My FULL Hermes Agent Setup (HermesOS)

The car is away. Of course the car is away. You have parked on a behind-the-back. What is the plan? Alex and his father are in a group of young children with disappointment, which break up just in the holiday. There they find more than just a障礼 and experience the summer of their lives. You're a nice guy, though. You're also a nice guy. The Some-It-Was. The Feel-Good-Comödie of the Year. From the 3rd September only in the cinema. My Hermes OS. It is a multi-agent system that autonomously builds software, acts as COO of my company, assistant, and my personal productivity and health coach too. Honestly, I love Hermes. I use it every single day and I have so much fun building new functionalities. So I hope this video gives you some ideas to build yours too. Now, without further ado, let's go. A portion of this video is sponsored by Hostinger. So first off, I want to show you Coder, a multi-agent software building system. So let's actually build something right now. This is my Discord where I talk to my Hermes mod. So I can say Hermes mod. Draft me a spec to add onto my Pomodoro app the ability to switch skins. So here's my little desktop Pomodoro app. I can just put in the time, label it, start the timer. And once it's done, it shows up in my Obsidian Vault. I think it's already very cute with this little avocado, but it would be even cuter if it had the ability to switch skins. Right now, it's only an avocado. Please create three other skins it can change to. For example, cat, dog, tomato. In the same style. The user should be able to switch skins easily. Keep everything else the same. And click enter. And then a few minutes later, we see that Hermes mod does its thing. It comes up with a Pomodoro skin switcher PRD. Tells me the goal is to add a skin switching system to the Pixel Pomodoro Electron app. So the user can pick between existing pixel arts with three different skins. And it goes on with a lot of details about the approach for doing this. And I think it looks pretty good. So I'm going to say looks good. Let's build it. And once it's done, it should show up in my build channel. So we'll let this build for now. And after it finishes, I promise I'll explain a little bit more about how this is all working under the hood as well. But in the meantime, I want to show you another functionality in my Hermes setup called Lifebot. So this is Lifebot. Now Lifebot, I talk to using Telegram. And as its name suggests, it is my productivity and health life coach. So I can ask it, how is my productivity so far today? It's running some terminal commands. Cool. And it tells me I've had three sessions today of 27 minutes. So film prep, run script, and it also logged a couple of test Pomodoros that I did. Can also ask it, how many steps have I done today? And it says no data yet, but the iCloud Placestorder hasn't downloaded locally. Okay, fine. I sync my steps from my phone every two hours to Lifebot. So hasn't done that yet. That's okay. You know what? Let's just say, I'm going to say, what about yesterday? How many steps? It said I took 11,502 steps. Strong for a Sunday. Above my average. Nice. So this is all great. You know, it's just like telling me information, right? But what is really valuable is actually hidden over here. You see, it has all this data that's being logged into it. And it's able to give me custom advice. I can ask it questions like, what's my average amount of log time right now? And what do you think I can do to help boost the amount of focus time I work? Like, should I consider switching around my times for when I do certain tasks? I'll take more breaks. And you can actually tell me like what the pattern looks like. Your strongest Windows by session count is around like 12 p.m. That's when I'm doing mostly tasks, organizing light work. Around 5 p.m. I'm doing mixed development and tasks. Around 7 p.m. I'm doing deep development, scripting and research and etc, etc, right? And it's telling me that the gap that I have is actually 9 a.m. to 11 a.m. It's saying that you're using 9 a.m. hour exactly once. And it produced two solid sessions. So it tells me to try out that 9 to 11 a.m. slot to try to get more deep work in earlier during the day. And it says to don't overthink breaks. TLDR, you don't need more breaks for longer sessions. You need one more session per day, ideally up in the 9 to 11 a.m. dead zone. Or by protecting the 3 p.m. to 5 p.m. from task creep. That alone will push you from 4 to 5 sessions to 5 to 6 and from around 90 minutes to 110 plus minutes per day of deep work. Anyways, I've dived a lot deeper into this with LiveBot as well. But I hope this gives you a good idea about how powerful it can be when you can get custom advice based upon the patterns that you have. As I'm working as well, you know, there's sometimes I'm just like, man, like I'm so tired. What should I do? Like, should I like take a break right now? Or should I like keep pushing through, right? So I can ask