01What is an agent operating system and how it works
An agent operating system is a custom control panel that lets you run and manage all your AI agents in one place. Instead of using a boring terminal, you get a visual mission control where you can see, organize, and build agents easily.
In this episode, Julian Goldie shows how he built his own agent OS using Claude, a local setup, and a few AI agents like Hermes and Open Jarvis. The system runs on your computer, so you don't need to be technical. You just need an idea and a conversation with Claude.
Claude writes the code for you, and the system becomes your personal command center.
The agent OS works by integrating different agents, each with its own skills, into a single interface. You can add workflows for SEO, video, text-to-speech, images, and more. It's like a control room where you can monitor everything you create and automate almost anything.
02Main features of an agent OS
An agent operating system comes with several useful features that make managing AI agents easier. Here are the highlights Julian demonstrates:
- Mission control dashboard – A visual grid showing all your agents, like Open Jarvis and Hermes, in one place.
- Team building – You can create teams of agents for different tasks, making it easier to organize and navigate.
- Custom workflows – Add tools like SEO, video, studio, NotebookLM, and Kanban boards to your OS.
- Rollback capability – If you don't like a change, you can delete the agent and restore the previous version.
- Memory integration – Plug in memory systems like Obsidian to give agents shared context.
- Cross-agent skills – Share skills across agents; for example, Hermes has 150 skills, OpenClaw has thousands.
The system also allows you to run local models, so you can use free tools like Ollama without needing an API. You can customize colors, add new workflows, and make it truly your own.
03How to set up your agent operating system
Setting up an agent operating system is simple and takes just minutes. Here are the steps Julian shows in the episode:
- Find an AI agent you like from GitHub or a new model release. For example, Open Jarvis is an open-source project with about 1,600 stars.
- Open Claude (or another local AI agent like Hermes) and paste the GitHub information or describe the agent.
- Ask Claude, 'Can we build this into my agent OS?' Claude will say yes or no and give you a recipe.
- Turn on bypass permissions in Claude so it can work freely without asking for approval each time.
- Tell Claude to build it. Let it run in the background—grab a coffee while it works.
- Check that the agent actually appears in your mission control. Don't just trust Claude's word.
- If there's an issue, paste a screenshot into Claude and troubleshoot back and forth.
In the episode, Julian uses Opus 4.8 for this process, noting that Opus 4.7 also works well. The system is built locally, but you can also use GPS platforms if you prefer.
04How to use it with examples from the episode
Using your agent operating system is all about talking to Claude like a team member. Julian demonstrates this with Open Jarvis. He asks Claude if they can build it into the system, and Claude responds with a plan.
Then he orders it to build, and Claude installs the CLI and sets everything up.
An example shows adding a new agent like Step 3.7 Flash, a recent free model with Hermes. You just tell Claude to plug it in, and it automatically adds it to your mission control grid. The whole process took about six minutes.
You can also use the system to generate videos, text-to-speech, and images with your favorite AI agents. Julian mentions having SEO tools, a video tool, a studio, NotebookLM, and Kanban inside his OS. To add a new workflow, like generating ads, you just ask Claude to build it.
The system remembers your style and makes new additions fit naturally, including branding and icons.
05Results and tests shown in the episode
Julian shows real results from building his agent OS. In the episode, he sets up Open Jarvis from scratch. Claude installed the Open Jarvis CLI and local models, and within about six minutes, the agent was ready to chat in the mission control.
He also tests the system by checking if the agent actually appears. Claude said Open Jarvis was showing in the side panel, but it wasn't there. Julian takes a screenshot, pastes it back into Claude, and troubleshoots until it works.
This proves you should always verify results yourself.
Julian compares the speed to human developers. A task that might take developers a few weeks, he says, can be done in minutes with Claude. He also demonstrates how you can roll back changes, deleting Open Jarvis and restoring the previous version if you don't like it.
06Tips, limits, and mistakes to avoid
Julian shares several important tips and warnings for building your agent OS. First, don't add too much. You don't need every agent or feature.
Even with just one agent, the system is powerful because it's better than a terminal and customized to you.
Another key point: always test your agents. Don't rely on the AI's word. Julian says, 'You've got to test this stuff out as well.
Don't just rely on the AI to tell you.' Paste screenshots into Claude to troubleshoot errors.
Keep a pinned conversation in Claude with all the context about your agent OS. That way, you can pick up where you left off without explaining everything again. Also, use bypass permissions carefully—you can always roll back changes if something goes wrong.
Finally, the system stays the same even as new models come out. You won't fall behind because you can plug in new agents in minutes. The loop method—find, ask, build, prove—keeps your OS evolving.
07Who should build an agent OS and why it matters
An agent operating system is for anyone who uses AI agents and wants more control. Julian says you don't need to know how to code—you just need an idea. This means content creators, marketers, and business owners can all benefit.
The system matters because it saves time and makes AI work more efficiently. Developers might take weeks to build something, but you can do it in minutes with Claude. It also keeps you ahead because you can integrate new agents as they release, like Step 3.7 Flash or Open Jarvis.
Julian emphasizes that 99% of people don't realize this is possible. By building your own agent OS, you become an operator, not just a user. You can see everything you've created and automate anything you want, making your workflow faster and more fun.
