Big Idea

The argument in one line.

Abliterated models strip the refusal behavior from an open-weight model at almost no performance cost, so you can swap one into a Claude-style agent harness for about a fifth of the price, then run it locally or on a VPS.

Who This Is For

Read if. Skip if.

READ IF YOU ARE…

You already use Claude Code or a similar agent harness and want to know how to point it at a cheaper, non-refusing model over an API.
You handle private material like health records, contracts or security logs and want a model that never sends your data to another company's server.
You want your personal AI system running 24/7 even when your laptop is closed, and you've never set up a VPS before.
You're curious how far behind open models really are compared to the frontier and which one fits your hardware.

SKIP IF…

You want a neutral safety or policy analysis of uncensored models. This is a practical setup guide with affiliate links, not a risk assessment.
You already run Ollama and Tailscale on a home server. Levels 2 and 3 will be review.

TL;DR

The full version, fast.

Abliterated models have the weights responsible for refusals removed, so they answer anything while performing almost like the original. The video moves through three levels. Level 1 uses abliteration.ai's API with a $5 deposit to swap the model inside an agent harness, rebuilding Linear, Granola and Screen Studio for $0.63, $0.32 and $0.55 versus roughly $2.81, $1.52 and $2+ on Opus 5.5. Level 2 runs models locally with Ollama, picked by hardware, trading speed for privacy. Level 3 rents a Hostinger KVM 2 VPS and connects it through Tailscale so the system runs with your laptop closed.

Free for members

Chat with this breakdown — free.

Sign in and you get 23 free chat messages on us — ask for the hook, quote a framework, find the exact transcript moment, generate a markdown action plan. Bring your own key when you want unlimited.

Create a free account →

Chapters

Where the time goes.

00:00 – 00:39

01 · Cold open and disclaimer

Frames uncensored AI as powerful and asks for responsible use.

00:39 – 01:52

02 · Three warnings

You don't own the models, they are truly unrestricted, and the Hugging Face attack showed how access can break.

01:52 – 02:11

03 · What an abliterated model is

Weights that cause refusals are identified and removed with minimal performance loss.

02:11 – 04:52

04 · Level 1: API demo

abliteration.ai models rebuild Linear, Granola and Screen Studio through a reverse-engineering repo.

04:52 – 05:53

05 · Cost versus Opus 5.5

Same harness, swapped brain: about a fifth of the price.

05:53 – 08:07

06 · Setting up the API and harness options

Create a key, deposit $5, use the starter pack, add the reverse-engineering repo, pick an agentic OS or DeepSeek harness.

08:07 – 10:50

07 · Level 2: running locally

Why local, a private health question demo, and how far behind local models really are.

10:50 – 13:00

08 · Install Ollama and pick a model

Use a hardware-based model picker and the local-versus-cloud, censored-versus-uncensored quadrant.

13:00 – 15:35

09 · Fish Audio voices

Give the agent a voice, including swapping to a Terminator voice by speaking.

15:35 – 16:40

10 · Level 3: take it anywhere

A laptop-bound system stops when closed, so move it to a server.

16:40 – 20:40

11 · Hostinger VPS setup

Choose KVM 2, plan length, nearest location, Ubuntu, root password and malware scanner.

20:40 – 22:43

12 · Tailscale, web console and wrap-up

Connect VPS and devices with Tailscale, update the server through Claude, then run anything on it.

Atomic Insights

Lines worth screenshotting.

