01What Is an Agent Operating System?

An agent operating system is a central platform where you can manage all your AI agents in one place. Think of it as a mission control for your AI tools. You set your goals, keep a journal, and watch your agents work online, all while they share a memory system.

In this episode, Julian Goldie answers community questions about this system. He shows how you can set it up, connect different agents like Hermes and Codex, and use free APIs to make it work. The goal is to have one system that controls everything your AI agents do.

02Key Features of the Agent Operating System

The agent operating system has several useful features that make it a powerful tool for managing AI work. Here are the main ones Julian discusses:

  • Centralized control: You can see all your agents working online in one dashboard.
  • Goal setting: Put your goals into the system so agents know what to work on.
  • Journaling: Keep notes about what you are doing and what your agents are doing.
  • Memory system: The system remembers past work and information, so agents can build on previous results.
  • Multi-device access: You can access the OS from multiple Macs or other computers.
  • Integration with tools: It works with tools like Google Drive, VPS, OmniRoot, and ChatGPT Work.

These features make it easier to manage complex AI projects and keep everything organized.

03How to Set Up Your Agent Operating System

Setting up an agent operating system can be done in two main ways, depending on whether you want to use cloud storage or a virtual private server (VPS). Here are the steps Julian explains:

  1. Option 1: Use Google Workspace API. Get your Google Workspace API key. Give it to your agents so they can write files to Google Drive. Share the same access token across all your Macs and OS instances, so every agent saves to the same place.
  2. Option 2: Set up a VPS. Rent a VPS from a provider like Hostinger. Install your agent OS directly on the VPS. Save files locally on the VPS, and access it from anywhere. Use Cloudflare to help set it up securely.
  3. For either option: Once you have your storage set up, connect your agents (like Hermes or Codex) to the system. If you use OmniRoot as a free API, give the agent the OmniRoot details and GitHub link so it can test and fix any configuration issues.

Both methods let you run your agents from multiple devices. The VPS method is good if you want everything in one place and don't want to rely on cloud drives.

04How to Use the Agent Operating System with Examples

Using an agent operating system is straightforward once it is set up. Julian shows several ways to use it in the episode.

First, you can use it with ChatGPT Work, which is the rebranded Codex. This tool has ChatGPT built in, so you can chat or use the Codex features to build apps and websites. You can even create schedules tasks.

To use it, just open the projects panel on the left to build, or use chat to talk directly.

Second, you can connect free APIs like OmniRoot. OmniRoot is a free API for coding across about 90 different models. To use it, go into your Hermes agent project and ask it to create a separate profile that uses OmniRoot.

Then test that it works and iterate until it is configured properly inside the agent OS.

Third, you can create AI avatar videos. Use the same system Julian's team uses to make training courses or one-off videos. You can add b-roll and set up a voice for the avatar.

This is useful for team training or reaching people on social media.

05Results and Tests from the Episode

Julian shares some results from using the agent operating system and related tools. He notes that AI avatar content is very effective. On his Twitter account, he posted only AI avatar content and got 550,000 impressions in one day.

This shows how powerful AI avatars can be for reaching a large audience.

He also mentions that many people prefer AI avatar content over regular content, which is surprising but true. This can be a key reason to use AI avatars in your own strategy.

For the agent OS itself, Julian does not give specific numbers, but he talks about community members who have successfully built AI employee platforms and learning platforms using these tools. These examples show that the system works in real projects.

06Tips, Limits, and Mistakes to Avoid

There are several important tips and limits to keep in mind when using an agent operating system. Julian offers these pieces of advice:

  • Use the right storage: If you work across multiple Macs, Google Drive sharing is easy. But a VPS might be more secure for local files.
  • Fix fetch failed errors: If you get a 'fetch failed' error with OmniRoot, give the agent the OmniRoot details and GitHub link. Ask it to keep testing until it works. The agent may route it incorrectly and just needs to iterate.
  • Update your tools: If you are using an older version of Codex, update to get the latest ChatGPT Work features.
  • For Windows users: OMI (a tool that listens to your microphone and watches your screen) does not run on Windows. Use an alternative GitHub project that syncs with Obsidian and uses a plugin system with the AI prompt journal.
  • Check final output: Always give the agent's work a final check after it says it is done.

These tips help avoid common setup issues and make your agent OS run smoothly.

07Who Should Use an Agent Operating System

An agent operating system is for anyone who uses AI agents regularly and wants to manage them efficiently. If you are building apps, creating content, or running a business with AI, this system can help you stay organized.

Julian mentions that community members use it to build AI employee platforms, learning platforms, and more. Even beginners can learn to use it with the training available.

The system matters because it brings all your AI work into one place. This saves time and reduces mistakes. If you are serious about using AI to get work done, having an agent operating system is a smart move.