something like, hey, I feel a bit tired. Should I take a break right now? Or should I push through another Commodoro, for example? And it's telling me it's like 11.06 a.m. You've done one real session and two one-minute tests. So that's like 27 minutes in total. So you haven't actually logged a single break yet. It tells me that my strongest window at 12 p.m. opens in less than an hour. So you're only at 27% of your typical an hour and 40 minute day of deep work. So the data says to push. But if you do break, make it 10 minutes, no more, no less. Because we went through this before. If I take breaks that are like five minutes, it's basically just procrastination. And if I take breaks that are 15 minutes long, I just like take the break, leave and just like never come back. Oh, cool. It also tells me that I has decided to do a self-improvement review. So that's another like Hermes specialty, which if you don't know what that means, definitely check out the video, which I'll link over here, which is the fundamentals video. Anyway, I hope this little demo showcases how useful it is to have something like LifeBot. I can really say it has honestly genuinely changed my life. Now let's talk about how LifeBot actually works under the hood. So LifeBot is an agent that is part of my entire multi-agent Hermes setup. And its job is to use data in order to give me life coaching advice, whether that be health goals or productivity goals. Of course, as you may have guessed, how good LifeBot is depends on the data it actually has. And right now it has three primary sources of data. The Pomodoro logs I showed you earlier, whenever I do Pomodoros, I use my little desktop app to track how many minutes I'm doing and what I'm doing during that time. And all of that gets stored locally into this Obsidian vault. So here's some of my logs. The second source of data is this typewriter desktop app, which I also custom build. And on here, I put like my to-do list, like today's video production day, also different study plans. And then I can like check things off as I do them. And they're also tracked over here in Obsidian with timestamps. And finally, the third source of data is from my Apple Health. I use iCloud in order to sync health data from my phone. Right now, I mostly only just track the steps, but I have been thinking of whether I should get like an Apple Watch or something so I can have more data as well. Let me know in the comments if you have an Apple Watch. Do you think it's worth getting? Because there are like a lot of health metrics that I could be tracking. But anyways, those are my three sources of data. And my Hermes LifeBot is able to access all of this data. And that's how it's able to analyze it. And it gives me very custom feedback based upon the data about me. Oh, and right now, the model that's powering LifeBot is DeepSeek V4 Pro. Although I do change that model out, like that's a hosted version from the cloud. But sometimes I also switch to a local model that I have on my Mac Studio 2. Anyways, I don't want to go into way too much more detail about this and exactly how I built it because that would be like a really, really long video. But I'm going to put a summary of LifeBot's specs on screen right now. So please do take a screenshot. This, as usual, would of course also be in the free guide linked in the description. And you know what? I have actually been thinking whether I should do like an entire Hermes workshop because I genuinely think Hermes is so useful. So if you're interested in something like that, where I'll actually explain how to actually build up these functionalities and multi-agent systems, do comment Hermes workshop because I actually do monitor all the comments. And if I see a lot of interest, then I will consider it. Anyways, let's actually go back and see if Coder, our software building agent, has finished building a new skins for our Pomodoro app. Okay, cool. So it says build successful and it gives me a bunch of information about it. That's great. Let's actually see what has built. Okay, so go on terminal and then let's see. Okay, and let's do NPM start. And here we go. We have our little Pomodoro here. Okay, great. So it's over here. We'll do one minute. We'll do a test again. Test skins. Start. And there's a button here. So here's the avocado. Drum roll, please. Cat. Hey! It's a cat. And dog. Hey, it's a dog. And tomato. There's a tomato. That tomato looks a little bit creepy, but that's okay. Yeah, so it did what it said it's going to do and now it has the ability to have different skins. So I don't know about the aesthetics, especially the tomato over here. Kind of looks a bit creepy. So probably going to need to like modify the aesthetics a little bit. Timers up. There you have it. Now let me explain how it actually works under the hood. So Coder is another agent in my Hermes multi-agent setup and it's specifically responsible for building software. So in terms of how it fits together, the way I communicated with it is