Full AI Employee Setup with Agent Skills (Hermes Agent Tutorial) transcript

My Hermes agent is one of the single most valuable assets in my business today. It manages my ads account. It posts on my social media. It responds to my comments. It even replies to all my DMs on Instagram, at school, and LinkedIn. And every night while I sleep, it looks for ways to improve itself and improve all of our marketing systems across cold email, paid ads, social media. So, listen, if you're sick of watching sponsored videos showing you how to set up a Hermes agent on a Hostinger, then you're in luck.

Cuz in this video, I'm going to walk you through not just deploying a Hermes agent, but we're going to actually do best practices in a not sponsored way. But, I'll also show you how to give it the skills and the tools to do exactly what my agent does for me. You're going to be able to launch and manage your agent from Cloud Code without ever having to log in to someone else's website or dashboard. You're just going to be able to talk to it, and it's just going to work, and it's going to be able to do everything.

No affiliate codes and no sponsors [music] here. And if you're new to the channel, I'm Lead Gen J. I run multiple eight-figure companies, including software companies and agencies. And I've traveled the world teaching AI. And today, I'm going to show you exactly how my Hermes agent works, and his name's Bob. You ready to meet him? So, here's my Jarvis dashboard. And as you can see, my Hermes agent, this is actually live.

It's actually running 35 scheduled jobs that happen throughout the day. Now, my Bob currently runs on a Mac mini, which I'm remoted into now. This is the dashboard of what Hermes looks like. Now, I never log into this. In fact, I'm going to be showing you how to operate your entire agent, to set it up, configure it, and improve it. Anytime it gets something wrong, be able to teach it new skills, new scheduled tasks. Not by going in here, but actually just by using your normal Cloud Code.

Now, mine is set up on a Mac mini because I like to iMessage it, but most people set them up on a virtual server. So, let's talk about step number one. Now, most Hermes videos that you've probably watched just focus on setting it up and launching one. Now, I'm going to show you what I did and what I do for my customers, which I consider to be best practices. Let's talk about Hermes for a second. Hermes is the most powerful agent harness by far.

I've used all of them, open claw, claw channels, paper clip. There's a reason Hermes is by far the most popular AI agent. It is the best configured AI agent for being able to handle multiple tasks, being able to operate inside of a single iMessage channel and just get things done, and it's just configured the best possible way. So, it's a free GitHub repo that you can install. Once you point your Claude code to this GitHub repo, it'll be able to just install Hermes and you'll be able to launch it, >> [music] >> and it'll install a dashboard that looks like this that you can log into.

You can see the files and models and cron jobs, which is all of the scheduled tasks. Here's all of the ones that I actually operate in my actual business. So, you can see some of the things that it actually does. Checks my emails, morning briefing, daily review, calendar preview, social report, infra quick check, so it can actually go log into my other computers and see what's going on. But, I don't want to spend too much time talking about this Hermes dashboard, cuz to be honest, you're never going to log in.

I want this video to be as practical as possible. So, let's talk about the initial setup, cuz the initial setup just really involves installing this GitHub repo into some server. Now, there are services like Hostinger, which I tried just cuz everybody's talking about it. I wanted to see what their process was, and I do this for clients. So, you can use one of these easy setups, but there's a lot of catches. For example, the server's going to be super weak.

In order to get the low pricing, you're basically accepting an incredibly low-power server that's controlled by Hostinger. So, because of the convenience that they offer you at Hostinger in setting up one of these agents, you're going to pay for it in the cost of running the agent. And if you're like me and you want it to do more and more powerful things, you're going to want a more powerful server. So, what you should do instead, and this is what I do for myself and for clients, is there's a site called Hetzner.

This is actually what developers use to launch virtual servers, and I never have to log into Hetzner anymore. You set this up once. I'm going to show you what this looks like, and once you set it up and get an API key and load your billing information, your Claude code is actually going to be able to not only deploy and configure the servers for you. It actually starts new subscriptions for me, and it can SSH, which means connect, directly into that server.

So, let's go into my Hetzner console. So, the first thing you need to do is create an account on Hetzner, sign up for an account, and put your billing information, and then go into this default project that's created for you. And within here, you'll be able to get an API key from Hetzner. So, I'm going to hit the settings, and then we're going to go into security, and under API tokens, you'll be able to actually create an API token, give that over to your Claude code, and not only is it going to be able to configure the server and install Hermes for you, but it's going to be able to install lots of virtual servers.