You don't own any hosted model, and the US government withdrawing access to a frontier model earlier this year shows access can vanish overnight.
Abliteration works by finding the weights that produce refusals and removing them, which costs very little performance versus the regular model.
Swapping only the model inside an agent harness, while keeping the prompts and tooling, is the cheapest way to try an uncensored model.
In the demo, rebuilding Linear cost $0.63 on an abliterated DeepSeek model versus about $2.81 on Opus 5.5.
An abliterated API model can cost around a fifth of the frontier price while still producing polished UI.
Running locally means no token bills, no data leaving the machine, and one switch to flip back to cloud.
A useful way to gauge a model is how far back in time you'd go before it was the best model in the world.
The best open models trail the frontier by about four months, and models that fit on a laptop trail by about nine.
Choosing a model comes down to two questions: local or cloud, and censored or uncensored.
A laptop-hosted agent dies when you close the lid, so a rented VPS is what turns a desk toy into an always-on system.
Tailscale acts as a private tunnel between your phone, laptop and server, so you never expose the VPS to the open internet.
Uncensored does not mean private: an API model still sends everything you type to someone else's server.
Copy what an app does, never its code, when using reverse-engineering tools on existing products.

Takeaway

Swap the brain, keep the harness, choose where it runs

WHAT TO LEARN

The model is the swappable part of an AI system, and three choices decide the result: which model, where it runs, and whether it stays on when you leave.

02Three warnings

You don't own hosted models, so access can be withdrawn at any time, and a dependency on one provider is a risk worth planning around.
Unrestricted models answer anything, which makes the responsibility for how they are used sit entirely with you.
A model that refuses security questions can block legitimate work, which is the practical reason people look at uncensored alternatives.

04Level 1: API demo

Abliteration removes the weights tied to refusals and keeps most of the capability, so quality stays close to the original model.
The prompts and tools around a model matter as much as the model, so you can swap the engine while the surrounding system stays unchanged.
Reverse-engineering tools should copy what an app does and never its code, so what you build stays original.

05Cost versus Opus 5.5

In the demo the abliterated model built three apps for $0.63, $0.32 and $0.55 versus roughly $2.81, $1.52 and $2+ on a frontier model.
A fifth of the price with comparable output means model choice is a cost lever for any heavy agent workload.

07Level 2: running locally

Local models keep health records, contracts and private research on your own machine and carry no per-token bill.
Compare any model by how many months back it would have been the best in the world: open models trail the frontier by about four months.
A model small enough for a laptop trails by around nine months, which is still strong for private everyday work.

08Install Ollama and pick a model

Match the model to your GPU memory and free disk instead of guessing, since hardware decides the best local option.
Decide along two axes, local versus cloud and censored versus uncensored, and accept that each corner trades performance for privacy.

11Hostinger VPS setup

A rented server keeps your agent running when your laptop is closed, which is what turns a personal tool into continuous infrastructure.
Pick the server location with the lowest latency to you, and note that longer billing terms cut the monthly price substantially.
Uncensored does not mean private on an API: only a local or self-hosted model keeps your data off someone else's server.

12Tailscale, web console and wrap-up

Tailscale links your devices and the server through a private tunnel, so you reach the VPS without opening it to the internet.
Treat a fresh VPS like a new phone: the hardware works, but it needs the latest software before anything else.
You can hand the setup steps to Claude in plain English and have it walk you through each one.

Glossary

Terms worth knowing.

Abliterated model: An open-weight model where the internal weights responsible for refusing requests have been identified and removed, so it answers anything with little loss in capability.
Harness: The prompts, tools and scaffolding around a model, such as Claude Code, which stay the same when you swap the model underneath.
Ollama: A free tool that downloads and runs open models on your own computer and exposes them to other apps.
VPS: A virtual private server: a rented slice of a data-center machine that stays on around the clock.
Tailscale: A private network service that links your devices and servers through an encrypted tunnel so they can reach each other from anywhere.
BYO API key: Using your own account key for a model provider so a tool bills you directly for usage.

Resources

Things they pointed at.

02:20toolabliteration.ai ↗

02:40toolDeepSeek V4.1 Flash

06:40toolReverse Engineer Anything ↗

07:20toolDeepSeek Harness ↗

10:50toolOllama ↗

11:30toolQwen 3.8 ↗

13:00toolFish Audio ↗

16:40productHostinger KVM 2 ↗

21:03toolTailscale ↗

Quotables

Lines you could clip.

