Transcription
In this video, I'm going to be setting up the new Hermes agent on my DGX Spark and setting it up for a real-world use case for BridgeMind. This agent is exploding in popularity, and it's actually growing faster than Open Claw. And in this video, I'm going to be showing you everything that you need to know about Hermes agent, why it's growing faster than Open Claw, and why it's actually better than Open Claw. But with that being said, let's get right into it.
So, let's dive into everything that you need to know about agent. So, what actually is this agent? For. So, first of all, this is built by a company called Nous Research. It's the lab behind the Hermes model family, which is an open-source AI lab that they have models that you can install locally, even on this DGX Spark. It's fully open-source, MIT licensed, and the big key differentiator is that it's a self-improving AI agent that creates skills from its experiences. Okay? So, this is one thing that positions it ahead of Open Claw and why it's rapidly advancing in popularity. It's at 57,000 GitHub stars.
So, how this works, this self-improving loop works by it receives a task; it then is able to execute that task, evaluate the results, extract a skill, and then it basically is able to store and refine that skill. So, this is the self-improving loop that makes this so much different than Open Claw. And when you think about AI agents, this here, this loop, is really what we need in an AI agent. We don't want to have to give it all of its memory, give it all of its information. We really need to give it tasks and be able to have it improve itself without a bunch of instruction from the orchestrator.
Some of the key features that this entails is skill documents. So, these are structured markdown documents that capture when X try Y patterns from successful tasks. It also has a longer-term memory than Open Claw, tool integration, self-evaluation, and it's model agnostic, which allows us to be able to swap models without changing anything about the agent, which is similar to Open Claw, but we'll, we'll dive into that a little bit later in the video once we kind of cover the basics here. But it's, it's really different because it doesn't just execute the tasks; it learns how to do them better.
But let's now dive in and actually get this set up on the DGX Spark. So, I'm SSH'd inside of Cursor into the DGX Spark that you see behind me on this shelf. That is a DGX Spark with 128 GB of unified memory, and we're just going to run Hermes setup here. And the most interesting thing is that you can actually import your data from Open Claw for this setup. So, if you guys are using Open Claw, you can actually just import all of your information from Open Claw over to your new Hermes agent. I'm not going to do that for this process.
The next thing that we're going to do is we're going to be able to do a quick setup. So, you can connect this with Nous Research subscriptions. You can use their portal, you can use Open Router, Anthropic, OpenAI Codex, and I'm actually going to use OpenAI Codex, okay? Because I'm just going to be able to use my existing OpenAI subscription for this. So, I'm going to, I'm going to go to the URL real quick and I'm going to authenticate, and then we'll move on to the next step. Okay, login was successful, and I'm now going to configure this with GPT-5.4, which brings me to the next step, which is actually configuring this with a messaging platform. So, I'm going to set this up with Discord because I already have a Discord server where I manage my Open Claw agent. So, I'm going to set this up with Discord, and that's going to be the next thing we're going to do, but this is going to take a little bit of time, but that's the next step. I'm going to just go here and select Discord. So, let me do that real quick, and then I'll be right back.
All right, guys, so check this out. I now have Hermes agent set up and configured inside of my Discord server where I have my other Open Claw agents. So, you can see my Open Claw agents are currently offline, but here's our Hermes agent right here. And as you guys can see, I'm able to just ping this agent, and just like it's very typical of Open Claw, but, "Hello there, how are you?" I can ping this agent, and it's going to be able to respond to it, and it's going to be able to respond to me. Like you guys saw, this is configured with GPT-5.4 via the Codex CLI. So, it's going to be able to respond there, but I now want to show you guys what we're actually going to be doing.
So, I'm going to be configuring this agent. The agent's name is going to be Vivian. And this agent is going to be tasked with doing cold outreach to businesses and brands in the vibe coding space to identify partners for BridgeMind to actually partner with for different events inside of the BridgeMind community.
So, as you guys know, BridgeMind is the home of the vibe coding movement. There are over 8,000 builders in our Discord server, and if you guys see here, this is one of the most active, if not the most active, Discord server in the vibe coding space, and we are going to be doing more events. The latest event that we did was a $5,000 vibathon that I funded personally, but the next events that we're going to be doing are going to be sponsored by brands and companies in the vibe coding space.