This is what I use to install virtual servers. It's going to be the lowest price, about $5 to $6 per month per server, and you're going to be able to easily upgrade the capabilities for the lowest possible cost, and no annual contracts, no nothing. Hetzner is the developer's choice. This is actually what a lot of companies like Hostinger use. They just upsell Hetzner. And all you have to do is get this API token, come over to the API docs, and they actually have a CLI already.

So, just point your Claude code to the Hetzner CLI, and you're going to tell it, "Listen, I want to set up a Hermes agent. Here's the GitHub repo. Here's my Hetzner keys in the Hetzner CLI. I would like you to find the correct configuration for a server for me to set up that's going to be optimized for [music] Hermes AI agent uses." And it will know exactly what configurations, it'll set the SSH key, and now it'll actually log in and install Hermes for you on that virtual server.

Now, the beautiful thing about doing it this way is when I want to talk to my server, to my Hermes agent, I never have to go into a dashboard. If I wanted to change it from Telegram to Slack for my main form of communication with Hermes Agent. All I would have to do is this. So, here's my Cloud Code terminal. I'm just going to open up a new session. And for me to talk to my Hermes Agent, I have a short command. I call it {slash} Bob, which is the name of my Hermes Agent.

And when I activate {slash} Bob, it knows, "Okay, I'm going to SSH into that server, and I'm going to be able to configure and talk to that Hermes Agent, even checking my past messages." So, if you do it this way, you're never going to have to worry about, "Oh, it's screwing up. It [music] forgot this tool. It's saying something that it shouldn't say. Or I can't talk to it." You can now debug everything by doing this.

So, if you connect your main Cloud Code to that virtual server first, it'll install everything, debug everything, and set it all up for you. In fact, if you're already using Cloud Code or Claude, you can actually have it train your Hermes Agent initially, right off the bat, with all of the things that you actually do in your day-to-day Claude operations. So, it can take that over in its scheduled tasks and skills. And this is actually what I'm doing for clients.

As I'm figuring out, "Okay, here's all the things that my Hermes Agent does on a daily basis. Research, image generation, inbox and calendar." So, that when somebody installs a Hermes Agent with us, it comes pre-trained. But, you can actually train it through your Cloud Code. And your Cloud Code has access to all of your local files and tools, so it can just install those into your Hermes Agent. In fact, let me show you how easy this is.

Uh right now, I'm going to tell my Cloud Code to use my Hetzner API and spin up a brand new Hermes Agent right now. And by the end of this video, it's going to be live, and I'm going to be talking to it by Telegram. So, let's go ahead and do that now. I'm going to go to I'm going to copy this GitHub URL for Hermes. I'm going to open up my Cloud Code. And let's just open up a new session and say, "Let's use my Hetzner API key and spin up a brand new virtual server and install Hermes on it.

And I'd like to configure it to talk to it by Telegram and install some basic skills. Here is the Hermes GitHub repo. Please use my Hetzner API and spin up a brand new server for this. I'm going to let that run and we'll come back to it during some points in this video, but I really want to spend the majority of this video talking about the hard parts, the stuff that isn't covered in most videos cuz it really is as easy as this for setting it up.

You don't need to use Hostinger. This should really be as simple as possible. Now, it wants to use a pretty powerful server, 8 GB of RAM, way more powerful than they're going to give you with Hostinger. I'm going to say just use the lowest cost VPS. This is a demo. But as you can see, it already knows the different virtual servers and configurations that it could use. There we go. It found one that's about €4 per month.

Awesome. Let's go ahead and do it. I found one that's less than that. Uh go ahead. So, it's going to start building that for me now. That's funny. It said filming mode is on. I basically have a mode when I'm doing content like this. It's actually brand new and it stops all background processes so I can focus. So, let's go ahead and do film off so that we can continue. All right. And now we're off to the races. It is configuring that Hermes agent for me.

