01What Is an Agent OS and How Does It Work?
Build agent OS means creating a single dashboard where all your AI agents live and work together. You can also create custom workflows that automate tasks like video creation, SEO content, or lead generation. Everything is one click away, and you can operate it just the way you want.
Think of it as your own control center. Instead of jumping between ChatGPT, Claude, and other tools, you put them all in one place. Then you add your own workflows, like a video agent or an SEO pipeline.
When you click a button, the system does the work for you.
In this episode, Julian Goldie shows exactly how he built his own agent OS, called Hermes Agent OS. He explains the process step by step, so you can do the same. The best part?
You can build it for free using subscriptions you already have.
02Main Features and New Additions
An agent OS can have many features, depending on what you need. Julian shows several key parts that make his system powerful:
- Mission Control Dashboard: This is the main view where you see all your agents and workflows.
- Custom Workflows: You can build specific tools, like an image generator, video agent, or SEO content pipeline.
- Memory System: Using Obsidian, you can store notes and memories that all agents can access.
- Group Chat: Let your AI agents talk to each other, bounce ideas, and collaborate.
- Pipeline: Move ideas from creation to implementation, with human approval steps.
- Gallery: Save and view everything you create, so nothing gets lost.
You can also add features like a Kanban board, voice-activated agents, and even local models for free. The system is flexible, so you can hide or delete anything you don't need.
One of the newest ideas is to add plugins from tools like Codex into your agent OS. That way, you can send emails or manage tasks without leaving the dashboard.
03How to Get Started Building Your Agent OS
You can start building your own agent OS today. Follow these steps to set it up, based on what Julian showed:
- Choose an AI agent to help you build, like ChatGPT or Claude. Julian uses ChatGPT for visual tasks.
- Open a new chat and explain what you want. For example, say: "Create a mission control dashboard where I can operate ChatGPT, Codex, and Hermes agent."
- If you have a design style in mind, take a screenshot and attach it. Say: "Use this style for the dashboard."
- Decide if you want to start from scratch or use a template. A template from a source like the AirProfit Boardroom is faster.
- If using a template, upload the zip file and say: "Install and configure this agent OS on a local port."
- Let the AI ask questions. Answer them, or say "Figure it out" to let the AI make decisions.
- Wait for the installation to finish. Then test it to make sure everything works.
You can run multiple builds at the same time by starting new chats in the same project. This lets you add features like an image generator or a voice agent while the main dashboard is still being built.
04How to Use Your Agent OS: Practical Examples
Once your agent OS is ready, you can use it to automate almost anything. Julian shows several examples:
- Video Agent: Create videos in one click, without spending hours editing.
- SEO Content Pipeline: Research keywords, generate content, and deploy it to your website automatically.
- Lead Generation: Generate leads, write email campaigns, send emails, and track replies all from one dashboard.
- Image Generator: Ask for an image like "a cat eating biscuits" and get it generated with custom settings.
- News Oracle: Pull the latest trending news and get topic ideas for content.
To use these, you just click on the workflow you want. For example, in the image studio, you type a prompt, choose an aspect ratio, and hit generate. The system does the rest.
You can also use a memory system. When you ask a question, the agent will look at your Obsidian vault for context. That way, it knows what you've worked on before and can give better answers.
If something breaks, like the agent not replying, you can drag a screenshot into the chat and ask the AI to fix it. That's how you keep improving your system.
05Results and Comparisons From the Episode
Julian showed that building an agent OS from scratch takes time. He used ChatGPT to build a new one, and after 10 minutes, it had only created the UI. It didn't even test the functionality, even though he asked it to make sure it worked.
In contrast, using a template got the system fully installed and configured in just 10 to 20 minutes. One of his students, Jacob, tried to build an agent OS for 12 hours a day, but he finished in 10 minutes using the template. That shows the big advantage of starting with a ready-made setup.
Julian also tested different models. He used GPT-5.6 for building and Codex for image generation. He noted that GPT-5.6 can handle almost anything you throw at it, but it might forget tasks if the project is too big.
That's why testing is so important.
He also compared running locally versus on a VPS. He recommends running locally for security, especially if your agent OS can edit files. A few members run it on a VPS, but they limit the tools to stay safe.
06Tips, Limits, and Mistakes to Avoid
There are several things to keep in mind when building and using your agent OS. Here are the key tips from Julian:
- Use one agent to build. Don't let multiple agents work on the same code at once. They might overwrite or delete each other's work.
- Test everything. After building, click around and make sure it works. If something breaks, ask the AI to fix it.
- Watch your tokens. Building many features in parallel can burn through your usage fast. You can reduce effort, speed, or switch to cheaper models.
- Keep it simple. Don't add every feature you see. Only build what you actually use. You can always add more later.
- Run locally for security. Public servers can be risky if your agent can control your computer. Use a local setup instead.
- Add a delete option. Make sure you can delete anything that doesn't work. That keeps your system clean.
You can also use free APIs like OmniRoot or HY3 to reduce token costs. Local models from Olama or LM Studio are free forever, and they're great for simple tasks.
One mistake Julian saw with other agents like OpenClaw was that people gave too much access to external devices. That's a security risk. Always limit what your agents can do.
07Who Should Build an Agent OS and Why It Matters
An agent OS is useful for anyone who spends a lot of time on repetitive tasks. If you create content, do SEO, generate leads, or manage a team, you can save hours every week. Julian says that after building his system, he can create more content, reach more people, and grow faster than before.
You don't need to be a programmer. You can build everything by telling an AI like ChatGPT what you want. No coding skills are required.
This means almost anyone can build their own tools and workflows.
Why does this matter? Because you can automate the boring parts of your work and focus on the creative parts. You can also sell AI agent services to clients.
They will be impressed by a custom system that solves their specific problems, like generating images or videos.
Julian also mentions that an agent OS can be personal. You can add a music generator, a training tracker for your hobbies, or anything else you like. It's your system, so you can make it fun and unique.
In the end, building an agent OS is about taking control of your AI tools. Instead of being limited by each app, you create your own super-app that works exactly how you want.
08How to Get the Full Agent OS Template
If you want to save time, Julian offers his full agent OS template inside the AirProfit Boardroom. You can download the zip file and install it locally. He updates it daily as he codes new features.
To get it, you need to join the AirProfit Boardroom, which includes a community where you can ask questions and get help. Members often share their own versions, like Sheena who made one for lead generation.
There are also weekly coaching calls where you can share your screen and get support. You can meet other builders and learn from their successes and mistakes.
If you prefer to build from scratch, you can follow the steps in this episode. But the template is a great starting point, especially if you want a system that works right away.