a technical service about AI models, hardware and inference economics. If you want the censorship and Big Tech material, that's on X — different room.
Start at $200. Upgrade to the Deep Dive within 14 days and your $200 comes off. Either way, you'll be running your own AI by the end of the week.
