Hermes Agent V0.18 Just Changed AI Agents Forever!

**Julian Goldie** (0:00)
So we have a brand new update from Hermes today, Hermes Agent v0.18, The Judgment Release. And we're gonna run through exactly what it means, how it works, show you examples of how it actually can be used in practice, and some of the systems we've built around this. So this is the new update that just dropped a few hours ago. And if you're wondering, okay, what does this mean? So essentially, we're gonna get straight into this. There's a bunch of new updates. The first one that's a big highlight here is Mixture of Agents. And this is a model that you can pick. So this is a new system, right? You see Mixture of Agents is now first class model you can pick. Bear in mind like Mixture of Agents, not really a model, but they describe it as that it's more like a system. So let me show you exactly what that means and how that works in practice. What you can see over here is that we have the Mixture section inside our Agent OS. And so Mixture of Agents is basically a way of getting better quality outputs by combining multiple models together. So for example, here we have Claude Opus 4.8 and Gpt 5.5 working together. Now, if we give the panel a prompt, what's going to happen is it will fuse the answers of two different models together to create a better output. And the idea here is like two minds are better than one. So you can see some of the stuff that we've created directly here. So we created like our own OS with this system. We created like a Minecraft alternative. We have generated all sorts of cool stuff, like even games and that sort of thing. And it was pretty good. Like it's doing pretty well. I also show you on the benchmarks how this performs. So if we go to Goldie Bench where I personally show all the tests that I've run and how it performs side by side.
You can see here that Hermes Mixer of Agents is right at the top. We're actually testing out Fable 5 today. But from what I've seen so far from Fable 5, I would say that actually Hermes Mixer of Agents outperforms it in many ways. You can see Fable 5 down here compared to everything that we've built with Hermes Mixer of Agents. The other thing to note here is that this is basically one shot because you don't go back and forth with Hermes Mixer of Agents inside a chat. You would just use it, combine all the models. It would take about 10 or 15 minutes to get an answer back, to get something decent, and then it's going to go into your workspace. So it's not like, for example, Claude CLI where you go back and forth because you have to wait so long for these outputs to come back. But once they come back, they look super nice. So that is one of the biggest updates and it's really good. It's kind of like an alternative to Fusion from OpenRouter. But if you're using Hermes anyway, you might as well just stick to using the Mixer of Agents instead of using theirs. But we've built in Fusion, Sucana Fugu, which is another system for this, and then Hermes Mixer of Agents directly into our agent operating system.
Next up, you can see every model's reasoning. Once the answers come back, not really too interested in that, to be honest with you. I don't think that's so exciting. You also got a system where it verifies its own work.
This is something called forward slash goal, but they've improved the forward slash goal system. Now, we've got that built into our agent OS as well. So we have goal mode over here. What that essentially means is kind of like a loop system. So you don't prompt the agents, you give the Hermes agent a task, and then it loops around, a separate judge would judge the quality of the outputs, and then it loops round and round until it's fully completed. This is something we also built in with the loop engine over here. So what we can do is define our definition of done. We have a builder API over here, we have a judge API over here. We can set how many rounds that iterates for. What that means is if the work from the builder is not good enough, because it's been judged by the judge as low quality, it will loop round and round for as many rounds as you allow it. And you set the rounds here and you can decide, okay, when does this stop and how many times you want this to go round. And it's autonomous because it's running without you. It doesn't require you to check the work or anything like that. It's just a loop system that goes round and round without you, which is pretty nice and it saves a lot of time, especially on quality control and that sort of thing. Now, one thing I will say here is if you're just using Hermes inside the terminal, it's not going to look like this. So if you're using the terminal here, this is why I don't use it anymore, it's just very difficult to see everything you've created. It's very difficult to segment everything inside Hermes. And so what I prefer to use personally is an agent operating system, where if I want to chat with Hermes with different profiles, I've got that over here. If I want to, for example, use a Hermes Oracle to get the latest news, I could do that over here. If you want to use Mixture Agents, no problem, we've got that here. If you want to do lead generation and outreach, no problem. And it's like you create a separate SaaS tool or mini app for each part of Hermes. We've got the full setup inside the AeroProfile Boardroom. But if you wanted to get the most out of Hermes, I would say that's much more powerful than just relying on the terminal where you can't really see anything that you've done previously. Next up, we have forward slash learn. This is pretty cool. So what forward slash learn means is basically you can take a guide and you can go into Hermes and say, okay, learn this system or learn this guide. So let me show you an example of that in practice. And I just want to show you examples of every part of this in practice. You can actually see how it works in reality.