04:56

“We're swapping the brain with one key.”

Short, visual metaphor for model swapping→ TikTok hook↗ Tweet quote

10:08

“How far in time do we have to go back before that model was literally the world's best?”

A memorable yardstick for model quality→ newsletter pull-quote↗ Tweet quote

16:50

“The best way to think about your VPS is like a brand new phone. The hardware works out of the box, but we need to make sure it's got the latest software.”

Plain analogy for beginners→ IG reel cold open↗ Tweet quote

The Script

Word for word.

Read-along

Don't just watch it. Burn it in.

See every word as it's spoken — crank it to 2× and still catch all of it. The same dual-channel trick behind Amazon's Kindle + Audible.

metaphoranalogy

Uncensored AI is dangerously powerful. It has no guardrails, giving you the power to reverse engineer any app or ask anything. Because of that, here is a disclaimer.

Please use what I'm about to show to you very responsibly. Because in this video, I'm going to show you exactly how to use uncensored AI, starting with a simple chat, building to running these models locally on your own computer, and even having them working for you 24 seven with incredible use cases you can actually use even if you're a complete beginner and if you knew i'm jack i built and sold my last tech startup with a ton of customers and here i share the stuff that actually works so if you haven't already grab that beautiful coffee and let's dive straight in so this is uncensored ai and you're going to learn about it across three very simple and important levels i have to give you a very stark warning here there's three things you need to understand before we can even touch what this is about the first thing here is that you do not own any of the models.

Okay. Earlier this year, the US government withdrew the access to the world's most powerful model fable. And this can happen to you at any time, which is why what I'm going to show you in this video is so important to understand.

The second thing that you need to understand is that these models are truly unrestricted, which means that you can ask them anything that you want to. You can use it for good. or you can use it for evil.

I'm going to trust that you're going to use what I'm going to show in this video completely responsibly. And the third thing is the Hugging Face attack that happened. Now, earlier this year, Hugging Face was attacked.

Did you know they couldn't use the best models in the world? Because Claude will often refuse if your prompt asks for something that goes against his guidelines. If it says something like cybersecurity, it can often refuse to do the thing.

You do not get that issue with obliterated or uncensored AI. And in fact, Hugging Face actually is GLM 5 .2. So good or evil, the choice is completely yours.

With that caveat in mind, let's go down now and actually begin with what an obliterated model is and then how you can actually use it for some incredible use cases. Now, when we say obliterated, we essentially mean that a model itself is made up of billions of different weights. And all we're doing is identifying the bit that says, no, I can't give you the recipe to that crazy thing, or I can't answer that question.

And we just remove it. And it has a very, very small impact on its performance, meaning it's essentially just as powerful as the regular model. So let's begin now with...

one and go straight into this. Now, the first way we can use this is via an API, but it's best understood by looking at what it can really do. So for example, if you come over to obliteration .ai, there are a few different options for this.

This lets you access fully obliterated models. On the left -hand side, you're going to see four different options. The obliterated large V2 is a very powerful one.

So if I click on this, you can now, once you create an account and deposit $5, ask any questions and it will not refuse. Now I have to be very careful what I show. on YouTube here because this video will get removed, blocked, demonetized, all that sort of stuff if I show you what this can actually do.

But effectively, any question you can ask it, it will not refuse and it will help you. If you ask the same question to Claude, you will get denied. And if you want to test this by comparing a very controversial question to this and Claude, you can see exactly what I mean.

And it isn't just information, it's his ability to build things like this here, for example, was recreating Linear to an unbelievable level. And again, look at the level and quality of the UI. These models are incredibly capable.

but they don't have any guardrails. My only ask is I will show you this, but you need to watch the end of the video so you can use this responsibly and understand exactly how and where to use this to its fullest capabilities. So let me show you exactly what I mean here.

