Claude Watermarks + Hermes + Agent OS (AI Q&A)

Claude watermarks, Hermes and Asian OS. Today we're answering the biggest AI questions everyone is asking right now. By the end of this video, your whole setup is going to be smarter, way more powerful. First, Claude just dropped a huge update in visible watermarks and now hiding inside your content.

I'll show you exactly what keeps a mark, what removes it and why you don't need to panic. Then I'll reveal the exact models I'd run Hermes on today. A fast one, a powerful one and a free one that runs right on your computer. We'll also be talking about obsidian as well and stick with me to the end because I'll show you how your Asian OS lets you run your entire team of agents from your phone.

Let's get into it. So the first question that we had was about watermarks and basically Claude recently announced that they're going to add watermarks new models released after the 2nd of August and due to the new EU act. And Morgan asks, how does everyone feel about the recent news about Claude adding watermarks? I don't quite understand what it looks like in practice.

What if, for example, we just voice transcribing in Claude, which is a great question. So basically if you were just voice transcribing in Claude and you use Claude to generate the content, the way that I understand it, and I think there's a lot of like gray areas here that aren't explained fully, but the way that I understand it is that actually it would put a watermark on anything that's added. So for example, even if like you wrote a human article and then you ask Claude to apply edits, it would still have a watermark inside the content that means that it's got edits, which means that if there's a watermark on Claude content, that doesn't necessarily mean it's AI generated because there's going to be signals where it looks like it's AI generated from watermark, but in actuality it isn't. I also expect like there's going to be a bunch of tools that help remove the watermark.

I can see that coming in the future too. Some things to note in terms of what keeps a mark and what doesn't with these Claude watermarks. So the first one is like, if it's just copied and pasted straight out of Claude untouched, then it would keep the mark. If it was like a super long post, very easy for Claude to add the watermark and therefore it would show up much.

If for example, you had a draft and then you sent it through the grammar fixes. So if you use Claude as like a grammar fixer, I actually think that would come out as well. So it's kind of harsh. And then also if you, for example, copied and pasted Claude into a document, into a blog, into an email, still have the watermark inside.

Now, apparently according to the guidelines that Claude released on this, rewritten content heavily in your words should remove the watermark. Also, it's very hard for Claude to add a watermark to short content. So if it's just like a two line social media content post, that's not really going to have a watermark inside of yet either older models from what I saw on the post before the 2nd of August, 2026 wouldn't have this change on them. So for example, if you're using Fable 5 now, I wouldn't expect that to have the watermark inside it, but I know they are trying to figure out a way to apply it retroactively.

And then also if you had a file with the watermark inside it, that could be, for example, reformatted or converted in a way where it doesn't have the watermark as well. I think the main thing to know is like the mark system really is not a big deal. I don't feel great about it, but also I know that for example, like Gemini's had this for a couple of years anyway. And also if you're doing something like SEO, it's not a big deal because Google have already said like, it doesn't matter if it's AI generated content or not.

And also I think social media is the same. Like it doesn't matter whether it's AI or not. What matters is the quality of the content is there. The people actually like it.

So that's the angle that I tried to focus on. Robert was asking, you know, what's everyone using for their models? Just curious how everyone has the Hermes and agents set up for models and that sort of thing. So they're trying to decide what's the best way to set up Hermes and the Hermes team.

So for me personally, I really liked Quen 3.8. If you're looking for a local model, depending on your setup, you can see, for example, Michael recommends Quen 3.6, 35B. That depends if you've got like an RTX. So they've got an RTX 5060.

For me personally, I would recommend Quen 3.8 Max if you're willing to use an API. That's like a frontier model that's pretty fast. And also DeepSea V4 Flash is really, really good with Hermes, like fast, responsive, reasonable right now. They'll probably change their price in the future, but I think those are two good options.

And then for local models, the only one that I've seen that's half decent on like a normal setup, like a Mac studio, for example, is using LFM 2.5, 2.6B. That was like the fastest, most responsive and actually useful API. And then you've also got stuff like Kimi K3, but it's a bit slow right now. If you have to pick one, I would go with Quen 3.8 Max.

