01What is an Agent Operating System and How Does It Work?

An agent operating system is a system that keeps all your AI agents in one place. It lets them work together, share memory, and save your workflows. Instead of jumping between tools, you have one tab with everything ready.

In this episode, Julian Goldie explained how his agent OS includes PaperClear, AI Agent Mastermind group chat, Applied Proline, Agent Kanban, a local engine, a loop engineering system, an SEO section, and a video agent. All agents plug into the same system, so they are synced and share context.

Old way vs new way: before, you juggled tabs and copied between ChatGPT, Claude, and others. You lost track of stuff and started from scratch. New way: one tab, all agents, saved workflows, and a memory galaxy that builds memories every few minutes.

You never have to re-explain yourself again. Julian said he has not had that problem for months.

02Main Features of an Agent Operating System

The agent OS has many features that make managing AI agents easier. Here are the main ones from the episode:

  • PaperClear – orchestrate all your agents together.
  • AI Agent Mastermind group chat – agents discuss ideas together.
  • Applied Proline – go from idea to implementation in one click.
  • Agent Kanban – local board to track tasks.
  • Memory galaxy – stores every memory and context automatically.
  • Video agent – creates videos, images, and voice.
  • SEO section – automates SEO workflows.
  • Music agent and game studio – extra tools.
  • Notebook lemon – research agent that generates reports, infographics, and podcasts.

You can customize everything. Ask your agent to remove sections you do not use. That keeps it simple and matches your needs.

03How to Set Up an Agent Operating System

Setting up an agent OS is not technical. In the episode, Jose built his entire system in one day and even translated it to Spanish. Here is a simple step-by-step based on what Julian showed:

  1. Start with a base agent OS template (like the one in the AI Profit Boardroom).
  2. Install the zip file that contains the system files.
  3. Plug in your preferred CLIs, such as Claude and Hermes.
  4. Open the system in one tab. Everything is already there.
  5. Customize it – ask your agent to remove sections you do not need.
  6. Use PaperClear to connect all agents in a group chat.
  7. Set up your workflows for tasks you do often, like SEO or video.

Julian recommends starting simple. Even just Claude and Hermes together is powerful. The best systems are usually the simplest.

04How to Use an Agent Operating System in Daily Work

Once your agent OS is running, you can do many things. Here is what Julian showed in the episode:

  • Use the group chat to brainstorm with all agents. Good ideas go to the pipeline.
  • Approve or deny plans in one click. Then it builds automatically.
  • Run scheduled tasks that work while you sleep.
  • Use the memory galaxy so you never repeat yourself.
  • Talk to a voice agent – with Hermes Jarvis or via 11 Labs.

For example, you can set up a voice agent by going to Hermes and saying, "I want to use Grok or Minimax. Can you set up the voice?" Or use 11 Labs to call your agent on the phone. Julian did that for fun but shut it down for safety.

You can also run a computer use agent that automates tasks on a schedule, like publishing to Substack. Julian shared a skill MD file to customize.

05Results, Tests, and Comparisons from the Episode

Julian tests all setups daily with his "Goldie bench." He benchmarks live demos, sees which models perform best, and gives medals. You can compare outputs side by side for each model and play with examples yourself.

Local models: Julian tested Mac Studio and Mac Mini and was not impressed. But an RTX 5090 with 32GB VRAM blew him away. Daniel from the community used that setup with 128GB RAM and 4TB SSD to generate an AI avatar video completely locally.

The quality looked like HeyGen, but it was local.

For token reduction, Michael used PaperClip. Julian gave three ways: free models like Hermes, free APIs on OpenRouter (e.g., N2 free until June 22), and local models. Combining these works too.

Julian also showed Substack growth from zero using a computer use agent. It is not mind-blowingly good yet, but it grows.

06Tips, Limits, and Mistakes to Avoid with Agent OS

Here are key tips and warnings from the episode:

  • Don't overcomplicate. Focus on one thing at a time. Understand basics first, then add new stuff.
  • You don't need every CLI. Pick what works for you. Julian uses Claude and Hermes mainly.
  • Open source tools like Hermes are not as smooth as Claude. Expect to document problems and fixes. That helps you improve.
  • Local models are not great on Mac. Use RTX or NVIDIA if you need local power.
  • Voice agents can be unsafe. Julian shut down his 11 Labs phone agent because he didn't want strangers calling it.
  • Keep it light. If you don't use local models, the OS runs fine on a MacBook Pro.

Follow the focus protocol to avoid overwhelm. Implement one thing, then move on. The more you simplify, the more you achieve.

07Who Should Use an Agent Operating System and Why It Matters

An agent operating system is for anyone who uses AI daily, especially entrepreneurs and marketers. If you spend hours switching between ChatGPT, Claude, and other tools, this saves time. It matters because new models and agents come out constantly.

You need a system to manage them all, not chase every new tool.

Julian built his own system and shares it with the community. He says non-technical people can set it up easily. In the episode, Jose did it in one day.

The system is updated daily, and you get video tutorials for each update.

If you want to run it on a VPS, John did that with Hostinger and Cloudflare, giving mobile and desktop access. Julian runs everything locally on a Mac Studio with 36GB memory. It is not great for local models, but it runs the OS 24/7.

In short, the agent OS helps you automate what you spend time on, keep everything in one place, and stay focused. That is why it matters.