But to do that, I need this Hermes agent to be configured with an email. It's going to have its own persona. It's going to be able to improve itself, and it's going to be having a writing skill and a cold outreach skill, and what it's going to need to do is do cold outreach to companies in the vibe coding space to reach out to the correct market, marketing, or partnership person and write a cold email to actually find partners for upcoming BridgeMind events. So, that's what we're going to be trying to do in this video, and we already are very quickly able to set up this Hermes agent. It's running on the DGX Spark, and you guys can see that I literally did this in like under the, under 10 minutes. So, let's keep going with this, and let's get it configured with our Outlook email.
All right, check this out, guys. So, I am now using Cursor to configure the soul user.md file, memory.md file, and agent.md file of Vivian. I was able to correctly set up my Azure credentials so that I could configure this so that Vivian will have email outreach capabilities. That was actually the most challenging part. It took me maybe 20 minutes to be able to get that set up. I had to get some environment variables and be able to configure them inside of Azure, but I now do have that set up properly, and you can see here that Cursor is finishing its plan now, and it's just about done, and it's, is done now. So, I guess we'll just be able to pull up Discord and be able to see if we can do some scraping and to send our first email using Vivian.
I want you to scour the internet and find a list of 10 brands that would be good potential partners for BridgeMind. I then want you to identify the correct marketing email or marketing person to reach out to for a cold email to send them an email about the upcoming BridgeMind event that they can sponsor for $10,000. Note in the email that BridgeMind is the home of the vibe coding movement with over 60,000 subscribers on YouTube and 25,000 followers on X and a Discord community of over 8,000 members, and note that we are currently doing outreach for sponsorships for this event, which is a buildathon.
All right, so we're going to send this out, and we're just going to see what is actually possible. So, let's just, let's just test this out. I'm going to, I'm going to send this out, and we'll go to this thread, and we'll see: is it, first of all, is it able to actually find a list of 10 brands, and then is it able to find the actual email? So, look at this. So, it says Hermes is planning three tasks currently. So, three tasks are currently being planned, and I just want to give you guys a look into what it's doing. So, it says, "delegate task," and I, this is, this is important, 'cause I just want you guys to see what it actually does. So, it's planning three tasks. Is there any way that I can actually see what it's doing? And, and this does function differently than, than Open Claw, for sure.
In my opinion, I think that this is actually better than Open Claw because this is just really interesting. I think that we could do email outreach, and then let's say if one of these is successful, right? Like let's say that we get good feedback from one of these companies. With this ability to actually improve itself, it will be able to improve and do better outreach and have better writing capabilities. Like I think that this is something that will have it just wake up, do cold email outreach, and it'll probably bring in a bunch of partnerships and sponsors for BridgeMind. So, I'm going to come back to you guys when this is actually finished, and we'll see the results.
It's been about 8 minutes, and right now what I see is we had that to-do and then the delegation of tasks. And right now it's actually executing code. So, I don't know if it's potentially has a snake, so I don't, I don't know if it's installing. It looks like it's using Python to maybe do some search requests or something like that. You guys can let me know, but it looks like it's still working, so we'll let it do its thing, but I just want to give you guys a quick update on the status here of the Hermes agent.
Our Hermes agent Vivian has now completed its task. It took under 20 minutes to complete it, but you can see here now that it was able to find a list of 10 potential partners for BridgeMind and find their partner address, right? So, look at this. Startup Partners, DDS FS Partners, but it found Notion, DataDog, Neo4j, Confluent, Algolia, Cloudflare. So, a bunch of different brands that it found, and we're going to be able to refine this. Here is the cold email draft that it has. So, the subject is "BridgeMind Build-a-Thon sponsorship," and it's reaching out from BridgeMind, the home of the Vibe Coding movement. We're currently doing outreach for sponsorships for an upcoming BridgeMind Build-a-Thon and thought that you could be a strong fit. BridgeMind has over... So, it perfectly did the... It did everything; the, the writing is really good, too. So, I'm really impressed. Here's another draft, better version, with one custom line. So, this looks good as well. Reaching out because your company is a strong fit for the audience we've built at BridgeMind. So, this is impressive.