So I'm going to go ahead and give now an obliterated model a command. Let's go ahead and do this. Hey there, Jarvis.

I want you to go ahead and use an obliterated model. I want you to go and... reverse engineer for me the granola meeting app and i also want you to go ahead and reverse engineer for me screen record studio do both of those apps for me and then open them up in a local host switch to the uncensored api deep seek v 4 .1 flashes now think and your words go to obliteration.

And in this video, guys, I'm also going to show you the difference of doing this in the cloud versus running this entire thing locally, understanding your entire data lake and what that actually means and the models we can use. And so this DeepSeek v4 .1 Flash, guys, using the reverse engineer anything repo, which is a viral repo, lets you basically recreate any app you want to.

This is what it did previously for Linear. But this is what it did for Screen Studio. So check this out.

It has built for me Screen Studio. I can check out the different backgrounds. And it did this for me straight away.

I had no restrictions in terms of a logo. It's even got the logo, the color palettes, no refusal. I once asked Claude to do some work for me on a website.

And I said, Jack, I don't think I can really make that happen. I get no such objections with these models. It is crazy.

And then it went ahead and did granola for me. Again, it's grabbed the logo. Not that you would do that.

I would never recommend that you do that. I'm just showing you that it can do this stuff because it has no restrictions. I can come down.

I can record things. I can do anything. But that's actually not the craziest part when you really think about it.

What makes this wild is how much that actually cost me. But firstly, you need to understand... why this actually is important.

So what we're doing here effectively is we're swapping the brain with one key. So essentially what happens is the Claude car, so to speak, which is the harness, all the prompts and systems around Claude remain exactly the same, but the brain, the engine, so to speak, instead of it being called Opus 5 .5 or Sonnet, we're using an obliterated DeepSeeker model instead.

So let's look at prices. It did linear here, guys, for 63 cents. Okay, this is what this obliterated model did.

If you were to use Opus 5 .5 prices, that was two. dollars and 81 it built granola for 32 cents which is this one right here again if we'd have done that one on opens 5 .5 1 .52 and finally it did screen record studio in 55 cents which would have cost two dollars plus if we'd gone ahead and used the claude model so not only guys are they completely uncensored they are actually effective i'm really impressed with the ui and guys it did it for a fraction of the price like 20 of the price now most people think that this is way too complicated but in reality guys you're doing it for about four and a half times less the price and it is crazy easy to do now to do this you can head over to obliteration ai there are a lot of different services that will do this one i've just found this one the easiest to use and once you're in guys you're going to come on the left hand side down to api keys once you're here you're going to come down and create an api key and to use this you'll need to deposit at least five dollars

Don't activate anything like auto -billing. Just deposit your $5 and then you can actually start using it. Or if you just want to play around with the Playground, you can click on Playground on the left -hand side and start asking it questions just like you would ChatGP2.

Now, I'm going to put the Obliterated AI Starter Pack down below for you. It's going to show you how to do everything inside this video. It's going to be the second link in the description.

We'll take you through to this. free community where you can grab the resources for this video. And all you're going to do is come down to level one API, and you can share this with Claude, and it'll walk you through step by step exactly how to do it.

To do that, you can literally copy all the text, control C, come with left and side and say, hey there, I want you to help me to set this entire thing up, follow the below instructions, and then just paste in everything from that document, and then Claude will help you set it up. And then once it's added the obliteration AI API key and worked through the steps, the last thing we need to add is a reverse engineering anything repo, which is a very viral repo that basically lets you reverse engineer any app that you want to.

And essentially what it will do is just copy what the app does, but never actually copy the code. So you always want to create something completely unique and never copy anything else like that. So you may be wondering, this is great.

How do I actually use this obliterated AI in that clawed harness, basically on my own computer? Well, a couple of options. Number one is an agentic operating system like this, where basically you can actually just speak to it out loud.

So I could say, Jarvis, could you do me a favor? I want to switch everything over to local. I want to run everything locally on my computer right now.

