Autores varios - AI
Hermes Agent w/ ChatGPT 5.5 is literally magic
🇬🇧 EN🇪🇸 ES
AI
Ask the tutor about this video
Sign in with a verified email and ask the tutor anything about this video.
TL;DR
Research summary
▶ Hermes Agent and its claimed advantages
The speaker introduces Hermes Agent as a 24/7 AI employee and says the video will cover setup, use cases, and whether to replace OpenClaw. He describes Hermes as focused on self-improvement through building and editing skills, lightweight, and performant. Based on his own experience, he says it has not slowed down and its updates have not broken for him. He contrasts this with his complaints that OpenClaw becomes slower as memory and context grow and that updates have repeatedly stopped it from working; he says he no longer updates OpenClaw. He also cites a $20-a-month Hermes plan, multi-agent support, and the team’s social media presence as advantages.
▶ Setup, model, and messaging
The speaker demonstrates installing Hermes by copying a command from its website into a terminal, then choosing a model. He recommends ChatGPT 5.5 over Claude Opus for this use, saying he considers them about equal for agent work and personality, while ChatGPT’s limits let him use a $20-a-month plan and OAuth; he says Claude Opus requires a $200-a-month plan. He recommends Telegram for messaging and explains that setup involves creating a bot through BotFather with `/newbot`, then entering its token in Hermes. He values Telegram because it works across devices and has features suited to agents.
▶ Building prototypes remotely
The speaker explains that Hermes builds or updates skills as it performs tasks, based on the user’s instructions. His first use case is building app prototypes while away from his computer: he sends a Telegram request for a 3D rocket simulator using 3.js, asking for a launch simulation with speed and altitude, then has it opened in a browser. He presents the example as a way to start an app on the go and continue developing it later at the computer.
▶ Steering work and checking results from a phone
The speaker says `/steer` can inject a new instruction while Hermes is working; he uses it to request that the rocket be blue. He shows the resulting simulator, including displayed altitude, speed, mission time, acceleration, and remaining fuel. He then asks Hermes to send a screenshot of the running simulator. He says the agent can open a browser, run the work, and send screenshots or videos for remote review, and describes using this workflow while away from his computer.
▶ Scheduled research and the speaker’s AI investment view
The speaker says he is a big investor who likes investing in AI stocks because he believes the AI buildout has not begun. He highlights Hermes cron jobs and says he uses them for recurring tasks, including checking every two hours that an OpenClaw agent is running. He schedules Hermes to research AI stocks with “massive moats” every 24 hours at 9:00 a.m. and produce a report. He says he thinks investing is especially important at the moment, expects money to go into AI companies, and believes there is a lot of money printing, so people should invest their money and assets. He says “not financial advice,” then jokingly adds “but uh honestly financial advice.” He names no stocks, describes no holdings or trades, and gives no allocation or other portfolio action.
▶ Learning a new workflow and creating a video
To demonstrate self-improvement, the speaker asks Hermes to make a 60-second YouTube Short about why to use Hermes with Remotion, which he describes as a plugin for agents to build motion videos. He says Hermes has no built-in Remotion skill and must research how to use it and build one. He reports that the task ran for over seven minutes, encountered problems it worked through, and created a reusable Remotion YouTube Short skill without being asked. The output also includes a thumbnail and a video with graphics; the speaker says he would remove the AI voice and add his own. He presents the result as a potential step toward an automated content pipeline.
▶ Finding personal automation use cases
The speaker suggests “reverse prompting”: give Hermes information about one’s career, interests, and job, then ask it for three use cases that can be implemented immediately to save time. He also recommends writing down manual tasks throughout the day, listing them for Hermes, and asking which can be automated. He says Hermes can use that list to suggest possible automations.
▶ Using Hermes and OpenClaw together
Asked whether to remove OpenClaw and use Hermes instead, the speaker recommends a multi-agent approach. He says the teams have separate roadmaps, both agents have distinct advantages, and users need not choose one; he uses both and recommends connecting both to a ChatGPT subscription. He says ACP lets them communicate, giving examples of one agent asking the other to perform a task. He concludes that users can set up a workflow and see which agent works best for them, then repeats his claims about Hermes building skills and improving with use. The ending asks viewers to like, subscribe, join his Vibe Coding Academy, and suggest future video topics.
Full transcript
[0:00] Hermes agent just might be the most
[0:02] powerful AI agent out there right now.
[0:04] It's your own personal AI employee and
[0:06] if you don't have one right now, you're
[0:08] missing out on so many productivity
[0:10] gains. In this video, I'll cover what
[0:12] Hermes agent is, the difference between
[0:14] it and open claw, how to set it up, a
[0:16] few use cases that will change your life
[0:19] and if you should replace your open claw
[0:21] with it. By the end of this video,
[0:22] you'll have your own full-time AI
[0:25] employee working for you 24 hours a day,
[0:27] 7 days a week powered by Hermes agent
[0:30] and chat GPT 5.5. If you're looking to
[0:32] get into AI agents, this is the video
[0:35] you've been waiting for. Now, let's lock
[0:37] in and get into it. This is Hermes agent
[0:40] in a second. I'm going to show you how
[0:41] to set it up and give it a whole bunch
[0:43] of use cases and prompts that are going
[0:45] to be really, really awesome. For the
[0:46] record, this is like the most visually
[0:48] aesthetic AI agent out of all of them.
[0:50] The people at New Research really have
[0:52] good taste. But before we do this, let's
[0:54] just quickly go into what Hermes agent
[0:57] is and what makes it different than open
[0:59] claw, some of its advantages. For those
[1:01] who are not familiar with Hermes agent
[1:02] or open claw, it's a 24/7
[1:05] AI agent employee that works on your
[1:07] computer, does anything a human being
[1:09] can do. Want to build apps? Want to send
[1:12] emails? Want to check your calendar?
[1:14] Want to send you files from one device
[1:16] to another? He's basically alive on your
[1:18] computer doing anything you want at any
[1:20] given time. The biggest advantage to
[1:23] Hermes agent, there's a bunch here, but
[1:25] it is fully focused on self-improving.