through Discord as a Hermes bot. Now this Hermes bot itself is powered by a local model on my Mac Studio called the Quen 3.635B model. Oh, specifically the Quen 3.635B A3B model. I can see it's a very popular model with 5.4 million downloads just last month. Anyways, I run this model locally and it's honestly a pretty good model. But because I do pay for a Clod Max subscription and I actually want to like use all of those tokens, right? What I actually do is I use this model to go and poke Clod code and for Clod code to actually write up the entire spec of what I'm trying to build. And it's actually Clod code that actually goes and builds everything. Then when it's done, it tells my local Quen model who then goes and tells me that the thing is done building. It is kind of complicated, but it's a workaround because Anthoppy doesn't allow you to use your Clod subscription directly. So that's why I gotta like do it this way. But hey, it works out. I do sometimes just power Hermes with other models as well, like Jupiter models, Gemini models, and GLM and DeepSeq. And that makes it a lot easier. I'm going to put all of that on screen right now, along with summary specs of decoder agent. Oh yes, just to mention the way I'm coordinating all of these multi-agents is with the Hermes combat board. So yes, I saw in the comments of my past video about what to do if you want to run Hermes 24-7, but you also don't have a dedicated machine. Well, fret not, because you can also rent it on a VPS, a virtual private server, which is kind of like renting a machine on the cloud. I actually also run another instance of Hermes on our team's VPS so that everybody else on the team can also play around and build on top of it too. And that VPS is through Hostinger, who are very kindly sponsoring this portion of the video. So just like a local machine, Hostinger keeps your API keys, data, and logs all on your own private infrastructure. And if you ever make like a config mistake, you can just roll it back because weekly backups are also included by default. And guess what? Hostinger actually has a pre-made Hermes template already. So you can go to hostinger.com slash Tina Hermes. Under the Docker catalog, there is the pre-built Hermes template pre-selected. So just click deploy. And that's it. You get instant CLI access without any manual configs. And since the entire VPS is yours, it's not per agent fees. You can run as many agents as your resources allow. Our team VPS, for example, runs Hermes agents and like six other internal tools too. So if you want a 24-7 Hermes agent, but don't want to deal with the hardware, you can use my link hostinger.com slash Tina Hermes and apply my code TinaHuang for an exclusive discount on the Hermes agent VPS plan. Link is in the description. Thank you so much, Hostinger, for sponsoring this portion of the video. Now back to the video. And with that, I want to move on to TacoBot, our Lonely Octopus COO agent and assistant. Lonely Octopus is the name of my company, by the way, in case you didn't know. All right, let's go on to Lonely Octopus team. So here is the Lonely Octopus Discord server, where all our Lonely Octopus team hangs out. And as you can see, there are many different channels. So TacoBot can actually do a lot of things. I will show you some of them. This one is one of my favorite functions. For example, I'm having this conversation with Ibrahim from our team about this teachable workshop that I'm doing. Hope you guys went to that one because you probably has already passed by the time you see this video. But I can say, hey, TacoBot, remind me at 9pm today to decide which demos I will do for the workshop. And then TacoBot does its thing. And it has set a reminder for 9pm today. Amazing. Also, look, TacoBot is a takoyaki. Anyways, I love this function. I use it so much. Like here, TacoBot is reminding me to review the deck for a timer that I set earlier. And then I can also reschedule it because we have so much stuff going on this Discord as well. I can also say go into this dev channel and ask it at TacoBot. What are people working on in the past week? And TacoBot is able to scan this Discord channel. And it's also linked to our GitHub as well. It's able to tell me that, okay, Rex is working on these different things. Internal tool side, Cothra is working on this. Jonathan is working on that. And then what's happening on newsletter as well. What's Jovi working on, etc. Really handy for me as a manager. And the final thing I want to show you is on the COO side. You see, the COO position for a startup, at least a startup like us, has always been the hardest thing to fill because it involves having to document things and make things sustainable while also operating really quickly at the same time. And honestly, I've been struggling to fill a COO position of this company for like years now until I realized that TacoBot could do this. For example, I can ask it, I can say, TacoBot, draft me an ops guide for how our team releases