Hermes might be the closest thing to an autonomous AI employee that a non-coder, non-technical person can actually use. You can give it tools, message it on Telegram, and have it do tasks [clears throat] while you sleep. But, if you want to truly unlock the power of Hermes, you need to know what Hermes can see, what it can remember, and what you’re allowing it to do. So, that means understanding the models, context windows, sol.md files, memory, skills, hooks, gateways, sub-agents, and [music] everything else. So, in this video, I’ll break down every essential Hermes concept in plain English, starting with the basics, [music] and then working up to how the whole system fits together. And by the end, you’ll understand what all of these terms mean, and how they work together inside of Hermes to do actual work for your business. Okay, so let’s get into it. There’s a lot of different concepts that I’m going to break down in this video, but the first one is going to be the Hermes agent versus the underlying model. So, this is the difference between a harness and the model that’s actually the brain and the intelligence. Now, I covered this in my previous video, but the summary is this, that think of it like a car, right? The car has an engine, which can be thought of as the model, right? That’s powering the locomotion. Now, obviously, to move the car, you need the wheels, you need the frame, you need the chassis, you need the brakes, you need the electronics, right? That’s the harness. So, the model is the engine, everything else that makes the car move efficiently and effectively is the harness. So, we’re looking at the tools, the rules, the sessions, the memory. As you can see here, we’ve got a few of these diagrams to walk you through, and I’m also going to show you some stuff in Hermes where I think it’s going to be relevant. So, that’s the first thing that you need to understand. So, when you send a request into Hermes, it’s going to gather the instructions, the context, maybe it’s going to load some tools, and then it’s going to pass those to the model; and the model is then going to decide what to respond with, what to say, what actions to take. Now, because those two layers are separate, the harness and the model, you can actually change out the model for whatever specific task that you want to do. And because Hermes is open source and completely free, and you can download it here hermes-agent.newsresearch, it’s completely free. You can install it in your terminal with with this one-line command, or you can grab the desktop app as well. So, because those two layers are separate, you could change the model, you can use different models, you can hook up open router and use some of the open source models like Kimiko 3, which I’ll explain how I’m doing that in a moment. And you can also work on the harness, you can delete tools, you can add your own rules, you can add hooks and things like this. So, it’s fully customizable, unlike Codex or Claude Code, where the system prompt and some of the wiring underneath the hood is proprietary, and you just can’t change that. So, this gives you a bit more flexibility with what you can actually do, and you can customize the harness to your own use case. You can remove stuff that is not useful, because Hermes comes preloaded with a bunch of different tools. And for me, I had to remove a whole lot of them, because I’m not going to use them. They’re not useful for my use case. So, if we head over to Hermes, which I’ve got in the terminal here, and I do {forward slash} model, I can see all of the models that I’ve got access to here. So, I’ve got OpenAI Codex, so I’ve enabled my OpenAI ChatGPT subscription, so I can access the Codex models. I’ve hooked up OpenRouter with the API key. I’ve also got my Anthropic subscription plugged in here as well. So, if you want to set these up, you can do it in the config. So, the way that you would do that is you’d open up a fresh terminal window, and I would do Hermes setup, and then you can customize your whole setup here. So, you could select whatever model you want, and you can add your API key and your authentication. Over in the desktop app, it’s a little bit different. So, this is a desktop app, you can see I’ve got sessions, my chats from Telegram, cron jobs, Discord. We’re going to get into all of this in this video. Don’t worry, but if you just come to your settings tab, and you can find all your configurations here as well. Everything that you would find in the terminal when I put that command in for the setup, you can find here. And if you head over to providers, this is where you’re going to be able to connect other providers in here as well. So, that’s the difference between the harness and the model, and I’m just showing you how you can change your model, whether it’s Claude code, whether it’s Codex, whether you want to use one of the open-source models. And so, the cool thing with Hermes is that you can route different tasks to different models. So, this is the benefit I see of Hermes over Claude code. So, in Hermes, when I’m creating my content, I’ll plan that content with GPT 5.6 Saul, right? Which is a powerful model. It will create the plan, and then I’m going to write that plan with Kimiko 3 using open router. And so, I’ve set that up in my agents.md file to make that happen automatically. And so, the reason why I do that is cuz