[1:28] Everything you do in Hermes agent, it
[1:30] self-improves. It builds and edits new
[1:33] skills as you go, which is really,
[1:35] really cool. So, the more you use it,
[1:37] the better it gets. But there's also a
[1:39] lot of other advantages as well. It's
[1:41] super lightweight, it's super
[1:43] performant. I've been using Hermes for
[1:45] months now. It hasn't slowed down at
[1:47] all. It hasn't felt like it got heavier.
[1:49] I love open claw, but it really has two
[1:52] things that piss me off. Number one is
[1:54] it gets less performant as you go. The
[1:57] more things you do with it, the heavier
[1:59] its memory gets, the heavier its context
[2:01] gets, it just slows down. It just feels
[2:04] like a very heavy AI agent. The other
[2:07] disadvantage to Open Claw at the moment
[2:09] is every time I update that thing, it
[2:11] breaks. I do not do updates on Open Claw
[2:14] anymore because every time I update it,
[2:17] it just stops working. That is really
[2:19] pissed me off to be quite honest with
[2:21] you. With Hermes Agent, it feels like
[2:23] they're a lot more selective and test a
[2:26] lot more the updates they put into it.
[2:28] It's never broken for me. It just works.
[2:31] So, a lot more reliable updates. There's
[2:33] a few other advantages as well. They
[2:34] have a new portal. Basically, what that
[2:36] is is its own AI plan. So, if you don't
[2:39] use ChatGPT or Claude, you can use their
[2:41] own very cheap $20 a month plan. It's
[2:44] built to be multi-agent. I have like
[2:46] five Hermes Agents going right now. It's
[2:48] fantastic. Really, really easy just to
[2:50] spin up new agents. And their social
[2:52] media team, I dare you to go on Twitter
[2:54] right now, tweet out Hermes Agent,
[2:56] you'll have five of their account
[2:58] members going in replying to all your
[2:59] tweets. So, really active social media
[3:01] team. And they appear just to be very
[3:03] nice, very nice happy people, which I
[3:05] like a lot too and I appreciate. So,
[3:07] that's the high-level, but as you know,
[3:08] this account's all about that action. We
[3:10] don't just sit here and look at charts
[3:11] all day. We get into actually using the
[3:14] tool. So, here we go, Hermes Agent.
[3:16] Let's talk about setting it up and let's
[3:18] also talk about the model you're using.
[3:20] As you can see here, a little spoiler
[3:21] alert, we're using GPT 5.5. So, let's
[3:24] talk about setting this up first. I'll
[3:25] put the link down below for Hermes
[3:27] Agent. Again, beautiful site. This color
[3:29] is my favorite color. This kind of
[3:31] forest green. Here's a little hint. You
[3:33] wear any clothes with forest green and
[3:34] then you're going to look better.
[3:35] There's There's no one on planet Earth
[3:37] that puts on forest green and doesn't
[3:39] immediately look 10 times better. So, I
[3:40] love the color. Here's all you got to
[3:42] do, super simple. Copy. Click copy right
[3:44] there. Boom, copy. You pop open your
[3:47] terminal. You paste in that command. You
[3:49] hit enter. You have it installed. Once
[3:51] you do that, you're going to want to
[3:53] select your model. This is the model
[3:55] that's going to be the brain for your
[3:56] agent. That's going to determine which
[3:57] task to do, which tools you should be
[3:59] using. I recommend chat GPT 5.5. Yes, I
[4:04] am recommending it over Claude Opus.
[4:06] Many different reasons. One is for agent
[4:09] work, it is just as good as Claude Opus.
[4:12] I don't think it's particularly better
[4:14] than Claude Opus for specifically for AI
[4:16] agents, but I don't think it's worse
[4:18] either. Personality-wise, it's just
[4:21] about the same as well. But here's the
[4:22] big advantage, is their limits are
[4:25] extremely high. You can get away with
[4:27] the $20 a month plan to run Hermes. If
[4:30] you use Claude Opus, you need the $200 a
[4:33] month plan. Chat GPT $20 OAuth works
[4:35] just fine for this. OpenAI very
[4:38] pro-consumer. They encourage you to use
[4:40] your OAuth. They encourage you to use
[4:42] your existing subscription. So that's
[4:44] great as well. And they don't make their
[4:46] models stupider. That's very nice and I
[4:47] like that as well. So Chat GPT 5.5 is my
[4:50] official recommendation for Hermes.
[4:52] Select that as your model. Use your
[4:54] OAuth login for that and you are good to
[4:57] go. I think the best messaging service
[4:59] for these AI agents, whether it's Hermes
[5:01] or Open Claw, is Telegram. So if you
[5:05] have a Telegram, set that up as well.
[5:07] That's super easy. All it's going to
[5:09] have you do is go to the Bot Father,
[5:11] which you can see right there, and do
[5:12] /newbot.
[5:14] That will give you a token. You take
[5:16] that token, you paste it into Hermes,
[5:18] and you're good to go. You have Telegram
[5:20] all set up. I like using Telegram cuz
[5:22] you can get it on any of your devices.
[5:24] You can get it on your iPhone, your
[5:25] iPad, any of your computers. This is
[5:27] going to be the best way to communicate
[5:28] with your agent. It just has so many
[5:30] features built specifically for AI
[5:32] agents. So set it up with Telegram as
[5:35] well. Super simple. So there it is. You
[5:37] have it set up. Now you can start using
[5:40] your agent. In a second, I'm going to go
[5:42] through how to personalize this to find
[5:44] the best use cases for yourself.
[5:46] Obviously, I'm the number one AI channel
[5:49] on YouTube. I have tens of thousands,
[5:51] hundred thousand people watching this. I
[5:52] don't know each and every one of you. I
[5:54] don't know each and every use case
[5:55] that's good for you. So, I'm going to
[5:56] show you what use cases are good for me,
[5:59] and then I'll show you how to do some
[6:00] self-discovery
[6:02] so you can figure out what use cases in
[6:03] Hermes agent are best for you. But, the
[6:05] one thing you want to know here is every
[6:08] time you use Hermes agent, it builds a
[6:12] new skill for what you're doing. So, you
[6:14] have a goal and build an app, create a
[6:16] video for you, whatever it is, it will
[6:18] constantly, based on how you instructed
[6:21] to do things, build and update its own
[6:24] skill. So, it is constantly
[6:25] self-improving. So, the more you use it,
[6:28] the better it gets. So, let's do this.