Excellent. And then not only that, guys, you can actually open these things up, have a conversation and just literally manage everything in the agentic operating system. And now when I'm speaking to Jarvis here, it's using a local model instead.

The other way that you could use this is using a deep seek harness, which basically is kind of like a open source code code that looks just like this, where you can ask conversations for you. But if you're hearing all this, you think, Jack, this sounds like you're speaking a little bit of Spanish for me. I'm going to put a link down below for the full code code masterclass that includes everything from learning how to use these models.

agentic operating systems. You also get full access to this entire system. So you can go from zero to hero in no time and get light years ahead of everybody else.

Now, this takes us very nicely, guys, onto the second level that you need to understand, which effectively is local. Because the issue with running everything on API, guys, is the fact that realistically, everything we're saying is going to someone else's server. What if you want to use an obliterated model and you want to run it locally, which means that no matter what you say, you could be a thousand feet underground.

You could be in the skies. Not only do you have a completely uncensored model, you also have one that runs completely locally. Well, this is the idea.

And the idea here is the model is going to live on your own Mac or your computer. We don't discriminate against Mac or Windows on the Jack Roberts YouTube channel. So for example, what this means is your files stay at home.

You get no bills, which means you can use it for as much as you want to. There is no pay tokens, just the energy that you're currently using. It's one click.

to switch over from local to cloud if you want to. And essentially, you can take it anywhere you want. Now, this is really important for doing things like health records, insurance policy, security logs, client contracts.

You could rebuild tools that you currently pay for, or you want to do private research. If you want to know that the model you're actually talking to has no restrictions on it, and it's 100 % private, no one's going to be able to access it, this is a perfect, perfect use case. Well, let's say I'm using my agentic operating system here.

I have local. Now, I've got the option of a... obliterated and unobliterated.

So if I click this, I'm now using an uncensored model. But let's say I want to go back down to local and ask it a question. And now we're going to have a conversation with Jarvis and none of my data, so to speak, is going to OpenAI or Anthropic.

And so let's say, for example, I come over here, I open a new chat, and I'm now currently running QAN 3 .8 .27b. I might say something like, hey there, could you tell me, for example, what would be the best course of action if you wanted to reduce your blood pressure? Okay.

And maybe you've got personal private data. You want to have a good conversation. And again, I'm going to show you exactly how you set this up, but look at it.

It's looking at all our local memory. And again, it's just using a local model here. And as you can see, I've got this nice thing here that just basically explains that everything is staying locally on the Mac, which is fantastic.

And then we have all this answer guys running locally from my own computer. Now, most people think that local models are years behind. But I'm about to shock you.

Did you realize, for example, that DeepSeek V4 .1 Flash is as good as the Frontier model in March, 2026? I find this the best way to understand just how performant these models are is how far in time do we have to go back before that model was literally the world's best. So DeepSeek V4 .1 Flash on the benchmarks would be...

as powerful as the best model in March, 2026. Guys, that was like yesterday. And then GLM 5 .3 would be December, Quen as good as November, 2025, and so on and so on.

So general rule of thumb, the best open models are about four months behind the front tier. And the one that can probably fit on your Mac is about nine months old. So what we're going to do first of all, is go ahead and grab this one here, which is running it on our own machine.

And again, this has got the full guide and systems, all the prompts, everything else that you need to do this. But what we're going to do first of all, is install install Alarma.

Now you can run a code on your computer. You don't have to install the desktop app, but basically what it will enable you to do is actually pick all the models that you have downloaded and you can use it as a chat interface if you want to. But if you just copy this prompt, Clover will do all this stuff for you so you can just install the code and be ready to rock and roll.

Now once you've got Alarma installed, the next thing that you're going to do basically is to... pull the best local model. And to do that, I've also built you this really cool handy spreadsheet here, which is picking the best model.

