---
title: 'AI Agent OS: Building Websites, Obsidian Setup, and Opus 5 Performance Issues'
source: 'https://youtube.com/watch?v=rlsQhVwCbsw'
video_id: 'rlsQhVwCbsw'
date: 2026-07-28
duration_sec: 561
---

# AI Agent OS: Building Websites, Obsidian Setup, and Opus 5 Performance Issues

> Source: [AI Agent OS: Building Websites, Obsidian Setup, and Opus 5 Performance Issues](https://youtube.com/watch?v=rlsQhVwCbsw)

## Summary

This video addresses community questions about building AI agent operating systems, using Claude Code for website replication, setting up Obsidian with the PARA method, and the performance issues of Claude Opus 5. The creator shares practical tips for iterating over agent OS builds, personalizing email outreach with AI, and leveraging community templates.

### Key Points

- **Getting Started with AI Agent OS** [00:01] — The video discusses how to get the most out of an AI operating system, connect AI agents, and sell these solutions to clients. It encourages iterating and testing rather than expecting a single prompt to work.
- **Building Websites with Claude Code** [00:26] — A GitHub skill allows Claude Code to replicate websites in three steps: install the skill, provide the target URLs, and ask Claude to use the skill. This method is effective for learning how to build similar websites.
- **Building Agent OS for a Client** [01:04] — To build an agent AI operating system for a fitness client with VPS access, the creator recommends using the AI Profit Builder template, goal mode, and iterating. For VPS setup, Tailscale or Cloudflare can be used.
- **Codex Voice Mode and Opus 5 Slowdown** [02:42] — Codex now supports voice mode that can orchestrate the entire screen, but it is slow. Claude Opus 5 is significantly slower than Sonnet 5 – taking up to an hour for tasks that Sonnet 5 completes in 20 minutes. The creator switched back to Sonnet 5.
- **Obsidian Folder Structure with PARA** [04:17] — Obsidian works well as a memory system for agents using the PARA method (Projects, Areas, Resources, Archives). This allows agents to store and retrieve memories, such as login details and project information, across the entire agent OS.
- **Community Progress and Automation** [05:54] — A community member set up three custom GPTs from classes, demonstrating that even one automation per month can save significant time. The creator encourages consistent action.
- **Building the Dashboard Agent OS** [06:20] — The dashboard agent OS was built using Claude models (Sonnet 5 and Opus 4.8). Community members get a zip file to install it and can use various models including Codex and Hermes.
- **Email Personalization with AI** [06:59] — For personalizing email opening lines, the creator recommends using Hunter API with Claude to analyze leads. Claude can write personalized sentences in a spreadsheet column. For large lists, it takes time, but the approach scales.

### Conclusion

The video provides actionable advice on building AI agent systems, leveraging tools like Claude Code and Obsidian, and personalizing outreach. It emphasizes iteration and community templates to accelerate progress.

