Hermes Agent Tutorial for Beginners (Full Step-by-Step Setup)

If you've been looking for a clear, step-by-step guide to setting up your own Hermes agent, you're in the right place. I'm Matt, and in this video, we're going to get your agent running on a small cloud server, connected to Telegram so you can message it from your phone, and walk through how to give it skills it can use again and again. I'll also show you how to schedule jobs so it can do things for you on its own even while you sleep. If you're worried the video is only going to cover the setup, it's not. Once your agent is running, we'll go through how to actually use it, what it costs to keep running, and how to make sure no one else can talk to it. By the time we finish today, you'll have an agent that listens on Telegram, runs work for you in the background, and gets a little better the more you use it. You don't need any coding experience to follow along, just the ability to copy and paste. Before we touch the server, let's quickly cover what Hermes agent actually is and why someone would want to run one. Hermes agent is a free, open-source AI agent made by a team called Nouse Research. The simplest way to think about it is this: it's like having a personal version of ChatGPT or Claude that lives on your own server, talks to you through chat apps you already use, and gets better the more you use it. There are three things that make it different from a regular chatbot. First, it runs on its own all the time. That means it can message you first when something happens, and it can do work for you on a schedule. A normal chatbot only does anything when you open the app. Second, it remembers how you work, not just the facts you tell it, the way you like things done. It builds up a picture of you across every conversation. Third, it writes its own skills. When you ask it to do something complex, it can save the steps and reuse them later, so it gets faster and more accurate the longer you have it. Quick note on where you can run Hermes. There are three options: your own laptop, a spare always-on machine like a desktop or a small home server, or a VPS, a small computer in the cloud. We're using a VPS in this video for one specific reason. Everything we set up here, the Telegram messages, the scheduled jobs, the agent reaching out to you when something happens, only works while Hermes is actually running. A laptop won't work because it goes to sleep. A spare always-on machine does work, but only if you already own one, and you'll be handling your own networking and updates. A VPS removes both of those problems for under $10 a month. Now, if you'd rather go the local route, the official Hermes documentation walks you through that path, and I've put the link in the description below. For everyone else, the rest of this video walks you through the VPS path step-by-step. Now, one more thing before we move on. If you've used a tool called Open Cloud before, Hermes is the same idea, but with stronger memory, the skills system I just mentioned, and tighter security. Hermes can also import your Open Cloud setup with one command if you ever want to switch over. Now, [snorts] the first thing we need is the server itself. So, let me show you how to get that set up. We're going to use Hostinger. The first link in the description opens a page that has Hermes agent already pre-selected for automatic installation, so we don't have to set that up manually. The link also stacks an extra 10% discount on top of whatever sale Hostinger is currently running, which makes this the cheapest way to get a working Hermes agent online anytime. So, go ahead and use the link on screen, or click the first link in the description below. When you click that link, you'll land on this page here, and if you click the choose plan button, that'll take you down to see the pricing options. You'll see a few VPS plans listed side-by-side. We're going to pick KVM two. It comes with two CPU cores, 8 GB of memory, and 100 GB of storage. I want to be straight with you about why we're not picking the cheapest plan. Later in this video, we're going to give the agent skills that let it browse the web, and the agent already runs inside a Docker container on Hostinger for safety. Both of these use real resources. The smaller plan can technically run Hermes, but it can't run all of that at the same time without slowing down or freezing up. KVM two also happens to be the plan that Hostinger themselves recommend for Hermes. So, go ahead and click choose plan on the KVM two option. You'll be taken to the cart page and the first thing you want to do is select the period for your registration. You can choose one month, 12 months, or 24 months and you generally get the best value, the lowest per month price overall, if you select 24 months, but if that's too much for you for getting started, totally understand. I recommend then selecting the 12-month option. That's the lowest term they'll let you select and still take advantage of our coupon. If you select one month, they don't let you use that extra 10% discount. Next, you can see that they already have the Hermes agent auto deploy set up for this particular option, so that's great. Last, you want to choose your server location. You generally want to pick a server location closest to you or whatever has the lowest latency. Here, I'll select one of these United States Boston options, which gets me only eight milliseconds of latency, which is fantastic. If you want to protect your setup, you could also turn on daily auto