01What Is an Agent Operating System?

An agent operating system is a central hub that connects multiple AI agents, tools, and memories so they can work together on projects. Instead of using one chatbot at a time, you get a whole team of AI workers in one place.

You can assign different agents to write, edit, and judge a video, for example. The system also lets you plug in new models whenever they come out. So you build it once, and it keeps getting better.

It works by giving each agent a specific role. They share context through a common memory, like an Obsidian vault. That way, you don't have to rebrief them every time.

02Main Features of an Agent OS

One of the best features is flexibility. You can add or remove models and agents without breaking your setup. For example, when GLM 5.2 came out, it was plugged in right away.

  • Manage multiple agents in one group chat, including Claude and Hermes.
  • Create separate profiles for each model, like GLM 5.2 or Kimi K2.7.
  • Make the UI visual with custom images and themes.
  • Use a Kanban board to track tasks, like a video agent iterating until it's approved.
  • Add skills like Hyperframes for video creation or Headroom to save tokens.
  • Control everything with voice via Hermes Jarvis.

You can also build your own versions, like an app builder that turns an idea into an app with just an approve or reject step.

03How to Set Up Your Own Agent Operating System

Setting up an agent OS is easier if you follow a step-by-step process. Start simple and expand as you learn.

  1. Get the agent OS zip file from the AI Profit Boardroom (or a similar source).
  2. Install it on your computer or a VPS using the installation guide.
  3. Open the setup and connect your AI agents, like Claude or Hermes.
  4. Add your free or paid API keys, such as from OpenRouter or coding plans.
  5. Create an Obsidian vault and use an MCP to plug it into your agents.
  6. Customize the UI and add your own images or themes.
  7. Set up one workflow at a time, like lead generation or video creation.

If you get stuck, ask Claude to help you fix or improve a part of the system.

04How to Use an Agent Operating System: Examples

In this episode, Julian shows several ways to use an agent OS. For video creation, he gives agents access to Hyperframes as a skill. Then he sets up a Kanban board with a video writer, editor, and judge.

The writer creates the script, the editor makes the video, and the judge checks if it's good enough. They keep iterating until the judge approves. This produced a fully edited video with a script, AI voice, and camera angles—no manual editing.

For SEO, there's an automation pipeline. For games, you can plug in models like Kimi K2.7 to build playable 3D games.

You can also use an app builder that takes a text idea, shows you a plan, and then builds the app when you approve. The system logs everything to Obsidian, so all agents share the same memory.

05Free APIs and Models to Keep Costs Low

One big question is how to manage costs. Julian recommends three ways to keep it cheap.

  • Use coding plans from Kimi, Minimax, or GLM. These plans let you use the API without extra charges.
  • Find free APIs on OpenRouter. Search for 'free' and you'll see many options like NEX2.
  • Use plugins like Headroom to reduce token usage, or Mark It Down to convert files to Markdown and lower input size.

For free models, log in to Nous Portal with OAuth. Then you can use Step 3.7 Flash and Nemetron 3 Ultra for free. Or run local models with an NVIDIA Spark.

Keep your model settings updated in the dashboard to switch to these free options.

06Organizing Your Obsidian Vault for Memory

Obsidian acts as a second brain for your AI agents. They take notes from every conversation and store them as Markdown files.

To organize it, have agents review documents and put notes into your Obsidian folder. This makes it accessible offline and searchable.

You can also create custom skills to help organize. Ask Claude to build one for you.

Using an MCP, you can connect your Obsidian vault to agents on a VPS. Then all your agents share the same memory and context, so you don't have to repeat information. This saves time and tokens.

07Safety and Limits: Keeping Your Agents Safe

When your agents have access to do things, you need limits. Julian suggests giving your agent its own inbox instead of your main email.

Use guardrails built into Hermes and Claude, plus set extra rules and permissions. This stops agents from doing anything unexpected.

Test and update your system regularly. If something breaks, like Hermes Jarvis, ask Claude to fix it by reviewing logs and adding improvements. Keep the agent OS version up to date because updates come daily.

08Who Should Build an Agent Operating System?

Anyone who wants to get more from AI should consider an agent OS. It's for both technical and non-technical users. Julian says he's not a coder, and many others have built amazing things too.

If you plan to create content, automate SEO, build games, or manage projects, this system helps you do it faster. It also future-proofs you because you can swap in new models as they appear.

If you're new, start small and focus on one workflow each week. You'll feel unstoppable as you build and customize your own system.