[6:29] I'm going to show you a few of my
[6:31] favorite use cases, then we'll go into
[6:33] figuring out your own best use cases. My
[6:35] number one use case, if you're anything
[6:37] like me, is I'm constantly thinking of
[6:40] things to build. I'm like I have a total
[6:43] builder mindset. I'm at the gym, I think
[6:46] of an app to build. I'm at the Whole
[6:48] Foods, I think of an app to build. I
[6:50] love using Hermes agents to build the
[6:52] initial prototypes for apps because I
[6:54] can be at the grocery store, I can go on
[6:57] Telegram on my phone, and I can say,
[6:59] "Hey, build this app for me." And I come
[7:01] back to my computer, and the app's up
[7:02] and running. So, let's do this. Say I
[7:04] wanted to build out a 3D simulation for
[7:07] me, maybe a 3D rocket ship launch,
[7:09] right? I'm at the gym, I go, "Oh, I'd
[7:11] love to come back and see like a 3D
[7:13] rocket simulator." So, I go and I say,
[7:16] "Please build me a 3D rocket ship
[7:21] simulator using 3.js."
[7:24] If you're a nerd like me, you like
[7:25] building video games, simulations,
[7:27] whatever it is, 3.js is the best library
[7:29] for this. It's going to build its own
[7:31] 3.js skill based on this. It should show
[7:34] a rocket ship taking off and show its
[7:38] speed and altitude. Then, pop it open in
[7:43] a browser and run it. Should be super
[7:47] detailed. I'm going to hit enter on
[7:48] that. It's going to get to work. It's
[7:50] going to build that out for me. The
[7:51] number one use case I'm showing you
[7:53] here, building out prototypes on the go,
[7:56] so that when you come back to your
[7:57] computer, you can keep building it out.
[7:59] And here's a quick one. Here's a quick
[8:00] tip for you. If any point you give it a
[8:02] task and you think it's something
[8:04] halfway through, you're like, "Oh man, I
[8:05] wish I would have said this." You can
[8:06] also do {slash} steer and it injects a
[8:10] message in there. We can say, "Make the
[8:12] rocket blue." And I'm going to hit enter
[8:15] on that and just going to give it a
[8:15] steer. It's just going to put a message
[8:17] in there as it's working that steers it
[8:20] in a certain direction. So, if you ever
[8:22] give it a task and you think while it's
[8:24] working, "Damn, I wish I would have told
[8:25] this that." Just do {slash} steer and
[8:28] you can steer your agent as it goes. By
[8:30] the way, as this builds, if you learn
[8:31] anything so far, leave a like down
[8:33] below, subscribe and turn on
[8:34] notifications. All I do is make amazing
[8:36] videos about AI. I'm thinking about
[8:38] making a video of powering Hermes agents
[8:41] by local models. If you're interested in
[8:43] that, let me know in the comments, too.
[8:45] And if you want a full live boot camp
[8:47] where you can ask me questions about
[8:49] Hermes agent, make sure to join the Vibe
[8:51] Coding Academy, link down below. It's
[8:53] the number one AI community on planet
[8:55] Earth. Go in and ask me questions, hang
[8:57] out, join a community of 1,500 people,
[8:59] totally up to you. Link down below. All
[9:01] right, so it popped open a browser on my
[9:03] computer. Let me drag this in so you can
[9:04] see this. Look at this.
[9:06] How sick is that? So, it's blue like I
[9:09] asked. Oh, wow. Look at this. So, you
[9:10] can see the altitude, the speed, the
[9:12] mission time, the acceleration, how much
[9:13] fuel I have left. I wonder what will
[9:14] happen. Will it explode when the fuel's
[9:16] out? Drag to orbit camera. Mouse wheel
[9:18] Oh, look at this. So, it built out the
[9:21] full 3D simulation for me. And so, the
[9:24] point here is is like, obviously this is
[9:26] really cool, but the point is is that
[9:28] you can be on the go wherever you are,
[9:31] come up with an idea for an app, and
[9:33] then you just go on your phone on
[9:34] Telegram and you message your Hermes
[9:36] agent and you build it out. And here's
[9:38] another really cool thing you can do.
[9:39] So, let's go in here, and let's just
[9:41] say, "Send me a screenshot of the rocket
[9:46] simulator running." So, say you're on
[9:49] the go, you have it building out an app
[9:51] for you. Now, you go and say, "Hey, just
[9:53] send me a screenshot of the work you
[9:54] did." Since it has full control over
[9:56] your computer, it can go, pop open its
[9:59] own browser, run the simulator, and send
[10:02] you video, send you screenshots, and
[10:04] confirm everything you're building is
[10:06] working. I love to do this when I'm in
[10:08] like full self-driving mode in my Tesla,
[10:10] and I'm just vibe coding on the go. I'm
[10:12] having it go, build things out, send me
[10:14] screenshots, send me videos, and I'm
[10:16] constantly able to be productive. And
[10:18] there you go, look at that. Sent me a
[10:20] screenshot of it running. That's really
[10:22] amazing. So, anywhere you are in the
[10:25] world, you don't have to download the
[10:27] ChatGPT app or the Claude app. You just
[10:29] have your Hermes agent go build things
[10:31] out and send you screenshots and videos.
[10:33] Now, here's another one I love to do.
[10:35] I'm a big investor. I like investing in
[10:37] AI stocks cuz I am a true believer the
[10:39] AI build out hasn't even begun yet. So,
[10:42] let's set it up. One big thing Hermes
[10:45] does really well is cron jobs. So, it's
[10:48] able to schedule its own task over and
[10:50] over and over again. I have it doing
[10:52] many, many cron jobs at once. I actually
[10:56] have it looking over my open Claude
[10:58] agent to make sure it's running well and
[11:00] doesn't crash. So, that's one thing you
[11:02] can do here if you're doing a
[11:03] multi-agent approach to say, "Hey, every
[11:05] 2 hours check on open Claude to make
[11:07] sure it's running still and not broken."