in my opinion and from some of the creators that have in my inner circle, we use Kimiko 3 for copywriting. It seems to be a bit better than Opus 5 and Fable 5, which honest, for me, have got worse when it comes to writing content. So, this is the kind of the new approach I’m trying out. Anyway, let’s move on to number three. The different interfaces that we can use to connect and talk with our Hermes agents. So, an interface is simply the place where you send your prompts and your work to Hermes and where you’re going to receive the answer. As I showed you earlier, the terminal is kind of a standard way to do it, but if you’re non-technical and you’re not used to working in the terminal, the desktop app is another way you can set this up. Now, I also have Hermes set up on Telegram. So, when I’m out and about, I can just message it through Telegram on my phone. Now, I’ve also mentioned the desktop app, which can show you a live preview, like an artifact of what you’ve actually built. So, that’s not something that you’re going to see in the terminal, and that’s one of the bigger features of Claude code on the desktop and Claude Co-Work. You can maybe build a landing page and you can see the artifact. Or maybe you create a presentation and you can actually see the artifact in the window. Now, one of the other features in the desktop app here is the voice conversation. So, I can have a backwards and forwards conversation with Hermes. So, I could speak to it and it can speak back to me. I don’t have to set up 11 Labs or anything like that. So, it’s going right now. I’m going to click stop, but if you could just reply to me with a few words just to confirm this is working. Pony Tail mode active for Pony Tail mode active. Which is something that I’m just installed. So, you can see it’s it’s talking back to me. Now, there is one other way that we can have Hermes have a conversation with us. And this is something I helped one of my clients set up at last week is in Discord. So, you can’t have the desktop app on your mobile phone. At least not yet. They are building that. So, you can see in my Discord I have my Hermes bot inside of here and I can add my Hermes to my voice channel. So, if I do {forward slash} voice, and you can see there’s some options. If I click here, I can go mode. We can do channel, join your voice channel. So, the way this works I need to go over to my voice channel I’ve joined and then I I need to put in the command voice mode join channel. Boom. My Hermes bot is in the channel. I can unmute it. Hey, can you just drop me a response um just to check that you’re listening in this channel. So, Hermes should speak back to me anytime now. You can see it’s transcribed my Mhm. Think yep, I’m here. Tom, could you loud and clear. Okay, so it’s talking back to us. Let’s move on. These are all the different interfaces you can use. We’ve also got Telegram like I mentioned. Uh you can use the API. There’s an SDK as well, but it’s getting a bit more advanced. Let’s move on to the next thing. So, let’s get into the fourth thing that I want to cover today and that’s prompts. You saw the MD file, your project rules. As you can see here, this all amounts to context. Now, context is the one thing that’s going to make your agent actually useful, right? Without any of this context, it’s just going to be a chatbot. It’s going to start from scratch every time. It’s not going to be particularly useful. So, the prompt is the request that you’re giving Hermes right now. And for me, I like to use WhisperFlow and just dictate what I want to happen. Maybe I’ll explain a project, give it a little bit of additional background context, maybe tell it where to look for specific files that I’ve already got created on my computer. Now, the soul that MD file is really the agent’s character, working style, voice, tone, communication. It tells Hermes what role to play and how direct it should be and how to make decisions. Now, your project instructions apply to a particular folder or workspace, and they hold the rules that Hermes should follow whenever it works inside of that folder. So, this is going to be your agents.md file. There are different files you can add in here pending on your own specific project. I often have a context.md file, and maybe if it’s a content folder, maybe a voice.md file. But, these can all be created by having a conversation with Hermes and asking, “What kind of context should we create so that you remember the project and you can create the best work possible?” Now, each layer has a different job, and the prompts covers the request, your project’s instructions, your agents.md file covers the workspace, and the soul that MD file is your agent’s general behavior. Now, before the model starts thinking, it’s going to grab all of these files. It’s also going to look for additional files and context that it thinks are relevant to the prompt that you’ve just submitted. Now, if Hermes keeps making the same mistake for you, you need to save this correction. You need to tell it what’s going wrong and how to fix it. Now, these fixes can live in a corrections.md file. You could add it to your agents.md file, totally up to you on how you want to structure this. But, if you want to see how I structure something called a second brain, which is a way that you can take all of your files, your documents, transcripts, all of that