backups. By default, you get weekly auto backups. For me, that's just fine, but if you want to make sure that everything's protected, you can turn that on. Whenever you're ready, go ahead and click continue. On the next page, go ahead and register your account. On the next page, enter your billing address and your payment information to complete the checkout. And I want to point out that there is a 30-day money back guarantee, so anytime in the next 30 days, if you have any issues, you can just reach out to Hostinger and get your money back. When you complete the checkout, you'll get this confirmation. Next, Hostinger drops you onto a short setup form. Now, we just need to choose the admin credentials for the Hermes web terminal and let the server build itself. By default, they populate the admin username as Hermes, which works just fine and they give you a custom generated secure password. Now, click on the eye icon to reveal the auto-generated password. I'm showing mine here because I'm going to delete this account later, but definitely make sure that you don't share this with anyone. This is how you access your agent. Go ahead and copy this username and password and save it somewhere safe. Then simply click deploy. Hostinger is now building your server and installing Hermes on it. This takes a few minutes. So, I'll go ahead and fast forward to when it's complete. When it finishes, you'll be routed into the Hostinger dashboard and you might be presented with a survey. You can fill it out if you want or simply click skip. There's also maybe a pop-up or two and you can simply dismiss them. Welcome to Hostinger's Docker manager. You can see here that your Hermes agent is already running in a new Docker project and it has traffic set up as a reverse proxy to make sure that you can access your agent from anywhere on the web securely. Now we need to open a window into the server so that we can talk to Hermes directly. We're going to use what's called a terminal. That's the text-only window where you type commands. Hostinger has one built right into the browser, so we don't need to install anything extra on your computer. To open it, you'll simply click the terminal button in the upper right. But before you do that, go ahead and copy the last four letters here from your Docker project name here. So, it should say Hermes agent and then have four letters. Go ahead and select and then copy those four letters. All right, now go ahead and open the terminal. A new tab will open with a black terminal window and you're already signed in. Now, if you've never used a terminal before, it can look a little bit scary or overwhelming. It feels a bit old school or even kind of hackery, right? But really it's simple. Instead of clicking around with folders and icons to interact with a computer, instead you're simply typing text. Kind of like interacting with a chatbot, but you need to enter very precise commands. You can't just type whatever you want. So, first what we're going to do is make sure that we have a flashing cursor here and if this isn't flashing, go ahead and click on your terminal screen and it should get that going. Next, we're going to type CD space {forward slash} docker {forward slash} hermes {dash} agent {dash} and then go ahead and paste in your four characters that you copied earlier. And then hit enter or return on your keyboard. We just did that because when you first land on the terminal, any commands you type are going to the VPS at the high level, almost like the desktop on your computer. To send any commands to your Hermes agent application specifically, we have to tell the terminal to access that program first. So, that's what we just did. Now, we need to access the Hermes container. To do this, type docker space compose space exec spelled EXEC, space {dash} IT, space hermes {dash} agent space {forward slash} bin {forward bash and hit enter. All right, now we're going to activate the setup wizard. To do this, type hermes space setup then press enter. This starts the setup wizard. It's a questionnaire that agent runs the first time it boots. It asks which AI provider you want to use, which model you want by default, and which messaging app you want to talk to it through. We'll answer each question together. To start, the wizard will ask if you want a quick or full setup. We'll go ahead and just do the quick setup. It's already selected here, so all you have to do is hit enter. The first thing the wizard asks is which AI provider you want to use. A provider is the company that runs the actual AI model the agent uses to think. There are over 20 options here. We're We're to use one called open router because it gives you access to many different models with a single account and a single bill. That means you can switch between AI models later without signing up anywhere else. In the wizard here, use the arrow keys on your keyboard up and down to highlight OpenRouter, and then press enter. The wizard now asks for your OpenRouter API key. Leave this screen open in the terminal. We'll come back to it in a minute. Go ahead and open up a new browser tab, and go to openrouter.ai. You can type that in manually, or use the link in the description of the video below. This here is OpenRouter, and it's the platform that allows us to access a bunch of different AI models in one place. Now, the wizard is asking for what's called an API key. An API key is a long secret string that tells OpenRouter that requests are coming from you. Hermes uses the key to call the AI model, and OpenRouter