[11:09] Another thing you can do, research. So,
[11:11] let's do this. "Every 24 hours at 9:00
[11:15] a.m. do stock research for me. Find AI
[11:19] stocks I can invest in that have massive
[11:24] moats." I'm going to hit enter on that,
[11:26] and it is going to schedule itself for
[11:28] every 24 hours to do stock research for
[11:31] me. I think it has never been more
[11:33] important to be investing at the moment.
[11:36] I think tons of money is going to be
[11:37] going into AI companies. I think there's
[11:40] a lot of money printing going on, so you
[11:41] probably want your money and assets. You
[11:43] need to be investing, not financial
[11:45] advice, but uh honestly financial
[11:47] advice. And so it is scheduled every day
[11:49] 9:00 a.m. next run. It is going to be
[11:51] building me a stock report out. Super
[11:55] simple to do. There it is. Now it's
[11:56] scheduled to deliver me that stock
[11:58] report every morning. I highly recommend
[12:00] you do the same thing. Now let's show
[12:02] off some of the self-improvement. So
[12:03] everything I've had it do so far, it's
[12:05] been building and updating new skills
[12:07] for itself. But let's do a command here
[12:10] that will actually have it do something
[12:12] it doesn't know how to do, so it can
[12:13] figure it out and build its own skills.
[12:15] Watch this. So if you're not familiar
[12:17] with Remotion, Remotion is a really cool
[12:19] plugin for AI agents that allow AI
[12:21] agents to build out full motion videos.
[12:24] I want to start making more shorts on
[12:26] YouTube Shorts, so I want to start
[12:28] building out short motion videos that I
[12:31] can use as YouTube Shorts. Let's do
[12:32] this. Build me a 60-second YouTube Short
[12:36] clip on why you should be using Hermes
[12:41] agent. Do this with Remotion. Figure out
[12:45] how to use I'm going to hit enter on
[12:47] that. And so there is no built-in
[12:49] Remotion skill. It is going to have to
[12:51] do research, figure it out on its own,
[12:53] build its own skill set, self-improve.
[12:56] And at the end of this, if everything
[12:58] works out here, we should have a
[12:59] one-shot 60-second YouTube Short using
[13:03] Remotion that's going to be beautiful
[13:04] and have all these graphics and look
[13:06] really cool. Just as a side note here as
[13:08] this is going, we are 7 and a half
[13:10] minutes into this. It hasn't stopped
[13:12] once. It hasn't asked me for one thing.
[13:14] It is just going and going and going.
[13:16] And boom, as I was talking to you, it
[13:18] finished. So it went for over 7 minutes.
[13:21] It got tripped up multiple times. And
[13:23] every time it got tripped up, it just
[13:25] figured out how to do it every step of
[13:27] the way, which is really amazing. And
[13:29] then as you can see here, it just said,
[13:30] "Also, this is enough of workflow that I
[13:32] will I will save a reusable remotion
[13:34] short skill for future clips." And so it
[13:37] is going and it is saving that skill.
[13:39] Look at that. Skill remotion YouTube
[13:40] short created. I didn't even have to ask
[13:42] for it. New skill created. It is
[13:44] self-improving as we go. Let's open up
[13:46] this video. Let's see how this goes.
[13:48] Look at this. It even made the
[13:49] thumbnail. I didn't even ask for a
[13:50] thumbnail. It just made a thumbnail.
[13:52] Let's check that out real quick. Ooh,
[13:53] that's nice. Good-looking thumbnail
[13:55] there. Action is better than answers. I
[13:57] like that.
[13:59] Let's pop open the video. Let's see what
[14:00] we got. This is one shot, one sentence,
[14:03] one prompt. Here we go. Let's hit enter.
[14:05] Let's see what we got here.
[14:06] >> Your AI should do the work, not just
[14:08] describe it. Hermes agent turns chat
[14:10] into action. It can use real tools
[14:11] >> This is beautiful. terminal commands,
[14:14] file editing, image generation, web
[14:16] research, and messaging from one agent.
[14:18] It can drive the browser,
[14:19] >> Look at this. inspect UI,
[14:20] >> graphics. Let's get that in. It says the
[14:22] side and I'll pause here. I don't know
[14:23] if the audio is coming through or not.
[14:24] There's like an AI voice talking. I'm
[14:26] just going to rip that out. I'm going to
[14:27] put my own voice in it. But I put in one
[14:29] sentence. I now have a piece of content
[14:31] for YouTube for a YouTube short.
[14:32] >> Catch console errors, and visually
[14:34] verify the result. Look at the
[14:35] transition of the colors.
[14:37] >> packages, run tests, build apps, start
[14:39] servers, inspect logs, and fix what
[14:41] breaks. That is amazing. I now I'm
[14:44] building like my automated content
[14:45] pipeline. That is really, really sick.
[14:48] Now, let's talk about Those are three
[14:49] use cases I love. Let's talk about how
[14:52] you can find your own use cases. I talk
[14:54] about this on a ton of my videos. If
[14:56] you've listened to my videos before,
[14:57] you've heard this. Reverse prompting.
[14:59] Reverse prompting. If I'm you, I'm going
[15:02] in. First thing I'd do when I set up
[15:04] Hermes, I'm saying, "Here's everything
[15:06] you need to know about me, my career, my
[15:08] interests, my job, everything you need
[15:10] to know. What are three use cases we can
[15:13] implement right now that will save me
[15:15] time?" That will allow Hermes to
[15:17] understand who you are and come up with
[15:19] use cases that will be helpful for you.