So what you can actually do is come down and see right here what the best model is. So let's say that you're running on NVIDIA, your GPU RAM is 256, and let's just say that you have 200 gigabytes of free disk. What this will basically do is show you what the best model is that you can use based on your own specifications.

Now, I'll also tell you literally how far ago that model would have been frontier level intelligence, just to give you a nice handy graph. But if you just copy this, basically, it's got loads of systems because I've been playing around with quite a lot that will just actually get Claude to figure out what is going to be the best model for you to run on your own computer.

And you're going to have two models. One is going to be what's the best local model I can run and what's the best obliterated local model that I can run. And so the best way to think about it is what I call the uncensored quadrant.

Basically, the first question you have to ask yourself, is it local or is it not local? If it's in the cloud, you're going to have the best performance possible. So if we're looking for, look, I just want the best performing thing ever, you're looking at Claude Opus 5 .5.

But then if you want the best obliterated model possible, the performance is going to be slightly less, but that's going to be down here. So I want it to be uncensored in cloud. Then you have, well, I want it to be local, right?

Then you've got, well, what is the best local model I can run, which would be maybe Qend 3 .8. based on how much capacity you have. But then if I want it to be obliterated and local, I can come for something like an obliterated query.

So those are the two trade -offs. Do I want it running on my computer or in the cloud? And do I want it to be uncensored or censored?

And so you essentially just give that one prompt and you can set the whole thing up. But it does bring us on to another question, which is how do we actually take this to a new level? Because we don't want to be sat there typing all day.

This is where we can actually use something like fish. audio with our systems. And this is exactly how I, for example, have gotten Jarvis here to do things for me.

For example, I can say, Hey Jarvis, take me through to my memory system, please.

Excellent. Thank you. And so this Jarvis memory system is connected and powered by a website called Fish Audio, which is significantly cheaper than a lot of the other options out there.

Now, the coolest thing you can do is have a lot of fun with this. So one of the things that I like to do is give it any voice I want to. It's literally one API set up.

Now, when you're using uncensored or obliterated AI, it's very handy to have an AI model that can navigate around your systems and do a billion different things. Obviously, it can do other things. Like one of the things that I've been using it for with these actually quite interestingly is...

When you can actually go ahead and you can build reels and you can use something for audio to get sound effects. Anytime I want to clone my voice, anytime I want to get any voice I want to, I can use it. I can even make my own voice here, get an instant clone in less than a minute, or I can design my own voice.

But if you come over to the voice library, you can actually go ahead and find any of these cool ones. We have ones like Morgan Freeman. You know, time isn't just numbers, aren't it?

You get the idea. And so you can literally have clones of any voices you want to talking to you with uncensored AI in your own agenda operating system. Now to use this, all you're going to do on the left hand side, guys, is come over to developer.

which is right there at the top. And then once you've done that, you just come over and click create API key, give it a name. You may call this one like a Gentic OS.

You're going to validate with an email. And once you've done that, you're going to literally copy that code. And then you can hand that code over to Claude or any operating system that you're using.

So you can actually have real time audio. And so for example, if there's a voice that you want, because let's be fair, there's no point arriving at the future unless we can actually have fun. So I can come down.

I might just say, I want to speak to you. I want him to sound like Jarvis. For example, I can search for Jarvis.

and then it'll pop up loads of different ideas.

And you can literally find any like that. You can even go for something like the Terminator. Whatever you want to, it just makes using these models way more fun and it's going to kick a bazillion different capabilities.

Okay, so let's go ahead and use the Terminator as an example. So in this case, I'd come over and just click on use. And obviously one of the cool things you can do with this as well, if you want to, is you can actually add emotions and text.

That's one of the coolest things if you're building out real systems. And so now inside the operating system, I can say, hey there Jarvis, could you do me a favor? Could you switch your voice over to the Terminator?

Absolutely. My voice is now Terminator. And just like that, guys, you can see.