And for anyone watching, like if you want to change the model that you have working on Hermes, you can just type Hermes model inside your terminal or go to Hermes dashboard and you can change your model from there. Depends as well what you do, right? So if it's like a fast everyday model, you want to go for speed, DeepSeek V4 Flash is pretty good. If you want a coding model, Frontier, something like Quen 3.8 is pretty good.

For a bulk worker, free local model like LFM 2.5, 2.6B is pretty good. And then also if you want like a reviewer or a quality controller, then I'd actually go with like something that just disagrees or is a different model to your main one. And that could be, for example, DeepSeek versus GLM 5.2. It could be Kimi K3 versus Grok 4.5, whatever you prefer to use.

But overall I'd go with DeepSeek for fast, or Quen 3.8 for Frontier. By the way, if you want me to answer these sort of questions for you, create a personalized video tutorial for you, like I'm doing today, feel free to post inside the ARProfitWarm community. Link in the comments description or just go to the ARProfitWarm.com to get access and for me to create personalized video tutorials for you like this. So we've got a question from Michael and this is, how do you clean up a messy second brain?

So they say my setup, Obsidian plus Graphify connected to my agents has got messy over time, too many notes, unclear links. My agents don't always pull the right context anymore when I hand them a task. The result is inconsistent output depending on what they happen to find. Has anyone successfully cleaned up and restructured a second brain that outgrew itself without losing the history?

So this actually happened to me. One thing that I'd recommend is that you go directly into Claude's desktop and then you can ask Claude to clean up your Obsidian memory and organize it. Now one framework that I like to use when it comes to organizing Obsidian is para, P-A-R-A. And that just helps organize your notes into something a bit more tidy and a bit more beautiful.

And then also don't ask it to delete anything. Just ask it to put it inside archive. So you still got a backup if you need to come back to it. So for example, projects, areas, resources, and archive, that's a very good way para to organize your whole setup.

And then you can ask Claude to just gradually fix the links and that sort of thing. One thing that works really well as well, you can give the documentation of Obsidian and how it works into Claude and just say, based on this documentation and based on my notes that I currently have, and based on the para framework that you already understand, can you organize and filter my notes so it's just much cleaner and easier and faster for my agents to organize. And that makes everything easier. Now we've got another question from Jay, who's a legendary member of the Air Profitable Audience.

Shout out to Jay. Now one problem he's actually facing is that he's running OmniRoot and he's running it with the agent OS. If you're not sure what an agent OS is, here's an example. So it's basically like all your agents plug into one system in one place and then you can use this on tail scale, remotely with your phone or whatever, and you can have your agents working together.

Now OmniRoot is running on a local host, which means that if you run OmniRoot but you're trying to run it in the cloud, I think it'd be very difficult as far as I'm aware. Now I don't use Windows personally, so I'm probably not the best person to ask when it comes to Windows, but I did ask the agent that set it up to come up with some ideas and I've detailed that for you on this guide. So it's recommended some steps. Again, I don't use Windows, so I'm not like 100% an expert on this, but that could be something to check out and I'll leave a link for you.

Bear in mind, as far as I'm aware, OmniRoot is local. So it runs for a local host. And then if you're trying to connect that local to tail scale and use it from separate devices, you'd probably have to use OmniRoot on the VPS, if that makes sense. But I'm not 100% certain on that.

Again, WSL experts, help me out here. So that's basically it for all the questions today. If you want me to answer your questions personally with a video tutorial like you've seen today, and also what I do is I link to the video tutorial, and then I also break down the questions and the answers into a step by step like this in case you don't have time to check it out. And usually I create a guide as well so that you can just follow along as you saw earlier.

Inside the community, we've got loads of great people. There's people online 24 seven to help you whenever you need to. Inside the classroom, we get access to all of my best trainings, courses, tutorials, et cetera. Inside the calendar, you can jump on weekly coaching calls, share your screen, ask questions.

And then in the map, you can meet people locally near you who are building with AI agents like you. So feel free to get this link in the comments, description or go to the AIprofitablewoman.com. Thanks for watching.