charges your account for what gets used. Treat the key like a password. Don't share it with anyone. On this site here, go ahead and click sign up in the upper right. Then, follow the steps to create an account or sign in. If you're signing up for the first time, you'll need to verify your email. Go ahead and open your inbox and click the link. Here's what that email looks like. Should be something like this. The subject will say your sign up link, and then there'll be a button inside that says sign up to OpenRouter. Go ahead and click that. On the page that opens, you'll get a survey. Go ahead and select an option, and then click continue. And you should get a confirmation. Next up, we'll need to add credits to our account. We need just a small amount to get started. So, let's go ahead and click buy credits, and then add credits. Go ahead and add your billing information, and then add a payment method. After you've got your payment method added, we'll need to add an amount to start with. It defaults to $10, which is plenty for us to get started. So, go ahead and click purchase, and you'll get this little confirmation. You can just click okay. When the transaction completes, you'll see that your account balance has been added to. Great. Now we have credit. Next, we need to get our key. On the left side, click API keys. Then in the upper right, click new key. You'll need to give your key a name. We can call it something like Hermes Agent. Now, you can set a credit limit here if you want to make sure that your account has a hard cap. For example, you could set a limit for $20, and then have that reset every day, every week, or every month. I'll set mine to monthly. And you can set that to expire at a certain date, or leave it alone. Now, this is a cost guardrail. This is great, so it makes sure that you're never spending more than you want to spend. Now, you do have to add credits. So, this mostly works as a guardrail for if you have more credit on your account, and you want to manage how it gets used over time. For example, if you've got 100 credits on your account, and you only want to spend 20 a month, this is how you can manage that. You can leave this blank if you don't want to have that limit. Either way, set it up the way you'd like, and then click create. Now, here you'll see a long string starting with SK. This is your API key. Again, it's like a password. Don't share it with anyone. Go ahead and copy it, and save it somewhere safe because you will not be able to see this again. If you close this window without copying it, you'll need to start over and create a new key. So, just make sure you copy it before you do anything else. All right, go ahead and switch back to your terminal tab. Then, click into the terminal, and paste in your API key. Now, I want to point out here, you won't actually see anything happen once you paste it in. This is just for security reasons so that anybody who's watching can't see private information being entered into your terminal, but it is in there. So, once you've pasted that in, go ahead and hit enter on your keyboard. And you'll see the API key has been saved. Next, the wizard asks which AI model you want Hermes to use by default. A model is the specific AI brain the agent will use to think. Different models have different strengths and different prices. Some are very smart but expensive. Some are cheap but not great at the kind of structured work an agent does. For a beginner setup, I recommend DeepSeek V4. That's this one right here. It's cheap, about 43 cents per million words in and 87 cents per million words out. And it's reliable at the kind of multi-step work that Hermes does. You won't get a surprise large bill on day one. Now, a quick tip, not every model on OpenRouter works well as an agent. Two to avoid for now are GPT 5.4 mini, which struggles with the tool calls that Hermes needs, and the reasoning versions of Qwen 3.x with thinking tags turned on. Those tend to talk themselves out of using tools. You can change the model later with a single command. I'll show you how to do this, so don't worry about this too much up front. We're picking DeepSeek because it's both cheap and reasonably powerful. To select it in the list, use your arrow keys until it's selected, and then hit enter on your keyboard. Next, it'll ask where you want to have your terminal backend. We're just going to keep the current local, so go ahead and hit enter on your keyboard. And now it'll ask if you want to connect a messaging platform. So, go ahead and hit enter on your keyboard to set up messaging now. You can connect Hermes to Telegram, Discord, Slack, WhatsApp, and a few others. We're going to use Telegram because it's free, it's fast to set up, and it works on every phone and computer. To connect Hermes to Telegram, we need to create something called a Telegram bot. A bot is a Telegram account that's controlled by software instead of a person. Hermes will use the bot to send and receive your messages. All right, so first in this wizard type one in your terminal to select Telegram and hit enter. It'll ask you to hit enter to confirm again, so go ahead and hit enter. And now it asks for that Telegram bot token. All right, leave the terminal where it is for now and switch to Telegram. You can use Telegram on your phone or your computer. If you don't have Telegram yet, I have a link in the description below where you can download it. So, download it on your preferred device. In my case, I'm going to run it on the computer here. And then sign up or sign