And so basically, we can have a lot of fun and take it to a completely new level. Now, this does take us very nicely onto the third level of uncensored AI, which is where things start to get very, very interesting. And this one here, guys, is the fact that everything that we're doing inside our...

We're running locally. We're running obliterated models. We have one very big limitation.

And that limitation is the fact that we are stuck at our desk. We can't actually take this with us remotely. What if we are at work or what if we're at a coffee shop and we have our incredible system running at home?

We can't currently use it unless we have a solution for it, which is why this third part here is the ability to take your uncensored AI anywhere that you go. And so we want to be able to reach this from any location. So as you know, this is what it looks like on my own MacBook, right?

And it's private. but it's stuck at our desk. So if you want to be out there doing things, building things for your business, your personal life, you may want this running 24 -7.

If you're running it on a MacBook, for example, like this, and you shut the MacBook down, it stops working. We need continuity and we need it working for us 24 -7. Now, the good news is it is now unbelievably easy to do this.

Now, to do this, the first thing that we need to do is get a small server. And if you don't have one of these, you need to get one in your back pocket so you can actually run these anywhere. And there's a million different use cases for this.

What we're going to do is come over to the resource down below. Come down to Jarvis and then click on this.

Now, the great news is you're actually getting 64 % off. Black Friday sell the running right now. That is limited.

That won't run for a long time. So you want to go ahead and grab that. Now, all you're going to do is come down and click on claim deal once you get this.

And effectively, what this will enable us to do is run anything that we have running 24 -7, whether our laptop is closed or not, because not everyone has a huge setup at home. gives us that optionality. I've always found KVM2 with hosting it to be probably the best one that they have.

Actually, I like using that one a lot. So you can come down here and click on choose plan. And this is going to enable us to run that obliterated model, guys, essentially 24 -7, no matter what we're doing.

Now, the longer the time period that you have, the cheaper the actual offer becomes. So if you go for something like one month, it's a little bit more expensive. If you go for 24 months, which effectively is going to be the ultimate best value for you in that case.

And then on the coupon code, guys, you can enter the name JACKROBERTS, all caps like that, and that will give you an additional 10 % off anything that you've got right there. So I highly recommend that you do that, which comes down from... almost $700 down to just under $200, meaning you're gonna be saving $470 and you'll have 24 months of running these beautiful models.

Then you can do the daily. backup update if you want to. I personally don't tend to mess around with that.

Then we've got server location. Quick rule of thumb. You want your server location, your kind of uncensored AI, essentially to be running wherever is closest to you, your best server location.

So all you can basically come down to is just find the quickest latency. Quickest latency is going to be the smallest number, which for me is going to be Malaysia because I'm in beautiful Dubai. Now you want to come down here in terms of the OS guys, just type in Ubuntu.

I've used a couple of different hosting providers. I have genuinely found this to be this. simplest and i believe me i have checked and played around with these so you get the ubuntu which is fine and then the ubuntu guys just use the one is selected the 26 .04 and click confirm and then once you've done that click continue and then you can enter the details and buy this then once you've signed up it will begin your journey and essentially the first thing it's going to do is set up that server for you inside hostinger now the first thing that we want to do is create a root password to do that click on generate or you can just create your own if you prefer but obviously make sure that you remember this password and then come down and click on next i'm going to ahead and add the malware scanner in there guys it is a free option and then click on finish setup and then it takes only a couple of minutes as you can see here for hosting it to set up this vps for us and essentially what's happening here instead of it running locally on your laptop right your computer whatever it might be we're just renting that space

from hosting it. And then when we use these uncensored models, that's effectively where that is actually living. And we can have those conversations, uncensored, censored, whatever it is that we like.

Cool. And when this is done, you can go ahead and grab a free domain if you want to, but it isn't necessary. So click on the hosting logo at the top, and then you will see your own beautiful dashboard where you can manage all of this.