Now, in the meantime, let me talk to you about what mine actually does, how I talk to it, and how I give it the tools and training so that it can actually help me in my business. So, there's a couple secrets that I'm going to be sharing with you here. Now, I talk to mine through iMessage. Now, you don't have to talk to it via iMessage. Mine's name is Bob. Here it is. This is Bob. And Bob like a real employee. I made him his own Gmail account.

He's got his own phone number. He's got his own Slack account. And he communicates with me and my team as if he's a real team member. So, the first thing I'm going to suggest to you to do is that. Treat it like a real employee. And that way all of the scheduled tasks, all the skills, all the things it does, it's doing under its own fingerprints. So, you can tell where it came from and how to fix it. You also want to make sure to give it its own GitHub repo and its own Superbase backend.

So, you need to push code and skills and updates to it. And the best place to do that is via GitHub. So, the first thing you're going to do if you don't already have a GitHub account, sign up for one and create one for your AI agent. Here's all of my GitHub repos. As you can see, there's quite a bit in here. And Bob command center is actually my GitHub repo for my Hermes agent. So, you're going to want to make sure it's connected to a GitHub repo.

And second, give it a Superbase back-end. Both of these things are completely free. You're going to create an account on Superbase and set that Superbase up. And I'm going to show you how to connect your Hermes agent to both of those. So, now it has a back-end and now it can remember code and remember updates. This is just the base layer of setting up a really good and effective Hermes agent. Sign up for an account on both of these.

And now you're going to want to connect it not just to the front-end and back-end, but connect it to all of your actual tools. So, the next thing you're going to do is go and sign up for a platform called Compozio. It's compozio.dev. And it's completely free to free you to use. In fact, I don't pay Compozio anything. Compozio is great because they've already built integrations into all of these different softwares and tools.

So, that way you don't have to worry about managing all of those integrations. You just do it through this one place in Compozio. You give your agent access to Compozio. And now it has access to all of these different tools. Now, you'll also notice that two of the top tools on this list are GitHub and Superbase. So, the first thing that you're going to do is set up your AI agent, go to compozio.dev, and integrate your GitHub account and Superbase account with your AI agent.

So, you're going to connect these now. It's literally as easy as pressing connect. I actually connect to Superbase through a different method. So, you're just going to go through one at a time and connect all of your main tools. Now, what you'll notice here, Google, Google Calendar, Slack. If you've ever tried to connect your cloud code to these things without Compozio or a second layer like Compozio, you you know it's a pain in the butt.

You have to create custom developer apps and all kinds of crazy keys. [music] The beautiful thing about Compozio, they handle all of that for you. There's even some apps that I'm just like, you know what, screw this. It's too hard. It's not worth connecting. Like Zoom, QuickBooks, I actually went through the pain and built the integration. But if you come in here and you look, okay, I want to connect my Zoom. Awesome, there's a Zoom integration.

I want to connect my QuickBooks. Awesome, there's a QuickBooks integration. Now I don't have to worry about building developer apps and managing the the connections at all of these things. So you're going to come through, you're going to connect all of your apps, and then you're going to come down here where it says connect my agent. And it's got all of the pre-filled agent responses that you can just copy. So as you'll see you can see Hermes right here.

You can literally just copy this. Now you can put that into the terminal in your VPS, or you can do this the easy way. Because remember, we're managing our Hermes agent through Cloud Code, through our normal Cloud Code. We're not doing anything weird. You'll never have to log in, you're just going to connect to it through here. So as you can see it's still working on actually setting this up. I'll give you another little golden nugget here.

You might see it asked me where to find the token. It asked about Doppler. So this is actually another platform that I use that's completely free. And I use this for managing tokens. So when you try and give your Cloud Code an API token, or you try and give your Hermes agent an API token, what you're going to notice is it says the same thing. It's like this token's now been exposed, you need to rotate it, and it's going to keep fighting back against that token.

It might not store it for later. Doppler handles that problem. So I actually have this shortcut set up on both my Stream Deck and my mouse, where I can just click here, and this is my API key command. And what this does is it copies the key and adds it to Doppler, which is connected to Vercel, which is like the front end of all of the apps. So it puts it everywhere, and it makes it really accessible for all of my agents, and it doesn't fight back, and I can manage all of my tokens from here.