[15:21] I promise you, if you do that right now,
[15:23] that exercise, your life will get much,
[15:25] much easier. The other thing you can do
[15:28] is, I do this all the time, get a piece
[15:30] of paper, write down on that piece of
[15:32] paper throughout your day every manual
[15:34] task you do. Every manual task you write
[15:36] it down here, then go into your Hermes
[15:38] agent, list out every task you did that
[15:41] you wrote down on your piece of paper,
[15:43] and say, "Which of these can you
[15:44] automate for me?" And Hermes will come
[15:46] and give you a whole bunch of use cases
[15:48] it can automate for you. Now, a question
[15:50] I get all the time is this, "Should I
[15:51] rip out Open Claw? Should I use Hermes
[15:54] instead?" The answer is truly, the
[15:56] future is multi-agent. You want to be
[15:59] multi-agent. Each team Hermes and Open
[16:02] Claw has a completely separate road map.
[16:04] There is no point in being tribalistic
[16:08] about this and going one or the other.
[16:10] I've had amazing experiences with both.
[16:12] Both have their own distinct advantages.
[16:14] Install both, hook them both into your
[16:17] ChatGPT subscription. ChatGPT 5.5 is
[16:20] great. You'll get ton of usage even on
[16:22] both. They'll be able to communicate
[16:24] with each other. There's something
[16:25] called ACP, which allows them to talk to
[16:28] each other. So, you can go to your Open
[16:30] Claw and say, "Hey, tell Hermes to do
[16:31] this." You can go to Hermes and it'll
[16:32] say, "Hey, tell Open Claw to do that."
[16:34] It'll work perfectly fine. Set up your
[16:36] own multi-agent workflow and find which
[16:38] agent works best for you. I use them
[16:40] both. They're both amazing. If you
[16:42] follow every step I went through in this
[16:44] video, you will have your own AI
[16:46] employee set up, your own Hermes agent,
[16:48] which will constantly be doing
[16:50] self-improvement, building its own
[16:51] skills, and getting better and better
[16:53] and better. If you learned anything at
[16:55] all, make sure to leave a like down
[16:56] below, subscribe, turn notifications.
[16:58] You want more Hermes content? You want a
[17:00] live boot camp? Vibe Coding Academy,
[17:01] link down below. And let me know in the
[17:03] comments, what tools are you using right
[17:05] now? What do you want more videos on?
[17:07] Hermes, Codex, Claude, Code, Opus 47,
[17:09] ChatGPT 5.5? Let me know in the comments
[17:12] below. All my videos are based on your
[17:14] feedback. I hope this was helpful. See
[17:16] you in the next video.
[0:02] powerful AI agent out there right now.
[0:04] It's your own personal AI employee and
[0:06] if you don't have one right now, you're
[0:08] missing out on so many productivity
[0:10] gains. In this video, I'll cover what
[0:12] Hermes agent is, the difference between
[0:14] it and open claw, how to set it up, a
[0:16] few use cases that will change your life
[0:19] and if you should replace your open claw
[0:21] with it. By the end of this video,
[0:22] you'll have your own full-time AI
[0:25] employee working for you 24 hours a day,
[0:27] 7 days a week powered by Hermes agent
[0:30] and chat GPT 5.5. If you're looking to
[0:32] get into AI agents, this is the video
[0:35] you've been waiting for. Now, let's lock
[0:37] in and get into it. This is Hermes agent
[0:40] in a second. I'm going to show you how
[0:41] to set it up and give it a whole bunch
[0:43] of use cases and prompts that are going
[0:45] to be really, really awesome. For the
[0:46] record, this is like the most visually
[0:48] aesthetic AI agent out of all of them.
[0:50] The people at New Research really have
[0:52] good taste. But before we do this, let's
[0:54] just quickly go into what Hermes agent
[0:57] is and what makes it different than open
[0:59] claw, some of its advantages. For those
[1:01] who are not familiar with Hermes agent
[1:02] or open claw, it's a 24/7
[1:05] AI agent employee that works on your
[1:07] computer, does anything a human being
[1:09] can do. Want to build apps? Want to send
[1:12] emails? Want to check your calendar?
[1:14] Want to send you files from one device
[1:16] to another? He's basically alive on your
[1:18] computer doing anything you want at any
[1:20] given time. The biggest advantage to
[1:23] Hermes agent, there's a bunch here, but
[1:25] it is fully focused on self-improving.
[1:28] Everything you do in Hermes agent, it
[1:30] self-improves. It builds and edits new
[1:33] skills as you go, which is really,
[1:35] really cool. So, the more you use it,
[1:37] the better it gets. But there's also a
[1:39] lot of other advantages as well. It's
[1:41] super lightweight, it's super
[1:43] performant. I've been using Hermes for
[1:45] months now. It hasn't slowed down at
[1:47] all. It hasn't felt like it got heavier.
[1:49] I love open claw, but it really has two
[1:52] things that piss me off. Number one is
[1:54] it gets less performant as you go. The
[1:57] more things you do with it, the heavier
[1:59] its memory gets, the heavier its context
[2:01] gets, it just slows down. It just feels
[2:04] like a very heavy AI agent. The other
[2:07] disadvantage to Open Claw at the moment
[2:09] is every time I update that thing, it
[2:11] breaks. I do not do updates on Open Claw
[2:14] anymore because every time I update it,
[2:17] it just stops working. That is really
[2:19] pissed me off to be quite honest with
[2:21] you. With Hermes Agent, it feels like
[2:23] they're a lot more selective and test a
[2:26] lot more the updates they put into it.
[2:28] It's never broken for me. It just works.
[2:31] So, a lot more reliable updates. There's
[2:33] a few other advantages as well. They
[2:34] have a new portal. Basically, what that
[2:36] is is its own AI plan. So, if you don't
[2:39] use ChatGPT or Claude, you can use their
[2:41] own very cheap $20 a month plan. It's
[2:44] built to be multi-agent. I have like
[2:46] five Hermes Agents going right now. It's
[2:48] fantastic. Really, really easy just to
[2:50] spin up new agents. And their social
[2:52] media team, I dare you to go on Twitter
[2:54] right now, tweet out Hermes Agent,
[2:56] you'll have five of their account
[2:58] members going in replying to all your
[2:59] tweets. So, really active social media
[3:01] team. And they appear just to be very
[3:03] nice, very nice happy people, which I
[3:05] like a lot too and I appreciate. So,
[3:07] that's the high-level, but as you know,
[3:08] this account's all about that action. We
[3:10] don't just sit here and look at charts
[3:11] all day. We get into actually using the
[3:14] tool. So, here we go, Hermes Agent.