And what? Look at this here. At the end of the conversation, at the end of this task, it also created a skill called sponsorship prospecting. And this is really, really interesting. So, you can see here that it created a new skill, and this is part of that, that feature of Hermes, to be able to refine itself, to create skills, to refine those skills based off of feedback and based off of performance. And this stands out to me more than Open Claw did. You know, I set up Open Claw with the three Mac minis that you see behind me, and you guys can see that even from the start of the video, these actually just got switched over here, 'cause from this experience setting up Hermes and getting all this configured, this is better than Open Claw in my opinion. And the self-improving loop, when you think about a employee, when you think about a person that's doing work for you, you really do need that self-improvement. You need your AI employees, your AI agents to actually be able to improve themselves. You can't micromanage them. They need to be able to learn from their experiences.
So, I am going to hook up these three Mac minis and set them up with different Hermes agents. And I even may actually fork Hermes and maybe make some refinements to it, but I'm very interested in this, and I can see the value of this to have an outreach agent that's able to do cold outreach for you, to find partnerships for your business. Like I said, partnerships with BridgeMind are worth over $10,000 at this point. So, we'll see how much money this Hermes agent is able to generate for us. But just to be able to instruct this agent to do this cold outreach, that's going to make a difference for me. And I think that the Open Claw hype has been a little bit, like, big hype. Obviously, you see the souls, you see the memory, you see the self-improvement with Hermes, and that's definitely very interesting to me. And I want to start getting into this, and over the coming weeks, I'm going to be showing you guys more of this on stream because I think that I'll be using Hermes agents to do things like cold outreach, to do things like partnership outreach, to be able to manage customer service.
So, you're guys, this is the first time that you guys are seeing this. So, even like with this one, I can now follow up and say, "Great work. I now want you to choose the best one and to actually send the email to that company." Let's actually see what it does here. I want to test this. So, I'm going to actually say, "Hey, I want you to now send one of those emails, so choose it and then send it and just test to make sure that not only is it able to do that, but it's actually going to be able to send the email for us." So, it looks like it's searching files, Outlook. It's now reading from the agent tools. It has that Outlook graph tool that we hooked up Outlook with. And yeah, guys, this is... This is pretty impressive. And like, like I, I notice things when they're useful, like, pretty quickly. Obviously, I've been Vibe Coding for over 3 years now, and I've been doing the Build in Public series for a while now, and this is going to be very, very valuable for me. And the fact that it's self-improving, again, that's what's sticking out to me. If you have Open Claw where you have to micromanage some AI agent, that's going to be challenging. But if you can actually get it to do this feedback loop where it's able to do the self-improvement, that's what you need in an AI agent. That's what you need in an AI employee.
So, I'll come back to you guys. We'll see what email it writes and who it chooses to write the email to. But initial thoughts are: I'm very impressed. It runs very well on the DGX Spark. And look at that. It looks like it already sent it. Look at that, chat. So, it sent it to AssemblyAI. It says why it chose it: strongest combination of audience fit and a clean public partnership contract. AssemblyAI x BridgeMind Build-a-Thon sponsorship. And it literally verified that it sent. I mean, guys, this is, this is very, very impressive. So, I'll update you guys with this on the streams this coming week for day 163 of Vibe Coding an app until I make a million dollars. But stay tuned on this because what you guys are seeing is like this is a legitimate AI agent, right? Like this is going to be able... Maybe I can set up some type of cron job so that it does this every day. But I am going to set up these three Mac minis as well and see if I can set them up with different personas, with different memories and different tasks, and maybe one of them can handle customer service for BridgeMind. But we're at $76,000 in ARR, and obviously we're going to need some AI employees to be able to help us in the management of BridgeMind. So, I'll keep you guys updated with this. But my initial thoughts are: I'm very impressed, and you guys are definitely going to see me using these more on stream.
But with that being said, if you guys enjoyed the video, make sure you drop a like and subscribe and turn on post notifications. And if you haven't already joined the BridgeMind Discord community, the home of the Vibe Coding movement, make sure you do so. There's going to be a link in the description as well as in the pinned comment below. And with that being said, guys, I will see you guys in the future.