stuff, and distill down the key insights that your agent can then tap into. I’ve done a full video on that and how to set it up, all of that stuff. It’s on my YouTube channel. Um you should be able to find it. It’s called second brain and Hermes. So, if you want to set that up, you can go over there after you finish watching this video. And one thing I do want to mention before we carry on with this video is that if you want to run Hermes 24/7, you’re probably going to need to host it where a machine is running 24/7. Cuz as soon as you close your laptop, you’re not going to be able to interface through Telegram or Discord. So, having a 24/7 setup like a VPS is probably going to be your better option. Now, if you do want to use a VPS, I use Hostinger. They are the sponsor of today’s video. And if you go over to the link in the description below, uh you’re going to head over to this page and you’ll be able to see you can install Hermes agent in one click. If I go ahead and click buy now, I’m going to give you an extra 10%off with Growth Lab 10. So, if we pop that into the checkout, just type this in here, you see you’ve got a nice 10% off there. You can select the time period that you want. Now, by no means do you have to use Hostinger. There are a bunch of different other tools and ways that you can host this. It’s just the easiest option that I found. And you can also use the desktop app and and connect to Telegram on your desktop as well. But if you do want an agent that’s running 24/7 where you can run scheduled jobs overnight that’s going to take up your usage when you’re not actually using it during the day for your work, then this is probably the the better option for you to check out. So, again, the link below is in the description if you want to grab 10%off over there. So, let’s talk about context and compression. So, there is something called a context window. And different models have different size context windows. And so, once you fill that up, and that’s going to get filled up by tool calls, responses, and your inputs and outputs. Now, when you start approaching the upper limit of your context window, so in code code, you’re typically going to have a million context window token limit. Whereas GPT 5.6 soul, if you’re using Codex and the subscription, it’s going to be like a 256,000 token context window. And for me, that fills up super fast. So, there’s a few ways that you can handle this. Now, in Hermes, you can actually set the limit where the session is going to compress and compact. What I mean by that is when you hit that limit, Hermes is going to compress and summarize and compact your whole conversation. It’s going to basically take notes on the important things that have happened. Then it’s going to insert that into the context window of a fresh session. Okay? Now, that’s one way of doing it. Now, what I would often do is create a handoff document when the context window gets about 50 60%full when I’m using GPT 5.6. And if I was using an Anthropic model, I would compact around 40%of the context window. Because once you fill that context window up till it’s nearly full, it’s going to have problems referencing all of the context and everything that you’ve worked on in that session. It’s not going to reference everything accurately. So, so when you do hit 40 50 60%of your context window, you’re going to want to create a handoff document, clear the session, or create a new session, and then load in that handoff document and continue your conversation. So, you could see here in Hermes in the terminal, we do have this bar at the bottom, and this is showing us how full the context window is, how many tokens we’ve used in this context window, and what the total size is. Okay, so I’m going to correct myself. 272,000 tokens typically in a GPT Codex model context window. Okay, so next, let’s talk about the agent loop. This is how all of your prompts and requests are going to work. You put in your prompt or or your request, Hermes adds that to the context window, and then it’s going to think about what needs to happen next. Do we need to run some tool calls? Do we need to do a web search? Do we need to tap into an MCP? Do we need to check some files in your folder that we’re working in? It’s going to take all of those actions, and then it’s going to start giving you the response. Now, what happens when we’re building something, we want to go through a few different phases. We want to go through the planning phase, which is going to involve searching of files, doing a bunch of research, and then coming up with the plan, and then we’re going to go into the build phase. So, at this point, Hermes will want to ask you a bunch of questions about the build, so you can finalize the plan that will then get pushed into the build stage. Once you’re building, Hermes can then run an autonomous loop and keep checking the work until it hits the goal of your plan, right? So, that’s one of the really cool things about having an agent build things for you. You don’t have to keep prompting it once you’ve created a plan and a goal of what that end state should look like. Okay, so let’s talk about tools for a second because this is what makes Hermes incredibly powerful, and also any of the other harnesses that you use, because most likely you’re going to have MCPs and skills that are all built into your workflows and the way that you