01What an Agent OS Is and How It Works

An agent OS is a personalized web app that runs AI agents like Claude and Hermes inside custom workflows. Julian Goldie built his own agent OS from scratch, starting as a basic chat tool and adding features over time. The agent OS works like a lightweight website, so you do not need a powerful computer to run it locally.

In this episode, Julian explains that an agent OS lets you mold AI agents to fit your exact needs. Instead of using a generic desktop app, you can design workflows that save time and produce better outputs. The system uses APIs to connect agents and automate tasks like image generation or lead generation.

02Main Features of Julian Goldie’s Agent OS

Julian Goldie’s agent OS includes several built-in tools. Here are the main features he shows:

  • Custom workflows: You can build your own image generator, lead generation tool, or any tool you need. Each workflow looks and works exactly how you want.
  • Memory knowledge graph: The agent OS automatically logs all memories into Obsidian. This creates a feedback loop where agents train each other with context.
  • Multiple harness support: The system works with Claude Code, Hermes, and other harnesses. Julian prefers Claude Code for reliability and Hermes for flexibility.
  • Obsidian integration: The agent OS connects to Obsidian, a free open-source app, to store and retrieve context. This setup is called the infinite context engine.

03How to Set Up Your Own Agent OS

Julian Goldie recommends starting simple and building up. Follow these steps to get started:

  1. Choose an agent harness. Julian suggests Claude Code for reliability or Hermes for flexibility. Focus on one harness at first.
  2. Set up Obsidian on your local computer. It is free and open source. Create a vault for your memories and context.
  3. Build a basic chat interface inside your agent OS. Julian started with a simple Claude and Hermes chat.
  4. Connect the agent OS to Obsidian. Tell your agents to update Obsidian every time they work. This builds your second brain.
  5. Identify your biggest time-consuming tasks. Automate those first by creating custom workflows in the agent OS.
  6. Add more features one at a time. Julian adds features out of curiosity, creativity, or to save time.

04How to Use the Agent OS for Custom Workflows

Julian Goldie demonstrates how to use the agent OS by building custom workflows. One example is an image generator. He created a workflow inside the agent OS that calls an API and produces beautiful images.

The UI is fully customizable, so it looks exactly how he wants.

Another example is a lead generation tool. Julian built this as a SaaS-style tool inside the agent OS. He started with an idea, built the workflow, and now it runs automatically.

You can also use the agent OS to log memories. Every time an agent like Hermes does something useful, it updates Obsidian. Julian shows a memory called "Hermes Astros" that analyzes competitors and gives content ideas.

This context feeds back into other agents, improving their outputs over time.

05Results, Tests and Comparisons Shown in the Episode

Julian Goldie compares using a desktop app versus an agent OS. He says desktop apps like Hermes desktop are less useful because they are not personalized. The agent OS, on the other hand, looks and feels unique to the user.

It produces better outputs because it has custom workflows and memory context.

He also compares harnesses. Claude Code is less buggy and works every time. Hermes is more flexible but can be buggier. Deep Sea harness has a basic UI that Julian finds not fun to use.

Julian notes that running the agent OS locally avoids security issues. Problems with VPS hosting, like those seen with OpenClaw, come from people actively searching for VPS access. Local hosting keeps your data safe.

06Tips, Limits and Mistakes to Avoid

Julian Goldie shares several tips for building an agent OS. First, do not use a VPS for hosting. Local hosting is safer because attackers can target VPS systems. Even a laptop or Chromebook works since the agent OS is just a web app.

Second, avoid using multiple harnesses at once. Focus on one at a time until you master it. Julian recommends Claude Code for beginners because it is reliable.

Third, do not use desktop apps straight out of the box. They lack personalization. Always build custom workflows inside the agent OS to get the best results.

A common mistake is not setting up the memory system. Without Obsidian integration, agents lose context. Julian says the feedback loop between agents and Obsidian makes both tools more powerful.

07Who an Agent OS Is For and Why It Matters

An agent OS is for anyone who wants to automate tasks with AI agents in a personalized way. Julian Goldie built his agent OS for the AI Profit Boardroom community, but anyone can build their own. It is useful for small business owners, content creators, and developers who want to save time.

The agent OS matters because it turns generic AI tools into personalized assistants. With custom workflows and memory, agents become more useful over time. Julian says the system creates a „positive feedback loop“ where agents improve each other’s outputs.

This leads to better results without extra work.