And I never even have to log in. I just use that shortcut command. If you're interested in installing that skill or any of my many amazing skills that I work so hard on, if you come to legion-ai.com/skills, you can see pretty much all of these. I think Doppler is a a skill that I've created. Yep, easy Doppler API. Boom. So, what this does, stop pasting API keys into your AI session. Put them all into Doppler, and it guides you through setting up a short key command like this.

So, anytime you want to share an API key, it's just one click. Okay, so the next thing you're going to do is copy this Composio CLI install, or honestly, you can literally just say to your Claude code, "Hey, I want to install Composio for my Hermes agent. Please install it. Here's the key." And then you just need to get your key, which is really easy inside of Composio. You're just going to come down to settings. You're going to come into sessions and API key, and there's your API key right there.

You're going to give it to your Claude code, give it to your Hermes agent, and now it's going to be able to use all of the tools and tooling from Composio. Boom. Now your AI agent has access to everything that you actually use in your real business. Now, another little hack here with Composio especially is maybe you want to give it access to its own email, but you also want it to access your email. Cuz maybe you want it to do what mine does.

I want it to review my emails every morning, tell me if there's anything important, filter it, and go through those scheduled tasks of the cleaning up my inbox. But I also want it to have its own email address, so it can send me updates and correspond with my team. Within Composio, you can sync way more than one account. So, as you can see here, I've got five different Gmail accounts synced, >> [music] >> and now it knows that it can access all of those.

And same thing with the other integrations. So, if you have your own personal integrations that you want to use inside of your Claude code, but separate integrations for your Hermes agent, you can link them all here. This is honestly such a helpful tool. Now, what you'll also notice here, social media is not really in here. So, I told you I was going to show you how to give your AI agent access to do everything that mine does, and mine replies to comments.

Mine manages my CRM. It talks to my leads. None of those are integrations that are in here. In fact, all of the integrations in here are going to be somewhat limited. They're limited because you only have the tooling that Compositio gives you, which in most cases is fine. But, if you want to integrate your GoHighLevel to do the things that mine does, like building workflows, like managing my sub accounts, things that GoHighLevel's API doesn't let you do.

For example, Instagram. There is no good Instagram connection in Compositio. Sorry. Now, you can connect it to ManyChat or GoHighLevel and do some workarounds. But, if you want to really unlock the superpowers of this thing, then you're going to want to give your AI agent access to a couple of other things. The next superpower that you can give it is Unipile. Now, what Unipile does that's really unique is they actually maintain the API connections with your social media platforms in a way that Compositio misses.

So, let's take LinkedIn for example. If I want to respond to DMs in my LinkedIn account, if I want to respond to DMs in my Instagram account that are more than 24 hours old, there's nowhere else to go. You have to do it through browser automation, and that's super sketchy. Or, you can use Unipile, and it's unbelievably affordable, about $5 per account. So, if you really want to use this to manage your social media, you can.

And it has a bunch of other connections as well. You can connect your WhatsApp. You can connect Facebook Messenger. You can connect Telegram and all of your email accounts, and you can manage all of those connections through Unipile. Now, where this really thrives is specifically in DMs and responding to comments, cuz those are two things that there's really no other good solution for. If I want to post to my LinkedIn or post to Instagram, those are easy.

I'm going to share the tool that I use to do that next. However, if I want to reply to comments on my Instagram, if I want to comment on somebody else's post on Instagram or on LinkedIn, there's not a lot of good solutions for that. Unipile solves that and gives your AI agent access to do way more stuff. And just checking back in on my Cloud Code session, server is up, Hermes is up, box is hardened, installing Hermes now.

I've literally said like three things to my Cloud Code session and it's already setting this up. I'm just on Opus 5, I'm not doing anything crazy. This is literally how easy it is to set up Hermes using this method. The only friction that you might face is on Hetzner, just like authenticating your account and adding your billing info. But once you do that, it is smooth sailing. Now, the next thing you're going to have to think about when you're launching a Hermes agent is how you're actually going to interact with this thing.

Now, it gives you a few different choices. So, let's actually open up the dashboard and talk about communicating with it. And then I'm going to give you some other sauce that you've probably never heard on YouTube. So, let's actually come into channels. This is how you talk to it. And I've got Telegram is disabled, Discord, Slack disabled, WhatsApp disabled. These are all options in which I can talk to my Hermes agent.