use your agent. And if you don’t, those are probably going to be some of the first things that you install. Now, Hermes does come with a bunch of pre-built tools, which I’m going to show you in a second. But essentially, a tool gives Hermes an action that it can take outside the model. So, that could be searching the web, it could be running a command, it could be calling a specific service. So, let me give you an example. Now, I write my newsletter inside of Hermes, right? I’ve got a few different skills that I pull from, and it’s got additional tools, right? So, it can search the web, it can go over to my YouTube channel, it can tap into my Instagram and pull posts and links that I can put in my newsletter. And so, once all of that has been completed, I can then push that to Beehiiv, which is where I’m sending my newsletter from. I don’t actually need to copy and paste anything over to Bee hive because there’s an MCP, right? And that MCP gives the harness a set of tools. And so, you can think of an MCP like an API call, right? API has a bunch of different end points and the MCP creates a tool for each of the different end points, right? So, you’ve got a toolbox and it’s like, okay, am I going to use a wrench? Am I going to use a spanner? Am I going to use the screwdriver? And each one has a different function for whatever it is that you’re trying to do. Now, if we come into Hermes in the desktop app, we can go over to our settings and we can have a look down here, tools and keys. So, we could see there’s a bunch of different tools here that have come either built in or ones that I’ve installed. And these need API keys to function. Now, if we click on one of these, we can see what this actually does. We can we can get a key by clicking on the link, go over to the website and decide if this is a tool that we want to use. Now, what I will say is there is a lot of stuff in here that you’re probably not going to use. So, for me, when I set up Hermes and I was monitoring my context window, for me it was filling up really fast. It would go to like 15, 20%immediately. So, I had to do a analysis, run a audit to ask Hermes, what are you loading when I’m sending the first prompt? Is are you loading up a bunch of tools or skills? Is there any hooks involved? Because there might be a bunch of stuff that’s getting loaded that I’m never going to use. And so, it’s just taking up context. It’s taking up tokens that I can then free up because I’m not going to use that tool. So, there is another thing called plugins and these are similar to tools. Now, a plugin is something you can grab from GitHub. Now, you can see here I’ve installed a few things like ponytail which you saw that it was referenced earlier in the chat. Now, a plugin is typically a collection of skills that maybe links up to an API key through an MCP. And these plugins are put together so that the skills can work synergistically with each other. Each skill will have a specific role. So in the superpowers plugin for example, which I’ve spoken about a lot on this channel, it does have like 11 different skills. So the plugin allows you to install those with just one click. And you can see here, MCP servers, hooks, and slash commands, skills, and bundles, plugins bundle all of those together. Okay, let’s move on to number eight, memory and skills. So memory stores useful information that Hermes may need in a future conversation such as your preferences, previous decisions, facts about your business and your work. A skill stores a repeatable method for doing a specific job. So it tells Hermes which steps, rules, or resources to use when that type of work comes up. So memory answers the question, what should Hermes remember? So when you start up a session, it’s going to check your memory.md file, and then it’s going to carry that information into your context window to use alongside of your prompt if it thinks that that’s relevant. And a skill answers the question, how should Hermes do this thing? So it gives the agent a process that it can follow over and over again. So neither one of these changes the underlying model. Hermes loads the relevant memory or skill into the working context window when it’s needed. Now you don’t have to write every skill yourself. Hermes can install skills that other people have shared. You can even create your own by doing {forward slash} learn, and Hermes will basically create the skill of a process that you’ve just gone through in the chat. The other cool thing about Hermes is that it will automatically create skills for you based on whatever it is that you’re working on. Now I don’t have a demo for that, but you’ll see in Hermes when you’re working, you can see it’s applying this principle of self-learning, and it will create a skill. It will show you a dialog box saying self-learning skill created, something like this. Now I have done a full video on memory versus skills. It is on my channel. Should be able to find it. It’s quite recent. Uh so if you want to do a deep dive on those two things, in my opinion, the most important things inside of Hermes are skills and memory. Now, now memory isn’t just a memory.md file. That can only contain a a small amount of information. That’s why I’ve created a a search brain. That’s why I also search system that vectorizes all my files and makes them searchable. So whenever I want to recall some information, all