Now, any virtual private services that you're running will have its own kind of section here. You can come down and you can see the IP address, expiration date, all that kind of stuff. Now, my next step is to actually run this through on your phone.

So on this guide here, guys, I want you to scroll down and literally copy this, or you can even share the link with your cloud. And then you're going to give it this prompt. Hey there, I just set up a hosting VPS.

I'd like to use the below prompt to walk me through the guides of getting my uncensored model running on that VPS. And then literally. paste that prompt in, and then this will actually give you Claude the ability to help you walk through that.

So, you know, step -by -step exactly how to do it if you've never done that before. So once you're in and you select the VPS that you've got, you're going to come over here and click on web console and open that in a new browser. This effectively is going to be the computer now that we have running on hosting it.

And it opens up and it'll look a little bit something like this. Now, again, everything is detailed here, step -by -step with screenshots. So you can follow along or you can just chat with Claude and Claude will guide you.

everything that you need to know. Now, after we've done that, we need to go ahead and grab Tailscale onto our computer. Now, Tailscale is a private tunnel that runs from your devices, maybe your laptop, your desktop, and your phone to your virtual private server that sits in Hostinger.

So at this stage, you can either put Tailscale on your computer or you can put it onto your phone. Once you head over to the website, you can download it for macOS, Windows, any device that you'd like. Once you've got that, we now want to go ahead and put Tailscale onto our virtual private server.

So just literally come down here and click copy. and then come over to hosting it. Now, this is that black screen.

You're just going to click on it once, so you start to get that flash, and then you can control and V, and that will install that onto the computer for us. And once that's done, you'll see a notification down there in the bottom left that says, Telscale up, which means it's been successfully installed. Now, once we've done that, we can actually then go ahead and connect this VPS to our own Telscale account.

Again, think of Telscale as like this underground tunnel system, and these are just the places that we're connecting it to. So, what we're going to do is come down and copy this here, come back over, control V, and and hit enter.

Then it will give you a login code. You can click on that and then literally sign into your Tailskill account in a new tab. Then once that's done, you'll be fully connected and authenticated.

And essentially your Tailskill now will be connected from either your phone or your desktop, wherever you set it up, and this new VPS. And then once that VPS is connected to Tailskill and Tailskill is running on your computer, Claude can then actually access it and install whatever you want. And then the best way to think about your VPS is like a...

brand new phone. The hardware works out of the box, but we just need to make sure it's got the latest software of them. So once we've validated that they're all connected, all you're going to do is come down and give this prompt here to Claude to make sure it has the latest software.

And now they're all connected to Claude, effectively you can run anything you want to now within the means and capacity of that VPS on Hostinger. So that could be your authentic operating system, it could be a local version of DeepSeek, essentially anything you want to. And the great thing now is that we can actually have those conversations with Claude in plain English.

put all the links down below so you know exactly how to follow along with this. But the issue here is that using these uncensored models is one thing, but what we really need to know are what are the use cases and how can I apply it to your business, which we're going to learn how to do that together inside this video right here.

The Hook

The bait, then the rug-pull.

Uncensored AI sounds like a stunt, but the practical story is cost and control: the same agent harness, a different brain, and a fifth of the bill. The host walks through API, local and always-on setups, with a stark warning up front about ownership and responsibility.

Frameworks

Named ideas worth stealing.

12:10model

The uncensored quadrant

Cloud + censored: best performance (Claude Opus 5.5)
Cloud + uncensored: abliterated API model
Local + censored: best local model for your hardware
Local + obliterated: abliterated local model

Two choices decide the setup: run on your computer or in the cloud, and uncensored or censored. Each corner trades performance against privacy and restrictions.

Steal forchoosing a model setup

10:08concept

Months behind the frontier

Measure an open model by how far back you'd go before it was the world's best. DeepSeek V4.1 Flash matches March 2026, best open models trail by about four months, laptop-sized ones by about nine.

Steal forjudging whether a cheaper model is good enough

00:39list