[3:16] Let's talk about setting it up and let's
[3:18] also talk about the model you're using.
[3:20] As you can see here, a little spoiler
[3:21] alert, we're using GPT 5.5. So, let's
[3:24] talk about setting this up first. I'll
[3:25] put the link down below for Hermes
[3:27] Agent. Again, beautiful site. This color
[3:29] is my favorite color. This kind of
[3:31] forest green. Here's a little hint. You
[3:33] wear any clothes with forest green and
[3:34] then you're going to look better.
[3:35] There's There's no one on planet Earth
[3:37] that puts on forest green and doesn't
[3:39] immediately look 10 times better. So, I
[3:40] love the color. Here's all you got to
[3:42] do, super simple. Copy. Click copy right
[3:44] there. Boom, copy. You pop open your
[3:47] terminal. You paste in that command. You
[3:49] hit enter. You have it installed. Once
[3:51] you do that, you're going to want to
[3:53] select your model. This is the model
[3:55] that's going to be the brain for your
[3:56] agent. That's going to determine which
[3:57] task to do, which tools you should be
[3:59] using. I recommend chat GPT 5.5. Yes, I
[4:04] am recommending it over Claude Opus.
[4:06] Many different reasons. One is for agent
[4:09] work, it is just as good as Claude Opus.
[4:12] I don't think it's particularly better
[4:14] than Claude Opus for specifically for AI
[4:16] agents, but I don't think it's worse
[4:18] either. Personality-wise, it's just
[4:21] about the same as well. But here's the
[4:22] big advantage, is their limits are
[4:25] extremely high. You can get away with
[4:27] the $20 a month plan to run Hermes. If
[4:30] you use Claude Opus, you need the $200 a
[4:33] month plan. Chat GPT $20 OAuth works
[4:35] just fine for this. OpenAI very
[4:38] pro-consumer. They encourage you to use
[4:40] your OAuth. They encourage you to use
[4:42] your existing subscription. So that's
[4:44] great as well. And they don't make their
[4:46] models stupider. That's very nice and I
[4:47] like that as well. So Chat GPT 5.5 is my
[4:50] official recommendation for Hermes.
[4:52] Select that as your model. Use your
[4:54] OAuth login for that and you are good to
[4:57] go. I think the best messaging service
[4:59] for these AI agents, whether it's Hermes
[5:01] or Open Claw, is Telegram. So if you
[5:05] have a Telegram, set that up as well.
[5:07] That's super easy. All it's going to
[5:09] have you do is go to the Bot Father,
[5:11] which you can see right there, and do
[5:12] /newbot.
[5:14] That will give you a token. You take
[5:16] that token, you paste it into Hermes,
[5:18] and you're good to go. You have Telegram
[5:20] all set up. I like using Telegram cuz
[5:22] you can get it on any of your devices.
[5:24] You can get it on your iPhone, your
[5:25] iPad, any of your computers. This is
[5:27] going to be the best way to communicate
[5:28] with your agent. It just has so many
[5:30] features built specifically for AI
[5:32] agents. So set it up with Telegram as
[5:35] well. Super simple. So there it is. You
[5:37] have it set up. Now you can start using
[5:40] your agent. In a second, I'm going to go
[5:42] through how to personalize this to find
[5:44] the best use cases for yourself.
[5:46] Obviously, I'm the number one AI channel
[5:49] on YouTube. I have tens of thousands,
[5:51] hundred thousand people watching this. I
[5:52] don't know each and every one of you. I
[5:54] don't know each and every use case
[5:55] that's good for you. So, I'm going to
[5:56] show you what use cases are good for me,
[5:59] and then I'll show you how to do some
[6:00] self-discovery
[6:02] so you can figure out what use cases in
[6:03] Hermes agent are best for you. But, the
[6:05] one thing you want to know here is every
[6:08] time you use Hermes agent, it builds a
[6:12] new skill for what you're doing. So, you
[6:14] have a goal and build an app, create a
[6:16] video for you, whatever it is, it will
[6:18] constantly, based on how you instructed
[6:21] to do things, build and update its own
[6:24] skill. So, it is constantly
[6:25] self-improving. So, the more you use it,
[6:28] the better it gets. So, let's do this.
[6:29] I'm going to show you a few of my
[6:31] favorite use cases, then we'll go into
[6:33] figuring out your own best use cases. My
[6:35] number one use case, if you're anything
[6:37] like me, is I'm constantly thinking of
[6:40] things to build. I'm like I have a total
[6:43] builder mindset. I'm at the gym, I think
[6:46] of an app to build. I'm at the Whole
[6:48] Foods, I think of an app to build. I
[6:50] love using Hermes agents to build the
[6:52] initial prototypes for apps because I
[6:54] can be at the grocery store, I can go on
[6:57] Telegram on my phone, and I can say,
[6:59] "Hey, build this app for me." And I come
[7:01] back to my computer, and the app's up
[7:02] and running. So, let's do this. Say I
[7:04] wanted to build out a 3D simulation for
[7:07] me, maybe a 3D rocket ship launch,
[7:09] right? I'm at the gym, I go, "Oh, I'd
[7:11] love to come back and see like a 3D
[7:13] rocket simulator." So, I go and I say,
[7:16] "Please build me a 3D rocket ship
[7:21] simulator using 3.js."
[7:24] If you're a nerd like me, you like
[7:25] building video games, simulations,
[7:27] whatever it is, 3.js is the best library
[7:29] for this. It's going to build its own
[7:31] 3.js skill based on this. It should show
[7:34] a rocket ship taking off and show its
[7:38] speed and altitude. Then, pop it open in
[7:43] a browser and run it. Should be super
[7:47] detailed. I'm going to hit enter on
[7:48] that. It's going to get to work. It's
[7:50] going to build that out for me. The
[7:51] number one use case I'm showing you
[7:53] here, building out prototypes on the go,
[7:56] so that when you come back to your
[7:57] computer, you can keep building it out.