of my files are in that mini vector database. And that information can be pulled when there’s a relevant prompt or request put into my agent. So number nine, let’s talk about hooks. Maybe you’ve never heard of these before. They’re kind of a bit under the radar, but in my opinion, they can be extremely powerful if you apply them in the right place and you know what they actually do. So a hook is an action that runs by itself automatically when something specific happens inside of Hermes. So it’s up to you to pick when that moment happens. It could be before tool usage, after a job finishes, when a new session starts, when you write a file to your system. When that moment comes around, the hook fires automatically every time. And it does that in a shell. It’s not a prompt that gets injected into the system. It’s, let’s say, a terminal command that happens in code. Now the difference between a hook and a skill is that a skill is invoked typically when you’re wanting to run a specific process. Now your agent can invoke the skill when it thinks that you need it, but the hook is going to happen automatically. Now as you can see on screen, this is how a hook works. So the moment that the hook is called, that could be writing to a specific folder or file, the hook checks your rules, right? So you can create rules of whether when it’s going to fire and when it’s not. And if the rules are broken, it’s going to get blocked. If it passes, that hook is going to run. Now you don’t need to be a programmer to set these up. You can just talk to Hermes and say, “Hey, can we create this hook so this process runs every time automatically without taking up any of the context. Remember this thing, this project, this information about me. A skill is when you want something done in a specific way, a process, and a hook is basically something that always happens when something else happens. So, when X happens, do Y automatically without me approving it or anything like that. Okay, let’s move on to number 10, profiles. So, you can think of these like individual agents. A profile is a a set Hermes setup built for a particular role, person, or type of work. So, each profile can have its own agents.md file, soul.md file, memory, skills, configuration, and conversation history. It can have its own models, its own tools, everything. So, that means, for example, as you can see on screen, you can have a research profile that can behave differently from, let’s say, an operations profile without mixing the instructions together. So, you can specialize an agent in a specific thing so that when you come to do that thing, so I could have a content creation agent that does that specific thing really, really well. It’s got access to all the context, all the previous content that I’ve created. It’s It’s got access to MCPs that are going to pull specific posts that are performing the best on X or on LinkedIn or on YouTube, for example. Whereas, if I had everything in a general agent, but if I had to do research, operations, content creation all in the main agent, listen, it’s possible, but better to have a specialist agent. And these can come in really handy when we get into agent orchestration and the Kanban, which I’m going to get onto a little bit later in this video. So, each profile is essentially an entire Hermes instance, and you can have all of your sessions underneath each individual profile. Now, these are different to sub agents. A sub agent is a temporary helper that’s created when you’re working on a project. Now, you’ll be able to see inside of the dialogue window when it decides to spin up a sub agent. So, for example, you could be working on something and say,”Hey, can you read 10 different blog posts and summarize the findings of each of those?”It can spin up 10 different sub agents, read 10 different blog posts, summarize those, and bring the findings back. That’s what sub agents are used for. Profiles are useful for separating roles and context, but they’re not a security boundary by themselves. Those kind of boundaries come from tools, permissions, and where Hermes is running. So, let me show you an example. If I come over to Hermes Desktop, and I can see at the bottom, you see I’ve got a bunch of different profiles. So, if we click on SEO Writer, for example, you could see these are all the sessions associated with this particular agent. You could see the cron jobs that I’ve set up for this agent, and this all ties together inside of my Kanban board, where I have my AI Search, I have SEO Editor, SEO Orchestrator, SEO Strategist, SEO Writer, and all of these agents are working together. They pass tasks between each other, which is orchestrated by the orchestration agent inside a Kanban board. Now, you can click manage profiles here, and this is where you can add additional profiles, and each one of these, as you can see, comes with a soldier.md file. Now, we can add additional context files in here where necessary, and you can add your own profiles just by clicking here. You can name it, you can clone it from a different profile as well, and then you can fill out your soldier.md file, which I would highly recommend for each of your agents. You can also add in context.md files, agents.md file inside of each of these profiles. It’s just not listed here in the interface, but it’s We’ll just know this is something that you can do when