in to your account. Once you've got Telegram open, go ahead to the search bar and look for Botfather. Or you can click the link in the description below to open a chat directly with the Botfather. It's the one here with this blue verified check mark. Go ahead and click that one and then click start. You can either type the command {slash} newbot or click it from the message. You'll get a response from the Botfather saying, "What are we going to call the bot?" So, I'll call mine Matt Hermes Bot. Now we need to choose a username for your bot. And the bot name must be unique in the Telegram ecosystem. So, try to pick something that's not just generic. Otherwise, you might have to retry a few times as you find a username that is unique. I'll call mine Matt Hermes Wonderbot. And we get a confirmation saying the bot's been created. That message contains this long string here. This is your token to access the HTTP API. This is what Hermes is asking for. So, go ahead and copy this. And one important thing here, if you've used another agent like Open Cloud before with a Telegram bot, do not use the same bot token. Make a new bot for Hermes. If you reuse the old one, both agents will fight over the same bot and neither will work properly. All right, now go ahead and switch back to your terminal tab. Then click in and paste it. Again, you're not going to see this being pasted in. Just go ahead and press enter on your keyboard. And the wizard will confirm the token and move on. Next, the wizard asks which Telegram accounts are allowed to talk to your agent. This is important. Without this step, anyone on the internet who finds your bot's username could message it and use up your AI credits. We need to tell Hermes that only your Telegram account is allowed. To do that, we need your personal Telegram user ID. Telegram has a bot that gives you yours in about 10 seconds. Head back to Telegram and then search for user info bot. It's this one right here with the green profile and the ID for a face. Then tap start. The bot replies with your details. Go ahead and copy your ID. Switch back to the terminal, paste in your ID, and hit enter. Next, the wizard asks if you want this ID set as your home channel. That's where Hermes will send scheduled job results and any messages it sends on its own. Press Y on your keyboard for yes and then press enter on your keyboard. And setup is complete. The wizard now shows you a summary of everything that it set up. Your tool availability, where your settings are stored, and a list of helpful commands. Next, we'll want to launch the Hermes chat interface. To do this, type Hermes and hit enter. All right, and the chat interface opens inside your terminal. You'll see your model, your tools, and your installed skills listed at the top. Now we start what's called the gateway. The gateway is the part of Hermes that listens for messages from Telegram and passes them to the agent. Once it's running, you can close the terminal and talk to your agent right from your phone. In the terminal at the bottom here, where you have the cursor, type Hermes space gateway install and press enter. This wires Telegram into the gateway. It can take a few moments, so just wait until it completes. Next, type Hermes gateway run and hit enter. Again, this can take a few minutes. Now we hit a permission error because we're running as root user and the gateway wants to run as the Hermes user for safety. Notice that the agent itself diagnosed the problem and gave me three ways to fix it. The simplest one is the first one, run the gateway as the Hermes user. So, that's what I'll do and you can simply copy the command here from the output, highlight it, hit command C, and then paste it into your terminal. Or if for some reason it doesn't give you the version that you can just copy-paste, here's what that command is. sudo space {dash} u space Hermes space {forward-slash} opt {forward-slash} Hermes {forward-slash} {dot} v e n v {forward-slash} bin {forward-slash} Hermes space gateway space run. I know that's a lot, but that's why it's easier to just copy and paste and hit enter to run it. Again, this will take a second to process. And now we have a confirmation saying the gateway is up and running. The gateway is live. Okay, let's send the agent its first message from Telegram. Open Telegram and search for the username you gave your bot or go back to your conversation with the Botfather and click on the Telegram link directly to your bot. Then click start and you may get this message saying unknown command. That's normal. Hermes doesn't use a built-in {slash} start command, so it's just letting you know. Now we'll send a real first message. Let's type, "You're running on a hosting or VPS inside Docker. Tell me what tools and skills you currently have available." And submit it. We're doing this to give the agent its initial context so it knows where exactly it's running. And you can see here that it's typing. Really this means that it's thinking and it's just letting you know. It'll come back with a response in a minute. And we got a reply. It's a pretty long one. It's a list of all the different skills and everything that our agent has access to. Now, if you got a reply, your agent is working. That's the hard part done. If you didn't get a reply, don't worry. The three things that most often go wrong are an API key that didn't paste cleanly, a missing user ID in the allow list, or the gateway not starting properly. The quickest way to find out is to return to your terminal and type Hermes space