[7:59] And here's a quick one. Here's a quick
[8:00] tip for you. If any point you give it a
[8:02] task and you think it's something
[8:04] halfway through, you're like, "Oh man, I
[8:05] wish I would have said this." You can
[8:06] also do {slash} steer and it injects a
[8:10] message in there. We can say, "Make the
[8:12] rocket blue." And I'm going to hit enter
[8:15] on that and just going to give it a
[8:15] steer. It's just going to put a message
[8:17] in there as it's working that steers it
[8:20] in a certain direction. So, if you ever
[8:22] give it a task and you think while it's
[8:24] working, "Damn, I wish I would have told
[8:25] this that." Just do {slash} steer and
[8:28] you can steer your agent as it goes. By
[8:30] the way, as this builds, if you learn
[8:31] anything so far, leave a like down
[8:33] below, subscribe and turn on
[8:34] notifications. All I do is make amazing
[8:36] videos about AI. I'm thinking about
[8:38] making a video of powering Hermes agents
[8:41] by local models. If you're interested in
[8:43] that, let me know in the comments, too.
[8:45] And if you want a full live boot camp
[8:47] where you can ask me questions about
[8:49] Hermes agent, make sure to join the Vibe
[8:51] Coding Academy, link down below. It's
[8:53] the number one AI community on planet
[8:55] Earth. Go in and ask me questions, hang
[8:57] out, join a community of 1,500 people,
[8:59] totally up to you. Link down below. All
[9:01] right, so it popped open a browser on my
[9:03] computer. Let me drag this in so you can
[9:04] see this. Look at this.
[9:06] How sick is that? So, it's blue like I
[9:09] asked. Oh, wow. Look at this. So, you
[9:10] can see the altitude, the speed, the
[9:12] mission time, the acceleration, how much
[9:13] fuel I have left. I wonder what will
[9:14] happen. Will it explode when the fuel's
[9:16] out? Drag to orbit camera. Mouse wheel
[9:18] Oh, look at this. So, it built out the
[9:21] full 3D simulation for me. And so, the
[9:24] point here is is like, obviously this is
[9:26] really cool, but the point is is that
[9:28] you can be on the go wherever you are,
[9:31] come up with an idea for an app, and
[9:33] then you just go on your phone on
[9:34] Telegram and you message your Hermes
[9:36] agent and you build it out. And here's
[9:38] another really cool thing you can do.
[9:39] So, let's go in here, and let's just
[9:41] say, "Send me a screenshot of the rocket
[9:46] simulator running." So, say you're on
[9:49] the go, you have it building out an app
[9:51] for you. Now, you go and say, "Hey, just
[9:53] send me a screenshot of the work you
[9:54] did." Since it has full control over
[9:56] your computer, it can go, pop open its
[9:59] own browser, run the simulator, and send
[10:02] you video, send you screenshots, and
[10:04] confirm everything you're building is
[10:06] working. I love to do this when I'm in
[10:08] like full self-driving mode in my Tesla,
[10:10] and I'm just vibe coding on the go. I'm
[10:12] having it go, build things out, send me
[10:14] screenshots, send me videos, and I'm
[10:16] constantly able to be productive. And
[10:18] there you go, look at that. Sent me a
[10:20] screenshot of it running. That's really
[10:22] amazing. So, anywhere you are in the
[10:25] world, you don't have to download the
[10:27] ChatGPT app or the Claude app. You just
[10:29] have your Hermes agent go build things
[10:31] out and send you screenshots and videos.
[10:33] Now, here's another one I love to do.
[10:35] I'm a big investor. I like investing in
[10:37] AI stocks cuz I am a true believer the
[10:39] AI build out hasn't even begun yet. So,
[10:42] let's set it up. One big thing Hermes
[10:45] does really well is cron jobs. So, it's
[10:48] able to schedule its own task over and
[10:50] over and over again. I have it doing
[10:52] many, many cron jobs at once. I actually
[10:56] have it looking over my open Claude
[10:58] agent to make sure it's running well and
[11:00] doesn't crash. So, that's one thing you
[11:02] can do here if you're doing a
[11:03] multi-agent approach to say, "Hey, every
[11:05] 2 hours check on open Claude to make
[11:07] sure it's running still and not broken."
[11:09] Another thing you can do, research. So,
[11:11] let's do this. "Every 24 hours at 9:00
[11:15] a.m. do stock research for me. Find AI
[11:19] stocks I can invest in that have massive
[11:24] moats." I'm going to hit enter on that,
[11:26] and it is going to schedule itself for
[11:28] every 24 hours to do stock research for
[11:31] me. I think it has never been more
[11:33] important to be investing at the moment.
[11:36] I think tons of money is going to be
[11:37] going into AI companies. I think there's
[11:40] a lot of money printing going on, so you
[11:41] probably want your money and assets. You
[11:43] need to be investing, not financial
[11:45] advice, but uh honestly financial
[11:47] advice. And so it is scheduled every day
[11:49] 9:00 a.m. next run. It is going to be
[11:51] building me a stock report out. Super
[11:55] simple to do. There it is. Now it's
[11:56] scheduled to deliver me that stock
[11:58] report every morning. I highly recommend
[12:00] you do the same thing. Now let's show
[12:02] off some of the self-improvement. So
[12:03] everything I've had it do so far, it's
[12:05] been building and updating new skills
[12:07] for itself. But let's do a command here
[12:10] that will actually have it do something
[12:12] it doesn't know how to do, so it can
[12:13] figure it out and build its own skills.
[12:15] Watch this. So if you're not familiar
[12:17] with Remotion, Remotion is a really cool
[12:19] plugin for AI agents that allow AI
[12:21] agents to build out full motion videos.
[12:24] I want to start making more shorts on
[12:26] YouTube Shorts, so I want to start
[12:28] building out short motion videos that I
[12:31] can use as YouTube Shorts. Let's do
[12:32] this. Build me a 60-second YouTube Short
[12:36] clip on why you should be using Hermes
[12:41] agent. Do this with Remotion. Figure out
[12:45] how to use I'm going to hit enter on
[12:47] that. And so there is no built-in
[12:49] Remotion skill. It is going to have to
[12:51] do research, figure it out on its own,
[12:53] build its own skill set, self-improve.