Now, I'm using blue bubbles iMessage because it runs on my Mac mini. But, if you don't have a Mac mini, you don't want to get a Mac mini, but you still want to talk to this thing via iMessage, there actually is a good way to do it. And I'm going to show you what that is now. This is something that you probably never heard on YouTube. But for most of you, just know that Telegram is probably the easiest way to do it. You just create a bot and it'll guide you through it.

And then you can interact with these things via Telegram. However, if you do want to use iMessage, you do not have to be restricted to Telegram and you do not need a Mac mini anymore. There's a platform, it's called Photon. It's called photon.dot.codes. And this actually gives your AI agent the ability to talk to you via iMessage and they have a free plan. So, if you're an individual and you want to iMessage with your AI agent, you can install Photon codes and get a number to use completely free.

So, that's communication, and I'll walk you through it once this agent is done getting set up. The next thing is what large language model's going to power your Hermes agent. So, mine is powered by my Codex subscription. Now, I have a $200 per month like maxed out Codex plan that it runs on, and it's ultra powerful, and it's ultra smart, and it's able to do everything. However, you don't need that. It can run on your $20 per month Chat GPT subscription, and it runs really smoothly, and probably still won't run out of credits.

What I'd recommend is not running it on the latest biggest best model. You don't need it to run on Sol 5.6. Mine does cuz I'm on the $200 per month plan, but if you're on the 20, use that. And then the other option is you can give it API keys. So, if you want it to run on Claude, unfortunately, syncing your Claude subscription's not very easy anymore, but you can give it an Anthropic API key, or you can get an account at openrouter.ai, and give that API key over to your Hermes agent.

So, it can use whatever model that you want. Kimmy K3, Deep Seek, whatever cheap Chinese model, but what I'd recommend doing is just syncing it with your Chat GPT subscription. It's really seamless, and that's currently the best option for powering these agents. Now, for me, since I'm just spinning this up for the first time, I'm going to go ahead and say reuse my open router API key. Now, if you choose to set this up with Chat GPT, you're just going to tell it you're going to use your Chat GPT subscription.

It's going to give you a login link. You're going to login on on the web, and then you just paste the code, boom. Now, it's synced with your Chat GPT, and you'll probably never have to do that again. Now, let's talk about Telegram. Now, you can have your Claude run you through creating a new bot, but it's really easy. So, let's just show you how easy this is. You're going to open up a conversation with Bot Father, just like this, and you're going to say new bot. /newbot.

All right, a new bot. What are we going to call it? Let's call it LGJ Hermes demo. Great. Now, let's choose a username for the bot. I'm going to use the same username and it must end in bot. Cool. So, _bot. Just like that, I've created the bot token. It's got a token that I can use or I can pass this on to somebody else to use. But, this is the token that you're going to need to pass into your cloud code and your Hermes agent in order to sync it with Telegram.

So, let's go ahead and do that now. I'm just going to give it that token, submit, and just like that, Hermes is installed. Both keys are in. The gateway defaults to deny all. I want my Telegram user ID to allow list me. Token valid. Model defaults to Opus 4.6 via Open Router. That's fine. Now, remember, because this is logged in to the virtual server that's running that Hermes agent, I can tell it to do whatever I want.

Switch the model, switch the Telegram, install this skill, install this scheduled task. I never have to literally log into that VPS or see that dashboard once. I don't need to go into cloud code on that virtual server at all because this already SSH'd in, communicated with that server, and now what I would like recommend doing, since you did this in your cloud code, is just set up a command. I would just tell it, "Hey, create a command for /hermes that automatically lets you SSH using the saved credentials into it to work on that Hermes agent and modify the files.

Here's the GitHub repo. Here's the Superbase." And this will allow you via cloud code to work on your Hermes agent anytime that you want. Now, while that's building, let's give you some more sauce cuz most of the tools that I gave you, Compozio, Unipile, you've got most of your bases covered. But, mine also manages my social media. It's able to post across all of the platforms. It's able to create stories. It's able to generate videos and images.

By the way, the way that it does that, coming back into Compozio, one of the integrations is called file.ai. As you can see, this is integrated here. You can create an account on filedep.ai, and now via Compozio, it can create images and videos and use the latest models. It's really powerful. Now, what if you want your agent to be able to post to your social media platforms, your stories, your Instagram, your Facebook, your X.