internal software. All the steps that we follow and sanity checks. And TacoBot is able to gather information from this Discord. And it also has access to our Notion where we have so much data stored there as well as our Slack, which we use for external contractors and external people we work with. So it's able to look through all of this to be able to draft up these ops guides. So you can see here that it specifically is searching through Notion right now. And we're able to know that, say, for example, pre-release checklist and sanity checks. So we know that for code and quality checks, we need to have at least one peer removed, minimum one peer approval to GitHub, peer against main, automated tests, all units integrations, blah, blah, blah, etc., etc., right? And release execution protocol. So this is the way that we execute things. So if anybody wants to know, they can just ask TacoBot. And we have new people who are going to be onboarding into our team. We can also give them access to TacoBot so they can learn how we do things. I hope you can see how powerful this system is, especially times 10 because that's the size of our core team. It has made our operations so much more efficient. So to summarize, the way that TacoBot works is that it has access to all these different data sources like Discord, Notion, Slack, Google Drive, YouTube, Instagram, social media accounts, databases, and all internal tools as well. It is able to use this data in order to inform the team of what's happening. And at the same time, it also performs assistance functions like being able to search for things, remind you of things. And it's also the CEO of this company where it's making documentations of how we do certain things, onboarding new members, and giving suggestions for how we can operate better as a team. Model-wise, I started off with Gemini 3.6 Flash because it's a very fast model and easy to test. But I'm slowly migrating to local models instead to make sure that everything is kept private, hosted on our Mac Studio. Put all of this summary on screen now. Please take a screenshot. Also, I'll link it in the free guide in the description and I'll provide a little bit more detail about the specs that's being built because you can't fit everything into one screenshot. You check it out. Okay, so I've covered, I would say, the three functionalities of Hermes that I use the most personally. Of course, I'm still learning and I'm building new things literally every day to my Hermes as well. So if you're asking me in a month, probably going to be more here as well. But before I end this video, I do want to have a section where I cover some of my regularly scheduled Chrome jobs that are maybe like less sexy, but they are still very useful and they also help keep the whole system running. And that is specifically the alerts channel. See, the alerts channel is not sexy, but it's where I get alerts from all the different agents that are running in my Hermes. And they all check in at a different schedule, like once a day or weekly, telling me that everything is operating as intended. There's no security breaches and any suggestions and things that you may come across that I could potentially be working on. Like I said, not sexy, right? But this part of the entire Hermes system is really important because it's how I make sure that everything is operating smoothly and I'm building things in a sustainable way. Not going to go into way too much details about this, but you know the drill. I will put on screen now a summary of all the different check-ins and alerts and security systems that I have implemented. And finally, before I end this video, I do want to mention that I am working on a more complicated agentic system as well within Hermes. The concept is to be constantly researching different topics related to like AI and tech and education because, you know, that's the niche that I'm in. Evaluating it and giving me suggestions on different content ideas. And software that I could be building. Either like internal software or external software. So I'm still in the very beginning stages of this. I'm not ready to show you guys yet, but I wanted to let you guys know what I'm working on right now because this kind of thing is totally possible. I'm just still like tuning it. I think once I managed to get it set up that I can autonomously be like scraping the internet, that will be a very, very cool and very powerful system. So yes, for now, this is my Hermes setup. I hope this was interesting for you guys to see what I've built. And I'm curious, what have you built in your Hermes? What do you want to be building in your Hermes now that you've watched this video? And again, please let me know in the comments, just write like Hermes workshop or something like that. And that will let me know how much interest there is in doing a Hermes workshop. And maybe we'll do one. But for now, have a wonderful rest of your day. And I will see you guys in the next video where live stream.