you’re setting these up, and it’s not complicated. You just have a conversation here with Hermes, and it will go ahead and set all of this up for you. Okay, so number 11, let’s talk briefly about the messaging gateways. Now, I have already covered these earlier in the video, so I’m not going to go into these in too much depth, but there are different platforms where you can access your Hermes agent. You can have it live inside of Slack, inside Discord, like I showed you, and inside of Telegram. So, you can see I’ve got my agent here inside of Telegram. This is something I can access on my phone. This is connected through the VPS, and then the VPS is then connected to my computer via Tailscale. And this is running cron jobs, is pulling in research every day to give me ideas for X posts. For example, so as you can see, these messaging gateways connect into Hermes. You can have different gateways for different agents. You can switch the agent profile inside of the messaging gateway. So, it’s totally up to you how you configure this. Some people like to have Hermes in Slack. Some people use Discord, Telegram, or you can just use the desktop app. They are bringing out a mobile app pretty soon. And then we’ve also got the classic terminal interface as well. Now, if you want to know how to set these up, you can literally just ask Hermes,”Hey, how do I set up Hermes inside of Slack?”And typically what this is going to look like is you’ll need to set up a Slack application. It’s then going to give you the authentication and the keys that you need to integrate Hermes into your Slack, and it’s going to create a bot, and that bot is then going to be active in whatever channels you want to add it to. So, yeah, you can see here we need to create an application, give it the permissions, and we can then go into the setup gateway setup, Hermes gateway setup in your terminal, and you can add in the configuration keys and get all of that set up. But this video isn’t designed to give you the complete setup tutorial, but you can definitely go through and set that up yourself if that’s something you want to do. So, you might be watching this video thinking,”Well, this all sounds great, Tom, but I don’t have the time, I don’t have the experience and the expertise to set this up properly.”You might be thinking,”I’m probably just going to spend a ton of time trying to figure this out, when in fact I could have somebody help me build this for me.”So, if you do want me to help you build this for you, and build out all of the different agents for the processes in your business, the first step is to book your free 30-minute AI audit call. On that call, I’ll figure out what your business problems are and which is the highest ROI automation opportunity to build an agent for in your business. And then we can look at the options of how you want to proceed and get that done, whether you want me to build it for you, whether we consult together, or I show you and teach you over a period of 90 days to be able to do it for yourself and implement it in your company, whether that’s yourself or somebody on your team. There is going to be a link below in the description if you want to do you want to book that call in. It is completely free. You’re going to speak directly to me for half an hour. I’m going to schedule that out in my calendar, so it’s important that if you are serious about this, that you show up to the call. And usually for my one-on-one consulting clients, I charge a thousand bucks an hour. So, you’re basically getting $500 of value for 30 minutes of my time. So, anyway, I’d love to speak to you if this is something that you are interested in. Anyway, back to the video. So, the 12th thing that I want to talk about is the different ways Hermes can run. Now, I have mentioned this briefly again in this video, but we’ll just dive into this a little bit deeper because you do have local methods, you’ve got VPS, you’ve got cloud. So, let’s just dig into these real quick. Now, you can run Hermes with Telegram or Discord or Slack, but you can connect that up in different ways. You can connect it to your local device, your laptop, what I’m recording on right now. You can connect it up there. You can also connect it up to a VPS. So, like I’ve been running on my Telegram, it connects to Hermes on a VPS. The VPS then connects to my computer using Tailscale. But it is important, in my opinion, to know the difference and the capabilities of each because if you run a team and you want to have multiple employees tapping into Hermes, build their own agents for their own process and roles, then you might want to set up that includes a VPS or some kind of way where all of your files and contacts are stored in the cloud. There’s a really cool tool called HQ for work. And that’s where you’re going to have a platform that contains all of your contacts, all of your tools, all of your MCPs, your API keys, and all of that. And your team can literally just plug in and access those through their device using the cloud. So, there are different ways of doing this. You’ve just got to think about your own use case and who’s going to be using Hermes inside of your business, and where do you want those files? Do you want your team members to have their own files on their own computer? There’s certainly some benefits to that, but also the drawbacks is there’s no shared knowledge