You need a way to integrate all of those platforms, and you need a platform that's configured for your AI agent. What I use is called upload post. And if you want discount codes and the links to all of my software and tools, it's going to be down below. But this is how you can sync all of these different social media channels, give your agent access, and now it can post across all of these channels, and it will already know what to do because you'll install the CLI.

You'll say compatible with any AI agent? Cool. You give it the tooling so it understands how to use the API. That's what it That's what this AI agent skill is. And then you give it an API key once you sign up, and now it's going to be able to post to all of your social media channels. Now, at the beginning, I told you that I was going to teach you how to make it self-improve. Well, there's a couple cool tricks here. Well, one of the things that I have my agent doing, and you can tell your agent to do this as well, is every night, once you're done working, go through the past 24 hours of work and see what you can do to self-improve and improve your results for the day after.

Now, there are all kinds of frameworks for self-improvement, for doing research, for searching for the right answer, for doing browser automation. And all of these are skills and tasks that you can teach your agent to do. And the trick is to refine the agent and refine what that agent is able to do and how it performs just by day-to-day going through these self-improvement passes, going into your cloud code, and saying, "Hey, it screwed this up.

Go through the past 24 hours of messages, see what you can improve." And continuing to give it new skills and new things to do for you. Now, important for you guys to understand, getting AI right is not as simple as install it and it should just work correctly. It takes a lot of refining. It's a combination of skills, meaning refined prompts, edge cases and rules, if this, then do that, look out for this, this, and this.

And most of those are only developed with time and repetition. And then the third thing is tooling. What does it have access to and how does it use those tools the correctly? And once you're able to create synergy between those things, the skills, the tooling, and all of the rules, now you get an AI system that's really awesome. [music] So, if you want to copy mine, I've got all of the skills and tooling and rules that Bob uses, my Hermes agent.

I've refined this over the course of 6 months and this is through open claw, claw channel, now Hermes agent. All of that's just been building and improving and we found ways to make search better, make reporting better, make the text and email it sends me look better. And I said this was not sponsored and it's not. No one's paying me to make this video. It's sponsored by me. So, if you do want to support and you want access to all of my AI agent files and configurations, I do have a program that you can join.

It's called the AI automation insiders, where you get access to all of my configurations and I'll actually deploy one of these for you. It's pre-trained with all of my stuff. Or you can then just install the zip folder, appointed at your claw code and say go nuts. But that's not all that you would get. I also have all of my AI systems available for you. So, I said that my agent understands how to run all my ads accounts.

Why do you think that is? It's not cuz of the integrations. The integrations are easy. It's the tooling, the rules, the gotchas, and all of the prompts that have been refined over long periods of time. So, if you want it to run your cold email systems effectively, you need to install our AI cold email system that has 10 skills, tons of tools, tons of rules and it's been refined by experts, including myself, over time.

Our fully automatic go high level. So, being able to give it access to build workflows, build funnels. I think we're on version four now of this. I'm always sending updates. We're always updating this stuff. So, if you're interested, I do have a link down below where you can go and join. We'll launch a Hermes agent for you. You'll get immediate access to all of this material, as well as our exclusive trainings. And just like that, as you can see here, the server's now spun up, Telegram is valid, and now I can start talking to this thing right away.

Send any message to this bot, and I'll pull the user ID and allow list me. So, it's ready to go. I literally did this kind of while I was making a video. That's literally how easy it is if you follow along the instructions that I gave you. And now you have a Hermes agent. That's just the setup part then. That's the easy part. The hard part is making it profitable, making it useful. In order to do that, you need the training, and you need to practice, and you need to install things.

You need to get things wrong. And you need to self-improve. To do that, you need to be part of a community, and you need to be actively working on this. So, we've got the community for you. It's the AI Automation Insiders. And if you found this information helpful, and you're like, "Thank you so much just for sharing the sauce. Nobody else does this. Thanks for sharing the tools that you use, the integrations." I'd really appreciate you liking the video, subscribing to the channel so that you see more of this, and leave me a comment.

Let me know what you think. Let me know what you're doing with your agent. I'd love to respond. I promise to respond to all of them personally, and I will see you in the next video. Thank you so much for the support.