Hermes Agent 2.0 is INSANE! artwork

Hermes Agent 2.0 is INSANE!

AI News Today | Julian Goldie Podcast

July 6, 2026

Hermes Agent v0.18 “Judgment Release” Explained: Mixture of Agents, /learn, /journey & Background Fan-OutHermes Agent released v0.
Speakers: Julian Goldie
**Julian Goldie** (0:00)
Hermes Agent just dropped their v0.18 update, it's called the Judgment Release. And today, I'm gonna walk you through exactly how it works, what it means, and what this update is about, so that you can get the most out of your Hermes Agent, and also how to plug it into an agent operating system, so you can build and automate anything else. You haven't seen the update, this is the update post. And so, some of the highlights of this are the mixture of agents set up. So what this means essentially, is that you can now combine multiple different models together to build and automate anything. So for example, all of these outputs you're seeing right here are WebOS, a Voxelcraft kind of Minecraft style game, this crazy wormhole thing.
Like all of these builds were created using Hermes mixture of agents. The interesting thing about this is that on one shot prompts, so not like multiple back and forths, on one shot prompts, it actually got better outputs for me than when I used Fable 5 Now that is one of the downfalls of this is like you can't go back and forth with it. It's just kind of like one shot and then you get an answer about 10 minutes later. But it can build pretty amazing things and it can outperform Fable 5 on the first generation. However, you can go back and forth with Fable 5 a lot more and a lot easier, which means that you tend to get better outputs. However, at the same time, Fable 5 is going to be removed from subscription on the 7th of July. So this is actually a really good alternative if you're thinking about using multiple models together. Now, next up, what we have here is the forward slash learn feature. So this means essentially you can teach your Hermes Agent almost anything just based on a link. So if we say, for example, forward slash learn, it will actually read the page, understand it, and then build that to a skill. So you can see an example right here, so forward slash learn gave it the link to our tutorial that we wanted to learn. It's like learned it and saved it for future sessions. So it breaks down what it is, the six pillars of what it learned, the main surfaces, the setup path and how to build it. And then also you can see here, it's actually logged it to our Obsidian vault. So this actually links to our memory when it's learning skills, which means that it breaks down the skill from that guide that we want to teach it. And that could be about anything, right? You could be like, okay, learn this guide about lead generation. So you can turn Hermes into lead generation agent. You can be like, right, learn how to edit videos and you give it a link to a video editing guide. And then once it's done that, you can save it as a skill so you can reload that skill in the future. So you can see here, it says AIPB forward slash agent operating system, load with skill agent operating system. So it's basically learned the skill of how to build and automate agent operating systems from this particular gut, which is pretty insane when you think about it.
Super useful. And it works pretty fast in just like a couple of minutes, you can teach Hermes a new trick in like one single click. Now we also have forward slash journey and forward slash journey basically recaps and as a timeline of everything that Hermes has learned about you. So here's an example forward slash journey and it breaks down the journey so far, what it's done, what it's worked on, et cetera. You know, so for example, what we did, when we do it, et cetera. And then also the good thing about this is that you can see the whole timeline of what it's learned in terms of memories, in terms of skills, in terms of things that's accumulated over time. And you can edit or delete that full journey right there. So it's kind of like a digital journal that your agent is keeping about you. And it can recap, you can read through it, you can see, okay, what has it learned so far? And what's that background fan app? Now, what this means is there's a new tool called Delegate Task with Hermes Agent, where it can fan out multiple sub-agents that run in the background. So you can fire off a pile of work and just keep going.
So essentially, with the fan out option, several sub-agents can run in the background, so your chat isn't blocked. So if you say, okay, like go off and do this task, it can carry on talking to you, you can carry on working with Hermes Agent, but in the background, it's going to fan out a task that you give it to multiple sub-agents. So let's say, for example, you give it a task of like creating a website. Well, it's going to have one agent that does the research, one agent that might do some auditing on what the website looks like so far. One agent that actually creates a draft, and one agent that gathers all of the Haitian mouth together and puts it together, so you get one result.

4 more minutes of transcript below

Feed this to your agent

Try it now — copy, paste, done:

curl -H "x-api-key: pt_demo" \
  https://spoken.md/transcripts/1000651996090

Works with Claude, ChatGPT, Cursor, and any agent that makes HTTP calls.

From $0.10 per transcript. No subscription. Credits never expire.

Using your own key:

curl -H "x-api-key: YOUR_KEY" \
  https://spoken.md/transcripts/1000775580805