[12:56] And at the end of this, if everything
[12:58] works out here, we should have a
[12:59] one-shot 60-second YouTube Short using
[13:03] Remotion that's going to be beautiful
[13:04] and have all these graphics and look
[13:06] really cool. Just as a side note here as
[13:08] this is going, we are 7 and a half
[13:10] minutes into this. It hasn't stopped
[13:12] once. It hasn't asked me for one thing.
[13:14] It is just going and going and going.
[13:16] And boom, as I was talking to you, it
[13:18] finished. So it went for over 7 minutes.
[13:21] It got tripped up multiple times. And
[13:23] every time it got tripped up, it just
[13:25] figured out how to do it every step of
[13:27] the way, which is really amazing. And
[13:29] then as you can see here, it just said,
[13:30] "Also, this is enough of workflow that I
[13:32] will I will save a reusable remotion
[13:34] short skill for future clips." And so it
[13:37] is going and it is saving that skill.
[13:39] Look at that. Skill remotion YouTube
[13:40] short created. I didn't even have to ask
[13:42] for it. New skill created. It is
[13:44] self-improving as we go. Let's open up
[13:46] this video. Let's see how this goes.
[13:48] Look at this. It even made the
[13:49] thumbnail. I didn't even ask for a
[13:50] thumbnail. It just made a thumbnail.
[13:52] Let's check that out real quick. Ooh,
[13:53] that's nice. Good-looking thumbnail
[13:55] there. Action is better than answers. I
[13:57] like that.
[13:59] Let's pop open the video. Let's see what
[14:00] we got. This is one shot, one sentence,
[14:03] one prompt. Here we go. Let's hit enter.
[14:05] Let's see what we got here.
[14:06] >> Your AI should do the work, not just
[14:08] describe it. Hermes agent turns chat
[14:10] into action. It can use real tools
[14:11] >> This is beautiful. terminal commands,
[14:14] file editing, image generation, web
[14:16] research, and messaging from one agent.
[14:18] It can drive the browser,
[14:19] >> Look at this. inspect UI,
[14:20] >> graphics. Let's get that in. It says the
[14:22] side and I'll pause here. I don't know
[14:23] if the audio is coming through or not.
[14:24] There's like an AI voice talking. I'm
[14:26] just going to rip that out. I'm going to
[14:27] put my own voice in it. But I put in one
[14:29] sentence. I now have a piece of content
[14:31] for YouTube for a YouTube short.
[14:32] >> Catch console errors, and visually
[14:34] verify the result. Look at the
[14:35] transition of the colors.
[14:37] >> packages, run tests, build apps, start
[14:39] servers, inspect logs, and fix what
[14:41] breaks. That is amazing. I now I'm
[14:44] building like my automated content
[14:45] pipeline. That is really, really sick.
[14:48] Now, let's talk about Those are three
[14:49] use cases I love. Let's talk about how
[14:52] you can find your own use cases. I talk
[14:54] about this on a ton of my videos. If
[14:56] you've listened to my videos before,
[14:57] you've heard this. Reverse prompting.
[14:59] Reverse prompting. If I'm you, I'm going
[15:02] in. First thing I'd do when I set up
[15:04] Hermes, I'm saying, "Here's everything
[15:06] you need to know about me, my career, my
[15:08] interests, my job, everything you need
[15:10] to know. What are three use cases we can
[15:13] implement right now that will save me
[15:15] time?" That will allow Hermes to
[15:17] understand who you are and come up with
[15:19] use cases that will be helpful for you.
[15:21] I promise you, if you do that right now,
[15:23] that exercise, your life will get much,
[15:25] much easier. The other thing you can do
[15:28] is, I do this all the time, get a piece
[15:30] of paper, write down on that piece of
[15:32] paper throughout your day every manual
[15:34] task you do. Every manual task you write
[15:36] it down here, then go into your Hermes
[15:38] agent, list out every task you did that
[15:41] you wrote down on your piece of paper,
[15:43] and say, "Which of these can you
[15:44] automate for me?" And Hermes will come
[15:46] and give you a whole bunch of use cases
[15:48] it can automate for you. Now, a question
[15:50] I get all the time is this, "Should I
[15:51] rip out Open Claw? Should I use Hermes
[15:54] instead?" The answer is truly, the
[15:56] future is multi-agent. You want to be
[15:59] multi-agent. Each team Hermes and Open
[16:02] Claw has a completely separate road map.
[16:04] There is no point in being tribalistic
[16:08] about this and going one or the other.
[16:10] I've had amazing experiences with both.
[16:12] Both have their own distinct advantages.
[16:14] Install both, hook them both into your
[16:17] ChatGPT subscription. ChatGPT 5.5 is
[16:20] great. You'll get ton of usage even on
[16:22] both. They'll be able to communicate
[16:24] with each other. There's something
[16:25] called ACP, which allows them to talk to
[16:28] each other. So, you can go to your Open
[16:30] Claw and say, "Hey, tell Hermes to do
[16:31] this." You can go to Hermes and it'll
[16:32] say, "Hey, tell Open Claw to do that."
[16:34] It'll work perfectly fine. Set up your
[16:36] own multi-agent workflow and find which
[16:38] agent works best for you. I use them
[16:40] both. They're both amazing. If you
[16:42] follow every step I went through in this
[16:44] video, you will have your own AI
[16:46] employee set up, your own Hermes agent,
[16:48] which will constantly be doing
[16:50] self-improvement, building its own
[16:51] skills, and getting better and better
[16:53] and better. If you learned anything at
[16:55] all, make sure to leave a like down
[16:56] below, subscribe, turn notifications.
[16:58] You want more Hermes content? You want a
[17:00] live boot camp? Vibe Coding Academy,
[17:01] link down below. And let me know in the
[17:03] comments, what tools are you using right
[17:05] now? What do you want more videos on?
[17:07] Hermes, Codex, Claude, Code, Opus 47,
[17:09] ChatGPT 5.5? Let me know in the comments
[17:12] below. All my videos are based on your
[17:14] feedback. I hope this was helpful. See
[17:16] you in the next video.