base. And I think if you’re working in a company with multiple employees in a big team, like one of my clients, Kyle, which there’s a case study on this video. He has a company of over 150 employees. So, in that case, it’d be pretty smart to have a second brain perhaps for each department that’s shared across their team. So, everybody when they’re building and using Hermes in the business, they can access all of the same context. If you have Hermes installed in the cloud with all your context and your API keys and your tools and your MCPs, everybody on the team is going to have access to that. If everything is running locally on somebody’s computer, then John who’s in operations is not going to access Steve in accounts. They’re both not going to have access to the same MCP or API key. They’re both going to have to set that up independently on their own devices. So, number 13, let’s talk about scheduled tasks and automated workflows. This is where agents like Hermes can be incredibly powerful. Now, this is not native to Hermes by any means. You can do this in Cloud Code and Code X, but a scheduled task lets Hermes begin work at a specific time without waiting for you to send a message. So, that schedule, as you can see on the screen, can run in a loop, right? It can run once. It can repeat at specific times of day on different days. And that interval is totally up to you. So, that scheduled job runs without you, which means the job needs enough context and instructions to stand on its own. So, an effective scheduled job might run a specific skill which contains all of the instructions and the context and maybe hooks up to different MCPs where it can pull additional information and research to actually complete the job. A great example of this is pulling your reports. So, let’s say you have data in Facebook Ads, your Google Search Console, maybe your email analytics. You can have MCPs pull all of that data and this skill can pull all that data, put it into a report, send that via email to whoever needs to see that report for Monday morning’s meeting, for example. Now, running a scheduled task like this we we also call it a cron job. That’s just one way to work. Hermes can also run a fixed script or respond to outside triggers when you connect. So, this is called also called a webhook, which you might have heard about. Now, if if a tools pings a webhook, you can ping that webhook to Hermes to trigger a specific skill or task or whatever it is that you want Hermes to complete. So, let’s say for example, you have a funnel and somebody hits the thank you page. You can create a webhook to ping when someone hits that thank you page and ping, let’s say, a message through to Slack or maybe into WhatsApp to notify you that somebody has hit the thank you page and you’ve got a new application coming. Now, typically the difference between running a script and running a scheduled job, I mean, you can have a scheduled job run a script. Let’s say it’s running a skill, that skill is going to be dynamic and it’s going to be able to do what you would typically do in a session if you run that skill. Whereas a script is more deterministic. It’s going to run a piece of code, it’s going to do a specific thing and that’s it. So, that could be good, for example, with accounting where you’re just moving numbers around, you’re making reports and spreadsheets and so on. So, let me give you an example. You could see here on my Hermes desktop app, I can see my cron jobs down at the bottom here. See I’ve got my daily AI insights clip library. And so, if we click on here, so we can see all of the different runs that have happened and if we want to change, we can go into manage. We could change the settings here. We do have a prompt and if we take a look at a prompt, we could see what’s actually running. So, I’m not going to go through everything here, but you can see it’s defined the workspace, the maximum run time, and it’s got the forced skills. This is what’s actually going to be doing the heavy lifting, the long-form video clipping pipeline. So, what’s happening with this skill and this schedule job is I put out posts on X at 9:00 a.m. Eastern every day. This skill is then going to run like you can see it run at 9:02 a.m. And so, it’s going to grab my X post, it’s then going to search on YouTube for relevant podcasts and interviews with X experts on the same topic. It’s then going to grab clips from those interviews, and it’s then going to send me those clips over on Telegram, and it’s going to add them into a folder on my computer. So, you can see here I’ve got this folder AI Insights clips, so you can see these are all of the clips that it’s pulled. I’ve got a clip library down here, and you can see, you know, I’ve got different clips from different podcasts, and I can include those in my next post. Let’s say I want to retweet the post that I put out the previous day, I can then add this video to see if that gets any more engagement and more reach, because video right now on X is performing really well. Now, if you do want to test your schedule job, you can literally just trigger it right now if you want to run it manually, but that’s how you set all of this up. I didn’t write this prompt at all, I got Hermes to write this after I went through the entire workflow in one of my conversations. I said,”Okay, let’s now schedule this as a cron