How to Build a Self Improving Agentic OS in Claude

0:00 How to build a self-improving agentic OS in Claude. What if your AI tools could actually remember you? What if Claude knew your goals, your projects, your notes? Why are you still copy-pasting context into every new chat? Most people are using AI completely wrong. They're juggling tabs when they should be running a system. By the end of this video, you'll see exactly how to fix it. Hey, I'm the digital avatar of Julian Goldie and I help you learn and actually use AI tools in your work. Today, I'm walking you through Agent OS. It turns Claude, Hermes, and every other AI tool on your computer into one connected system that learns about you and gets smarter the more you use it. I'll show you what it is, two use cases I tested, and how to install it tonight. Stick around, there's a new update most people haven't seen. So, what is Agent OS?

0:47 Agent OS is a locally hosted dashboard that runs on your own machine. Think of it like a mission control center for every AI agent you use. Right now, most people have Claude in one tab, ChatGPT in another, and another AI tool in a third. Nothing talks to each other, so you repeat your context every new chat. Agent OS gives you one local dashboard that connects all your AI agents in one place. You get a chat panel, session history, agent status, control rooms for each agent, skills and plugins, Kanban integration, and analytics. Everything is local, so your data stays with you. Now, here's the part that flipped my brain. Every conversation across every agent gets logged into one memory layer built on Obsidian. Obsidian is a free note-taking app, but inside Agent OS, it

1:33 becomes the brain. It stores your goals, journals, and personal knowledge base. So, when you talk to Claude tomorrow, Claude already knows what you worked on today. When you ask Hermes a question, it pulls context from your actual notes and projects. The agents stop being random chatbots and start acting like a team that knows you. That's the self-improving part. The more you use it, the smarter it gets. So, why use Agent OS with Hermes and Claude together? Claude is great at thinking, planning, and writing. Hermes is an open-source agent framework by News Research, strong at running tasks. There's a third agent called Open Claude that handles execution. I call this the four-layer Goldy mission stack. Layer one is intelligence, Claude. Layer two is execution, Open Claude. Layer three is research and orchestration, Hermes.

2:19 Layer four is self, the Obsidian memory layer. When all four run together, you stop having scattered AI conversations and start having one system that compounds over time. Local first matters, too. Across five cloud platforms, you'd spend more time fighting logins than shipping work. Now, let me show you the new update inside Agent OS. The dashboard now has a per agent control room. Each agent gets its own space inside the OS. You can see what Claude is working on, what Hermes is researching, and what Open Claude is executing, all from one screen. There's a goals tracker built in. You type your goals, and your agents align their work to them. The journal lets you log what you did each day, and it feeds the memory layer. There's voice input now, so you can talk to your agents instead of typing. And the Kanban integration means you can drag tasks around

3:04 visually. Let me show you two use cases I tested. Use case one was a single HTML page for an SEO ROI calculator built for the AI Profit Boardroom. I gave Claude the prompt for a calculator with inputs for monthly traffic, conversion rate, average order value, and estimated SEO lift. Claude built it in one shot, form fields wired up, and math running cleanly. Use case two was a single sales page for an AI SEO consulting offer. I asked Claude for an HTML page with a hero section, problem section, solution section, pricing, and a clear call to action. It produced a clean single-file page with all five sections in order. Two pages both shipped fast through the same workflow. Now, let me walk through how this gets built. I built mine in one

3:50 Claude session. I opened Claude and pasted in the docs from the Hermes GitHub and the open Claude GitHub. Then I asked Claude to scaffold the dashboard in Next.js and Tailwind. Next.js is a web framework. Tailwind is a styling library. Claude handled it. After about an hour, I had a working dashboard on my Mac. It had the chat panels, goals tracker, journal, memory layer wired into Obsidian, three agents connected, and voice input through the microphone. You can ship version one tonight in a single Claude session. You don't need to be a coder. You need to copy, paste, and follow a prompt. Here's a real example of how this changes your day. I opened Agent OS yesterday and typed into Hermes, "Based on my Obsidian vault, give me some ideas on what I should automate today." Hermes pulled context from my agency notes, community notes,

4:36 and current build work. It came back with a personalized list of automations tied to projects I'm actually working on. Not generic, specific quick wins for my real work. That's the difference between using AI as a chatbot and using AI as a system. Now, real quick, if you want the Agent OS system without figuring it all out on your own, the AI Profit Boardroom is where to go. Inside the Boardroom, you get the Agent OS zip file as a free bonus, plus 100 prompts, and a 30-day roadmap built around this exact stack. You get the Hermes plus Claude OS launch kit that walks you through every step from blank screen to working dashboard. You get weekly live coaching calls where you can ask questions about your specific setup if you get stuck. And you get access to 27 other launch kits that plug into the

5:22 same system. If you're building a self-improving Agent OS, the Boardroom was built for exactly this. Okay, let's keep going. How do you install Agent OS? Install Hermes itself first. The Hermes installation guide for 2026 is linked in the description. Hermes is free, open source, and runs on Mac, Windows, or Linux. The install is a single command in your terminal. Paste it in, hit enter, and Hermes sets itself up. You'll need an account with at least one AI provider like Anthropic for Claude. Once Hermes is running, install Ollama to run local models. Ollama lets you run models like Gemma or Qwen on your own machine. After that, set up your Obsidian vault. It becomes your memory layer. Then build the dashboard by opening Claude, pasting in the Hermes and Open Claude docs, and

6:09 prompting it to scaffold in Next.js and Tailwind. Inside the AI Profit Boardroom, there's a starter zip so you don't have to build it from scratch. Now, how do you update it? Updating Agent OS is modular. Hermes updates when you pull the latest from its repo. The dashboard updates when you reprompt Claude with new features. Obsidian updates through its app updater. Your memory layer keeps growing because every conversation gets logged automatically. You don't have to train it, you just have to use it. Now, some pro tips for beginners. Tip one, start small. Get Claude in the dashboard first, add Hermes second, add Open Claude third. Tip two, write your goals into Obsidian before anything else. The memory layer only works if there's something to remember. Tip three, use the journal every day, even one line. That journal

6:55 becomes the most valuable file in your system within 30 days. Tip four, don't try to make it pretty on day one. Ship it ugly, use it daily, then ask Claude to clean it up on day seven. Tip five, treat your prompts like assets. When one works, save it to a prompts folder in Obsidian. Tip six, trust the local first model. Local first is what makes the memory layer powerful. Tip seven, don't skip the four-layer stack. Intelligence without execution, research, and self is just a chatbot. When you build your Agent OS tonight, you're going to hit a wall, a config issue, a prompt that doesn't scaffold right, or the memory layer not connecting to Obsidian. If you want the Agent OS system handed to you instead of building it from scratch, the AI Profit Boardroom is where to go. You

7:40 get the Agent OS zip file as a free bonus, the 100 prompts pack, the 30-day roadmap, the Hermes plus Claude OS launch kit, and weekly live coaching calls where you can ask questions about your exact setup. Head to AI profitboardroom.com to get inside. And if you want the full process, SOPs, and 100 plus AI use cases like this one, join the AI success lab. Links in the comments and description. You'll get all the video notes from there, plus access to our community of 58,000 members who are crushing it with AI. That's it for today. Build your Agent OS tonight. Make Claude your intelligence layer. Let Hermes do the running. Let Open Claude handle the execution. And let Obsidian be the memory that ties it all together. I'll see you in the next one.