## Transcript

operating system, how to get the most out of it, how to connect your AI agents, and also how to sell this to clients. So, let's get straight into it. This question's from the AI Profit Bot and community, where we help people with
AI automation. So, if you do have a question, feel free to ask and post this everyone daily. So, one of the first questions we've got here was from Faizan, and he was was asking about like what's the best, the most efficient way
to build websites similar to these. This is actually a really good skill for this on GitHub, which you can see right here. And basically, this works really well with Claude Code. So, you can use Claude Code to learn from other websites, and
that. And what you can actually do is just say that GitHub, for example, then you can go into Claude Code, you can that, and then give it the websites that you'd like to build similar websites to.
So, it just works in three simple steps. You install this skill to Claude Code, then you're going to give Claude the websites you'd like to replicate, and you're building, and then just ask Claude to use the skill from step one to
another question here from Nick, and Nick is talking about how to build an agent AI operating system for his client who owns a fitness. His client also wants to be able to access it on a VPS. So, this is the prompt that he's asked
say at this point, it's probably not going to be just one prompt that builds loop, it's got to iterate, it's got to test, it's got to check its own work. It's got to keep improving that. And bear in mind that with an agent OS, you
just constantly iterate and improve it. So, our own agent operating system, we're building out daily. So, what you can actually do instead is you could So, let's say, for example, it's Claude Code that's setting up the agent OS. You
could actually use our template inside the AI Profit Builder to set it up, give client wants, and then use something called goal mode to build it out. Now, the reason I recommend using go mode is because it needs to run on a loop to
build this out, right? So, you can use {forward slash} go inside Claude of Codex whenever agent uses to build this. Give it the zip file of the agent operating system, and give it the specifications of the client's
needs a fitness app and a social media app, then you can build it out from there. Then, the final step would just be making sure that it actually works, when it comes to set up on a VPS, you could use Tailscale, or you could use
Cloudflare. I actually linked some examples of how I've already done that for example, you see here John Mesa, he created the agent OS on a VPS, and the And then, you can actually use it from
mobile app, from desktop, but also, this means that you can give your client it from there and host it. And they were also saying about Codex having Java now. So, if you haven't already checked this
the app I've already done, but basically, Codex can now use voice mode, to operate your computer, and you can also use browser using, can use all these tools. But basically, the voice mode can orchestrate your whole screen.
It's pretty cool. The one thing that I would say is that it's a little bit slow future of it. I can see how powerful it is and how useful it could be, but right going to break, it's still going to struggle on some things. But we do have
a full demo testing out, as you can see right. And when that is actually really powerful it will be. This is a really interesting question. I literally I had exactly the same problem earlier today. So, when I was using Claude Opus 5, it
can do pretty good outputs, but it is super slow. Like, something that would normally take Sable 5 about 20 minutes, I've seen Opus 5 take about 1 hour to do. It's crazy. So, it's great that we've got a model that is Sable 5 level,
and it's cheaper, but right now is not fast enough to build anything substantial with. So, I actually changed all of my settings since I Claude just to switch back to Fable 5. I think that if you do run out of Fable 5 tokens,
either just use the credits cuz it's going to save you time anyway, or switch over to Codex cuz GPT-3.5 is really good, too. But yeah, for sure, Claude Opus 5 is super slow, particularly over the last couple of days. Reach posted
about Obsidian folder structures and the best way to set this up. So, basically with Obsidian, Obsidian works really well inside of an agentic operating memories, all of the folders, all of your files set up in system like this so
you can visualize. Now, these are just markdown folders and files. The way that we actually set this up is with something I call the infinite context engine. So, our agents upload all of their memories into Obsidian, and then
all of our agents across our whole agent operating system can pull in those memories and use them. So, for example, we were using Buzz earlier today, and Buzz had instant context of everything that we work on, every project we use,
every website we work on. Even have, for example, login details to our WordPress websites, as you can see right here. I could like post directly to our websites for whatever keyword we give it. So, the way that we set this up is we use
something called PARA. This stands for projects, areas, resources, and archives. It was actually designed by a productivity X so that it can sort information by actionability rather than topic. So, there's four different
it's really easy for your agents, particularly in a business context, to voice, your goals, your brand, projects, and any companies you have to. So, you here. We've got the AI Profit Room. We have our funnel strategies. We have Open
inside one beautiful system. So, I definitely recommend checking out an something that's really cool about this community is like people are actually what they learn. So, you can see, for example, Muhammad, shout out to
Muhammad. He set up three custom GPTs from some of the classes, and that's like three new automations they set up based on what he's spending time on. So, imagine, for example, he set up three automations per week, that's 150 by the
that many, but even, for example, if you do one a month, like it's going to make a massive difference in terms of the time you save, in terms of the progress you make, and also it's just great to see people taking action. So, shout out
to Muhammad. Jeremiah asked, "Which brain built your dashboard agent OS?" So, for me, personally, I actually use Claude to build it out. It's actually called Fable 5. Claude Fable 5, I think, is a four. And we began it with Claude
ago, cuz all these five-generation models weren't out. But, yes, Fable 5, Opus 4.8, it wouldn't have been possible to build out the agent OS system without those. But, the great thing is that everyone inside gets access to a zip
file to install it, and you can use any of these models set up. You know, you could use Code XO, you could use Hermes, you could use Claude Fable Opus, whatever you We've got another question from Uzi, basically focused on email
personalization. He's asking like, "What's the best way to efficiently "What's the best way to efficiently personalize the opening line for each prospect that he's reaching out to?" So, the way that we actually did this is we
the way that we actually did this is we used a combination of Hunter API with Claude. And it analyzes each of the leads, so we can type in any type of automatically enrich those leads using a combination of Hunter API and Claude.
So, if I had to do this all over again, I'd probably use something like Claude, to reach out to, and just have Claude write like the first sentence of for each lead. And bear in mind that Claude can create and edit documents, so
do have thousands of prospects, then obviously it's going to take a long you would have it working on hours. And then, the way that I set it up on the spreadsheet is I would have a name, the email, and then also the one sentence
have more information that you can plug into the spreadsheet to give Claude more context on the lead itself so that it can personalize the message better, couple of skills. So, there's something called lead generation strategy on MCP
Market. I've not tested out. And there's another one called lead research assistant. It's part of Awesome Claude skills. I've not tested those out. But, the research on the leads, and then use that to write the personalized sentence,
out, you can just use, for example, Hunter or to whatever software you that Claude is filling for you. And that's basically it for all the question inside the community. If you want me to create a video tutorial for
like I did today, you can get that inside the AI Profit Portal and just loads of great training. It's all my best training. So, I'm for example, how to go from a beginner to expert in AI automation, our new daily updates. So,
I've actually created new guides on bots, on ChatGPT voice, Claude Opus 5, and run it for free, Communicate 3. And then, we've also got the full agent to get it from us. So, we update this regularly with a video tutorial full
install with any of your AI agents. Inside the calendar, you can jump on get all the support in real time. And then, inside the map, you can meet people in your local area who are building out with AI agents like you.
comments description, or just go to the AI Profit Portal.com. Thanks for AI Profit Portal.com. Thanks for watching.