01What the Agent Operating System Is and How It Works

The agent operating system is a powerful system you can use to build and automate anything. It links all your AI agents together so they work as one team.

In this episode, Julian Goldie explains how the AgentOS uses features like a memory galaxy, Kanban boards, and a local AI agent engine. You can use it for tasks like lead generation and outreach campaigns. You can even voice control it with Hermes Jarvis.

The system keeps all your agents aware of what you are working on. It also lets you orchestrate them on visual boards. This makes it easy to manage complex workflows.

02Main Features of the AgentOS

The AgentOS includes several key features that make it useful:

  • Memory galaxy: all agents know what you are working on.
  • Kanban boards: orchestrate agents visually.
  • Local AI agent engine: build and automate anything locally.
  • Codex section: chat, goal mode, workspace, and saved sessions.
  • Hermes agent: voice control and integration with GPT 5.5.
  • Studio: generate images and videos with GPT 5.5.

You can also plug in external tools like OpenClaude or use the Codex plugin inside Claude Code. This gives you multiple ways to use your Codex subscription.

03How to Set Up the AgentOS

Setting up the AgentOS is straightforward, even if you are not technical. Julian mentions that a user named Rick set it up in about 30 minutes. Here are the general steps:

  1. Get access to the AgentOS inside the AI Profit Boardroom.
  2. Download the zip file that contains the latest version.
  3. Unzip the file and look for the update.md file.
  4. Install the features you want—you can pick and choose.
  5. Log in to your Codex account to connect it.
  6. Set up Hermes for voice control and model access.
  7. Connect your Obsidian vault if you want to use the memory feature.

Once installed, you can start customizing workflows and hiding parts you do not need.

04How to Use the AgentOS with Codex and Claude Code

You can use the AgentOS in four ways with Codex. First, plug it into the Hermes agent. Second, generate images using OpenClaude and Hermes with your Codex auth.

Third, use the Codex tab directly inside the AgentOS. Fourth, use the new Claude Code plugin that lets you bring Codex into Claude Code.

For local models, you have two options. You can change the model in the config.toml file for Codex. Or you can use Olama with GLM 5.2, which works in one click with Claude Code or Codex.

In the episode, Julian also shows how to compare models on GoldieBench. You can see side-by-side outputs and pick the best one for your tasks.

05Results and Comparisons Shown in the Episode

Julian compares GLM 5.2 to other models using GoldieBench. He says GLM 5.2 is up there with Opus 4.8 and actually outperforms it on some tests. For example, compared to Kimi K217, GLM 5.2 is miles ahead.

He also notes that open models like GLM 5.2 are overtaking frontier models. GPT 5.6 is gated, and Fable 5 was taken down after three days. GLM 5.2 remains open with published weights.

In another example, a member named Davide calls GLM 5.2 a genius. Julian agrees and says it is very comparable to Opus 4.8 in performance.

06Tips, Limits, and Mistakes to Avoid

One big tip from Julian is to focus on systems, not prompts. He says the real skill is building repeatable systems and debugging them properly when updates arrive. That way, you can swap models without impacting your workflows.

For teams, simplify the AgentOS. Hide parts you do not want them to use, and write an SOP for them. Only give them the workflows they need.

You can also use the settings section in the top left to hide features.

When using local models, stick with GLM 5.2 if you want the best option. Minimax M3 is older, and Kimi K Super 7 is not as good. Master one model to get better outputs.

For memory, skip the 'remember' skill and use Obsidian instead. Julian plugs his Obsidian vault into all his agents for better recall.

07Who the AgentOS Is For and Why It Matters

The AgentOS is for anyone who wants to build and automate AI workflows without being a technical expert. Julian says non-technical users like Rick can set it up in 30 minutes.

It matters because it saves time and helps you scale. One member named Ben says Julian's responses saved him a lot of time. The system also allows teams to share a common brain, like a shared Obsidian vault.

Whether you work alone or with a team, the AgentOS gives you a central place to connect agents, use multiple models, and automate tasks. It is designed to grow with you as new models and features drop.