doctor and submit it. It'll check your setup and point out what's wrong. And if that doesn't catch it, run Hermes space setup again and walk through the wizard. All right, from here we're going to make the agent do real work for you. The two things that make Hermes more than a regular chatbot are skills and memory. A skill is a saved workflow the agent can run on its own. Some come built in, others get created when you ask the agent to do something complex. The agent saves the steps so it can run the same workflow again without thinking it through from scratch. Memory is how the agent keeps track of things about you across separate conversations. I want to show you something that makes setting up your own agent worth the effort. The agent is going to do real research for you on the open web, save what it found to a file on your server, then automate the same job to run on a schedule. Now, I make YouTube videos. So, for me, I want to have my bot research the top five YouTube videos ranking right now for something like email marketing and then present me with a report. But, you can have your own bot research whatever you'd like. So, think about that now and then I'll show you a prompt you can run that you can modify to make it work better for your situation. Here, I've typed, "Find the top five YouTube videos ranking right now for email marketing tutorial for beginners. For each one, save the title, channel, and what it covers to a file. Then, tell me which gaps you noticed." We'll go ahead and hit send. And it decides to use the skill YouTube content. It notices that it doesn't have Chrome installed yet, so it's going to set it up for itself. And you'll see it's thinking output as it's working. The agent will spend a minute or two doing the research. Just let it work and when it's done, it'll reply with a summary in the chat and confirm where the file is saved. Now, if for some reason your agent doesn't have access to the tools it needs to run the task you've asked of it, it'll tell you and present you with options for what you can do together to make it so that it actually can do the job you've asked. For me, after a few minutes, it comes back and says where it saved the file here, opt/data/email-marketing-top-5-youtube.md. MD is just a markdown file, it's just a text file. So, here's the summary. It came back with five different YouTube videos. One from Alex Hormozi, HubSpot Marketing. Oh, and I guess uh my video here on the Mediakix Media channel is number three, so that's pretty cool, I guess. There's number four and number five, Jade Beeson and Life Marketing. It identified some big gaps of things that need to be covered in future videos if anyone were to make a video and it highlights all the different things, why it matters, bottom line, and asked if we want any further follow-up steps. So, let's pause on what the agent just did. It went out on the open web, opened five different pages, pulled the same structured information out of each one, made a judgment call about what topics they're all covering and what they're missing, and saved everything to a file on the server. The file is now sitting on your own server, not in someone else's cloud sandbox, not behind a connector to a third-party service, on the machine you control. It'll still be there the next time you talk to Hermes even in a completely different conversation. And in a minute, we're going to set the agent up to do this same research on its own on a schedule so the file builds up week over week without you doing anything. Now, let's save that whole workflow as a skill so the agent can run the same research on any topic without you having to type the long prompt again. So, now we'll say, "Save that workflow as a skill called competitor-watch. From now on, when I give you a search term, find the top five YouTube videos, save the same info to the file, and call out the gaps." We'll go ahead and submit that. All right, the agent confirms it saved the skill and tells us what the skill does. It lists the trigger, says you give me any search term, for example, SEO tips for SaaS, and I'll follow these steps here. There's four steps and includes a reusable script, a Python script. That's super cool. Now, one thing worth noting about skills, when the agent writes a skill from your own work like we just did, it's safe by default. But, if you ever install a skill that someone else wrote, Hermes will scan it for unsafe code first and you should still only install skills from sources you trust. All right, last piece, memory. The agent can save things about you and pull them into every future task. Let's give it a real preference that's going to shape how this research workflow actually works for you. You can tell your bot whatever you want about you. But for me, here's what I'll say. Remember that I make YouTube tutorial videos about software, mostly for complete beginners. When you analyze gaps in competitor research, frame them in terms of what would help a complete beginner specifically. We'll go ahead and send that. And it confirms that it saved that preference. Now, let's check its memory. Let's send it, what do you remember about me? And the agent will reply with what it has stored, including the niche description. It says that I'm a man, I make YouTube videos, it needs to frame the competitor gap analysis around beginner specific needs. And it even goes on to include details about the fact that this is Hermes agent running on a Hostinger VPS, that we're using Telegram, etc., etc. So now that preference is saved for