Hermes Agent Fundamentals In 29 Minutes — Transcript
Full transcript
- 0:00Welcome to the Hermes Agent's
- 0:01Fundamentals Guide. Setting up your
- 0:03Hermes Agent correctly is so important.
- 0:05It's the difference between having a
- 0:06terrible time versus being able to build
- 0:08something sustainable that you can build
- 0:10on top of and become more magical every
- 0:12day. It's been over 4 months now that
- 0:14I've been using Hermes Agent
- 0:15consistently. Here's my multi-agent
- 0:16system doing research and building
- 0:18workflows and softwares while I'm
- 0:19chatting with it using local models to
- 0:21keep things free and private. Not that I
- 0:23have anything to hide, of course. And
- 0:24here is my Obsidian second brain and
- 0:26getting smarter every day. So, in this
- 0:27video, I'm going to walk you through my
- 0:29Hermes setup. I have tips and tricks
- 0:31along the way that can make a really big
- 0:32difference for your experience. And I'll
- 0:34be giving you exact prompts and
- 0:35workflows so you can also customize and
- 0:37build your Hermes Agent to exactly what
- 0:39you want it to be based on your budget,
- 0:41requirements, and use cases. Now,
- 0:43without further ado, let's go. All
- 0:45right, outline for today's video. We're
- 0:47going to start off with talking about
- 0:48hardware choices, where your Hermes
- 0:50Agent is going to live. Then, I'll walk
- 0:51you through the download and setup
- 0:52process. We'll walk through Hermes major
- 0:54features like tools, skills, and cron
- 0:56jobs, and the memory management system
- 0:57including how to optimize it. I'll also
- 0:59talk about use open-source models. And
- 1:01finally, after you've mastered these
- 1:02fundamentals, I want to give you a taste
- 1:04for the multi-agent frameworks, how
- 1:06powerful those can be, especially in
- 1:08combination with OpenClaw or Claw code,
- 1:09which is absolutely a game-changer if
- 1:11you're building software. By the way, I
- 1:12just want to say thank you guys so much
- 1:14for suggesting me moving over from
- 1:15OpenClaw to Hermes. I honestly was and
- 1:18still am a big fan of OpenClaw and will
- 1:19probably revisit it again later. But for
- 1:21now, Hermes has become my major local
- 1:24agent. So, yeah, you guys were right.
- 1:26Okay, let's start off with talking about
- 1:27hardware.
- 1:32For me, I primarily run my Hermes Agent
- 1:33on my Mac Studio. It has 64 gigs of RAM,
- 1:36an M4 chip, and it's on 24/7. This is
- 1:39powerful enough for me to run some
- 1:40pretty great local AI models and only
- 1:42accessing more powerful cloud-based
- 1:44models when I need to do something like
- 1:46really fancy. This is a good balance for
- 1:47me and my team because even though the
- 1:48Mac Studio did cost me a lot of money,
- 1:50it is still way cheaper than when I was
- 1:52using a Mac Mini or like even my laptop
- 1:54and having to rely really heavily on
- 1:56cloud models. But do not fear if you
- 1:58don't happen to have a Mac Studio. You
- 2:00can also have very powerful Hermes Agent
- 2:02setups with other options. I would say
- 2:04that there are four primary options that
- 2:05you can choose. The first one is a
- 2:07dedicated powerful local machine like a
- 2:09Mac Studio, for example. The second one
- 2:11is a VPS, a virtual private server,
- 2:13which is kind of like renting computer
- 2:14space on the cloud. These can be as
- 2:16cheap as just five or six dollars and be
- 2:18on 24/7 as well. The third option is
- 2:20just to use any old computer or laptop
- 2:22that you happen to have around, like
- 2:24this MacBook Pro, for example. 16 gigs
- 2:26of RAM. This is actually where I
- 2:27started. Just make sure that you wipe
- 2:28everything and I just kept it like on
- 2:3024/7, so it was basically like a
- 2:32dedicated machine. And finally, if you
- 2:34really don't want to spend any money and
- 2:35you don't happen to have any other
- 2:36computers lying around, you can also use
- 2:38your own personal computer. But, if
- 2:40you're going to do this, I highly,
- 2:41highly recommend that you use Docker,
- 2:43which containerizes your agent, so it
- 2:45doesn't allow it to go bananas on the
- 2:47rest of your computer. All right, so I'm
- 2:48not going to go into way too much detail
- 2:49about each of these options, uh because
- 2:51I have actually covered those in some
- 2:53videos, which I can link in the
- 2:53description you can check out. Now,
- 2:55let's actually move on to the next step,
- 2:57which is setting up our Hermes Agent.
- 3:00>> [music]
- 3:02>> So, to set up Hermes Agent, you want to
- 3:05go to hermesagent.newsresearch.com
- 3:07and you have two different options. You
- 3:09can install it via the desktop app if
- 3:11you have a Mac or you can install it via
- 3:13the terminal. You can have a Mac as well
- 3:15or if you have a Windows. Most tutorials
- 3:17I see actually go through the terminal
- 3:18route. So, I actually want to show you
- 3:20how to do it via the desktop app, which
- 3:22is a great option for people who don't
- 3:23feel comfortable typing stuff into
- 3:24terminal. But, if you are going through
- 3:25the terminal, it's super easy as well.