job, scheduled task, and let’s create skills that you can use to run this process efficiently.”Number 14, sub-agents versus Kanban and agent agent coordination or multi-agent orchestration. Now, I already gave you a definition of sub-agents. These get spun up temporarily inside a main agent chat in a session. It spins up the sub-agents, they do something, they pull the information back, and they feed it to the main agent. Now, the Kanban is very different. It’s like a Trello board, and I’m going to show you it in just a second. It’s going to move tasks along the pipeline, and at each stage there’s going to be a different agent involved. It’s going to assign that agent to a specific task that it’s been designed to help you with. So, let me show you how this works. So, what you’re seeing on screen right now is a custom gateway, very similar to the desktop app, and I’ve set this up on my local device. Now, there are a lot of different options in here. You’ve got the chat, which is This is what appears in the chat window. You’ve got your sessions. You can tap into previous sessions in here. Files, models, logs, cron, schedule jobs, skills, MCPs, all of that stuff. But, the thing that I wanted to show you that I was talking about, remember the Kanban? If you click into here, you can see I’ve got a couple of different boards. Now, a Kanban works where it’s passing tasks through different stages of a pipeline. So, here, my SEO article pipeline has something ready for me to review. You could see it needs an assignee, and this is for me to take a look and manually review this post. You could see we’ve got a description. We’ve got a list of all the files that it’s used and created. So, I can come in here, and I can check out one of these files, for example, and I can download it. Obviously, it is already on my device. So, I can basically just search for this file in my folders. I can add comments. I can say,”Approved. Move to next step.” And so, this is going to get picked up outside of a session. So, I don’t need any sessions running to run these All these agents are running independently of the main Hermeus chat, the sessions, and the agents that I have set up. These are all going to work in the background and pass work between each other to complete this pipeline that I’ve defined already. So, you can see here, this is everything that’s happened in this particular pipeline. I’ve also got one set up for my AI insights clip. You can see nothing’s running right now because it runs once a day. You could see all of the previous tasks here. So, this is a bit more of an advanced concept, but for me, this is how I visualize multi-agent orchestration. There are tasks that get passed through different stages of a pipeline that different agents are involved in that I don’t have to interact with unless there is an approval step for me to get involved with. Okay, let’s move on to number 15, the security layers. So, we’re talking permissions, approvals, and sandboxing. So, Hermes has several security controls because no single setting can cover every different type of risk. User authorization controls who can send requests to the agents. The tool set controls which actions the agent is able to request. An approval can add a checkpoint, so Hermes can pause before a risky action can be taken and waits for you to either allow or reject that. A hook, like we spoke about earlier, can actually guard the same moment automatically. So, an approval waits for your answer while a hook applies a rule that you set in advance and can block that action on its own. So, for example, if you have a hook set up that prevents your AI agent from reading your .env file in the conversation, then that’s going to happen automatically. And that’s that is a security layer that you can set up, and it’s what I have set up so that my agents don’t put API keys in the chat. Now, it can run bash commands and terminal commands to pull that information into API calls, but it’s not going to put that information inside of the chat, if that makes sense. So, in my opinion, the safest setup is to give each of your agents or your profiles inside of Hermes specific permissions and specific tools and files that they can access, and then keep everything else outside of its reach. Now, like in Claude Code, you can have different approval settings where you either manually accept all of the edits and changes that the agent wants to make, or you can put it into auto mode, or have your own separate rules according to the different types of tasks and tools that it calls, whether it’s going to ask you for permission or not. So, listen, I have gone through a lot there, and there is a whole bunch more information on the Hermes documentation site, including stories and case studies and a bunch of good stuff. But, I wanted to pull out the key concepts to explain in this video. I know it’s been a long one, and if you’ve got to this point in the video, drop me a comment below cuz I’d love to see who’s actually stuck with me on this. This is all important information to know and understand and internalize if you’re going to be building with Hermes or indeed any agent because these principles can apply across the different harnesses like Codex code code for example. So now you understand how the main parts of Hermes work together, but the next step is using them to solve a real problem in your business.