every future conversation. The next time the agent does this research, it'll already know the additional context you provided. In my case, it'll know who I make videos for and the gap analysis will be framed around what beginners need. The longer you have Hermes, the more of this it builds up, the more context, the more knowledge, and the less you have to repeat yourself. One of the best features of Hermes is that it can do work for you on a schedule, not just when you ask. These are called scheduled jobs or cron jobs. The agent runs the job at the time you pick, even if you're asleep, even if your laptop's off, and sends you the result on Telegram. You don't need to write any code or schedule syntax. You just tell the agent in plain language what you want and when you want it. Let's automate the research flow we just set up. In Telegram, we can say something like, "Every Monday at 9:00 a.m. run my competitor watch skill on three search terms: email marketing tutorial for beginners, WordPress tutorial for beginners, and SEO tutorial for beginners. Send me a summary on Telegram with the gaps that stood out." Now, of course, modify the research task or the schedule task to be whatever it is you want done on a schedule. This is just my particular use case. The agent confirms the schedule and tells you when the job will run next. It may ask you to confirm your time zone. You can answer with yours. For example, I can say, "Let's use US Central Time." And we see that it was updated to run every Monday at 9:00 a.m. Central. The time zone's been saved, so other future jobs will default to it. To see all your scheduled jobs, you can simply ask. You can say something like, "Show me my scheduled jobs." And the agent will list them with their run time. Right now, it's just this Monday morning competitor watch. Now, if you want to see a sample of what that full output looks like, you can also ask the bot to run a job anytime. So, we could say something like, "Run that full job now." And it kicks off the cron job. This is a great way to review the output of the job to make sure you're getting everything in the right format and the way you want it presented. If it doesn't come back the way you want, you can simply tell the bot and it'll make the adjustments for the next time. Let's take a second here to acknowledge what we just set up. Every Monday at 9:00 a.m., this agent is going to research three topics on its own, save the findings to a file on the server, and push a summary to Telegram. All on the AI model that was picked earlier, DeepSeek. And that file is going to grow week over week, so you're going to start to get a really robust research picture. You can stack as many of these jobs as you want to. You could get a morning brief, a weekly report, a price check on something you're watching. Anything you can describe in a sentence, the agent can schedule. All right, there are two places to keep an eye on what all this costs you. First, your OpenRouter activity page for AI usage and your Hostinger dashboard for the server itself. Back in OpenRouter on your API keys page, you can see the usage so far for that specific key. You can also click activity on the left side under account and see your current spend, number of requests, and total number of tokens used and it breaks it up by hour in the day by default. You can also change it to different time periods. So, you can really get a good sense for what kinds of activities and requests are costing a lot of money. Your AI model is the single biggest lever on your bill. DeepSeek V4, the one we picked, is one of the cheaper reliable options. If you ever need more reasoning power for a hard task, in Telegram, send {slash} model as a command to switch to a different model. From there, you can click OpenRouter and get a list of all the models that are available to you. If you need more power, you can step up to something like Claude Opus 4.7. Just keep in mind that more power comes with more cost. Or if you want to step down to something cheaper, you can do that, too. For me, I'll go ahead and switch us back to DeepSeek since it's a great LLM for daily use. You can also tell the agent to use a specific model for a specific scheduled job. So, your routine work stays cheap while heavier tasks get the smarter model. Now, back in the Hostinger dashboard, you can keep track of your costs by clicking on your profile in the upper right and then clicking billing to see your past information as well as upcoming subscriptions and the renewal price. That's what you'll pay when the introductory term ends. I recommend checking the open router activity page every few days while you're just getting started. That way you can see your real usage and adjust the models you use before anything surprises you. Now, the research workflow we built is just one example. Hermes can do a lot more. The easiest way to find what's useful for you is to ask the agent directly. Send it something like, "What kind of tasks are you good for?" And then let it suggest ideas. Here it comes back with a list of what it's good at. Hands-on execution, long-running background work, scheduled recurring tasks, code generation and editing, research and synthesis, and memory across sessions. It also lists some stuff it's decent at and stuff it's not built for. All right, your agent is up and learning from how you use it. The first link in the description gets you an extra 10% off your VPS if you haven't grabbed it yet on top of whatever sale Hostinger's already running. Thanks for watching.