Agent Operating Systems: Automate ANYTHING!

**Julian Goldie** (0:00)
Today, we're going to be answering some of the latest questions on the Agent Operating System. This is a powerful system where all of your agents work together. Actually, I did a bunch of new features recently. So for example, there's a new wake word system for Hermes that just dropped today. We've added that in. Also got, for example, free Claude Code with two separate free API routers plugged in, which is OmniRoute and Nine Router. And then additionally, we have a new free AI Coder set up here. So you can code over here. You can preview what you've created. You can view the workspace. You can see the history and you can actually view all your previous creations over here too. And then, for example, we have a memory system plugged in. We have, for example, Notebook LM plugged in or Gemini Notebooks. This is now called Anti-Gravity Gemini 5.6 Flash plugged in. So it's an amazing system. And today we're going to answer some of the latest questions inside our AI Profit Boardroom community, because I know if people inside the community are setting this up and using it, then I know that you probably have similar questions about these systems. So let's get straight into it and we'll start answering some of the latest questions about the Agent Operating System. So the first one we got from Stephen was about sub-agents. And it was basically like, how would you consider developing a agent for a large company who want to develop a HR agent? The agent would need to reference different stuff, use it, et cetera. It would be good if it stored the history as well. Is there a solution for this? Hope that makes sense, right?
So for this set up, I would actually recommend either using a Claude project that you can give members access to, but if you need it to be like much wider, then you can actually create a custom GPT. So a custom GPT allows you to create a custom trained AI agent that you can share with other people. And then it's also really easy to use. And that's one of the best things about a custom GPT is like, it's shareable, anyone can access it. It could be trained on all of your policies and anything else you need to add to it. And also, it's super easy for anyone else to use too. We actually have an example, as you can see right here. So you can just custom train it. You can edit the GPT. You can add custom instructions, upload knowledge files, et cetera. That's probably the easiest, simplest solution, right? If you want something more advanced, then you could probably give access to your team to Buzz or something like that. We've got a new tutorial inside the classroom of the AI Profit Boarding, but if you want something super simple, I would say with custom GPT, usually the simplest solution is the best. If you're wondering how to create a custom GPT, we actually have a full tutorial.
If you're wondering how to create a custom GPT, it's right there, but that's the way that I would approach it. Great question. Another question we got here was from Sue about Pi. Now, Pi is pretty cool. It's like an AI agent that you can custom train, similar to, for example, OpenClaw or Hermes. In fact, OpenClaw was actually built on Pi, so it's a pretty cool setup. We've actually got a full tutorial on Pi if you want to learn how to use it. And we've also got a tutorial on how to use Pi to build your own OpenClaw, which I'll link to too. But yeah, it's a pretty powerful setup. It just takes a lot more work to set up from the ground running. Whereas, for example, if you use something like Hermes, Hermes is a lot simpler and much more ready out of the box. But this as well looks like you could build your own custom UI or you could integrate it.
If there's enough demand, we might actually add it to the Agent OS. I just don't know many people that actually use Pi compared to Hermes or something like that. But if you want to see me build it, feel free to direct message me inside the iProffort Boarding. Another great tip from Danish here about using Goal. He's given some examples here. So this is like a really nice in-depth tutorial. Thank you so much Danish for posting this. I've actually tested Goal mode with Codex and also with Hermes. We have forward slash Goal built into the system. So for example, in Hermes here, you can see we've got Goal mode. And then also over in Codex on the Agent OS, we have Goal mode as well. It's a really powerful feature. It can also code out for days, which is super exciting. We've got another question from John, which is, did Anthropic remove Fable from Claude Code? They actually didn't. So you can see, for example, inside the Claude Desktop, here we have Fable 5