01What is an Agent Operating System?

An agent operating system is a central control panel for your AI agents. It replaces a plain terminal with a visual dashboard where you can manage, organize, and run agents like Hermes, OpenClaw, or Open Jarvis. In this episode, Julian Goldie shows how to build your own agent operating system using AI like Claude to do the heavy lifting.

The key idea is that you don't need to be technical. You talk to Claude like a team member, and it writes the code and sets everything up. The result is a custom tool that looks like a SaaS product, but it's local and made just for you.

02Main Features of a Custom Agent OS

Julian highlights several features that make an agent operating system better than a terminal. The interface includes a Kanban board, workspace tabs, and a side panel. You can build teams of agents, see everything in one place, and navigate easily.

  • Kanban board: Organize agents visually and see what you've built.
  • Workspace tabs: Keep different projects and workflows separate.
  • Skills sharing: Agents like Hermes have 150 skills; OpenClaw has thousands. You can share skills across agents on the same layer.
  • Memory integration: Plug in Obsidian memory or other tools.
  • Local model support: Run free local models like Ollama or add APIs.
  • Custom styling: Claude remembers your preferences for colors, icons, and UI layout.

03How to Set Up Your First Agent OS

Julian walks through the process using Claude (Opus 4.8) and Open Jarvis as an example. The setup is done locally, but you can also use GPS if you prefer. The key is to build in the same place where you store your agents.

  1. Find an agent on GitHub or a free model that you like.
  2. Copy the information from the GitHub repository or the model's page.
  3. Open Claude and paste the information, asking if it can be built into your agent OS.
  4. Let Claude propose a recipe, including any downsides or friction.
  5. Order Claude to build it—say, "Great, set that up for me."
  6. Turn on bypass permissions in Claude so it can run freely.
  7. Wait a few minutes; Claude installs the CLI and any needed local models.
  8. Check the result in your agent OS. If it's missing, paste a screenshot and troubleshoot.

04Using Your Agent Operating System

Once set up, you can control everything from the dashboard. For example, with Hermes, you no longer need to stare at a terminal. You see it on the Kanban board, can assign tasks, and monitor progress.

For Open Jarvis, you can chat with it directly in the side panel.

Julian shows how to add new tools quickly. He recently added a free model release from 3.7 flash with Hermes. The system automatically added it to the agent grid.

You can also remove agents by saying "delete Open Jarvis" and restore the previous session.

  • Keep a pinned conversation in Claude with all your past context. That way, you don't have to re-explain anything.
  • Run agents in the background while you do other tasks.
  • Add workflows for ads, SEO, video, or any idea.

05Results and Comparisons in the Episode

Julian shares that a developer might take weeks to build this, but with Claude, it took about six minutes to install Open Jarvis fully. He emphasizes that the system is faster and smoother than using a terminal.

He also compares using Claude directly versus using an agent OS. With Claude alone, you can't run custom agents together or control the UI. The agent OS gives you a mission control that constantly improves.

He tests every build; sometimes Claude says it's done, but it's not actually there. So he verifies himself and troubleshoots with screenshots if needed.

06Tips, Limits, and Mistakes to Avoid

Julian warns not to add too much. You don't need every agent or workflow. Even one agent on the OS is powerful because it's better than a terminal and customized to your needs.

He also stresses that you should always test the agent yourself. Don't just rely on the AI's claim. If something breaks, paste a screenshot into Claude and explain the error.

He advises turning on bypass permissions for smoother builds, but you can always roll back changes.

  • Keep it simple: Start with one agent and one workflow.
  • Verify the build: Look at the side panel or grid yourself.
  • Use a pinned conversation to save context and time.
  • Remember the system never falls behind: New agents can be plugged in within minutes.

07Who Should Build an Agent OS?

This is for anyone who wants to automate tasks without coding. Julian says 99% of people don't realize this is possible. You just need an idea; Claude does the coding.

It matters because you turn your AI agents into a powerful system that evolves daily. You become an operator, not just a user. Whether you're into SEO, video, or any field, an agent OS gives you mission control and a fun way to manage your AI tools.