- 3:27Just click the copy, stick it into your
- 3:29terminal, and just go through the
- 3:30onboarding sequence. So, click download
- 3:32for macOS. All right, so once you have
- 3:33it downloaded open, it would look like
- 3:35this. A clean Hermes Agent. This
- 3:37interface probably looks very familiar,
- 3:39it's very similar to your normal chatbot
- 3:41interfaces. And the first thing you want
- 3:42to do is come over here and look at the
- 3:45different models. So, there are some
- 3:46models that are already here because I
- 3:48have like an Anthropic key that's set
- 3:49up. This probably going to look
- 3:50different depending on what you already
- 3:52do have set up over here. So, you can
- 3:54just choose what you have and if you
- 3:56don't have anything that you like, you
- 3:57can also choose edit models and then
- 3:59click add provider. Here you can connect
- 4:01to any providers that you have. If you
- 4:03have an open AI ChatGPT subscription,
- 4:05you can just log in to that and be able
- 4:07to use your subscription here or any of
- 4:08the other ones here. If you want to go
- 4:10fully local, that also works. I will
- 4:12show you a little bit later in the video
- 4:13how to set up your local model. But, if
- 4:15you don't happen to have like any kind
- 4:17of provider subscription, the easiest
- 4:18option is probably just go with the news
- 4:20portal. It is a monthly subscription
- 4:22that gets you access to hundreds of
- 4:23different models. Very convenient. And
- 4:25once you have that connected, you can
- 4:27just start off by saying, "Hi."
- 4:30Yay! Just had your first interaction
- 4:32with Hermes agent. By the way, the
- 4:33general consensus is that the Sonnet 5
- 4:36model is the best driver for the Hermes
- 4:38agent. But, because of Anthropic not
- 4:40letting you use your subscription and
- 4:42you actually have to use the API key, it
- 4:43can get pretty expensive. For me
- 4:45personally, because I use Hermes a lot,
- 4:47my main driver is an open source model
- 4:49called the Quant 5.6 35B model, which I
- 4:52run using my Mac Studio. There's a lot I
- 4:54can talk about when it comes to model
- 4:55selection and what you can be using, but
- 4:57to keep ourselves on track, I'm just
- 4:59going to put on screen now some of the
- 5:00common models that people choose to use
- 5:02as their main drivers, including their
- 5:03pros and cons. And I'm also going to put
- 5:05in the free guide below more detailed
- 5:08recommendations as well as a prompt that
- 5:09can help you choose what is the best
- 5:11driver model for your specific setup.
- 5:13Now, let's continue with talking with
- 5:15the agent for a bit. The first thing
- 5:16that I actually recommend that you do is
- 5:18tell your agent a little bit about
- 5:19yourself. Now, instead of actually
- 5:21typing it because of AI psychosis and
- 5:23I'm incapable of like writing stuff
- 5:25anymore, what I like to do is go to a
- 5:27chatbot like Claude, for example, and
- 5:29then just write like, "Give me a
- 5:33brief description about what you know
- 5:35about me that I can give
- 5:38to Hermes agent." So, yes, it will just
- 5:41tell me about myself, which I can just
- 5:42copy that and I'll write, "Here is a bit
- 5:46about myself. Please remember this.
- 5:49There you go. And you can see that it
- 5:51says running memory because it's
- 5:52actually going to remember
- 5:54this. Now, you may be asking, "How is it
- 5:57remembering it? Where is it storing this
- 5:58information?" Great question. Let me
- 6:00show you. I'm going to actually going to
- 6:01jump now to my Mac Studio, which has my
- 6:03like actual like proper Hermez Agent
- 6:06running up. I I, you know, redownloaded
- 6:07and installed it again to show you guys
- 6:09like the download process.
- 6:10Uh but let's actually see what it looks
- 6:11like. I'm going to move there now. Now,
- 6:13this [clears throat] is the Mac Studio.
- 6:14I don't know if you can actually notice
- 6:15a difference, but
- 6:17anyways, let me actually show you where
- 6:19Hermez Agent stores his memories. So, if
- 6:21you go to finder, if you have a MacBook,
- 6:24it is stored under the users, your name,
- 6:28so my username is Tina, and then
- 6:29.hermez. And you can literally see
- 6:32everything that's contained here. But
- 6:34this is not very pretty, so I'm going to
- 6:35actually show you using Obsidian. So, it
- 6:37looks better. So, here is the .hermez
- 6:39folder, and it will be telling
- 6:40information about the user, me. It's
- 6:43stored under memory, and you can click
- 6:45user. So, this is literally like human
- 6:47readable. You can read it yourself what
- 6:49Hermez knows about you. It has specific
- 6:51information that it saves into the
- 6:54.users. It's just a markdown file. And
- 6:56since we're talking about where it's
- 6:57storing stuff, there's also a memory
- 6:59markdown file where it saves information
- 7:02that is very crucial to your current
- 7:03setup. This is the built-in Hermez
- 7:05memory, so you don't have to do anything
- 7:07here. So, I will talk a little bit more
- 7:09about the memory system and how it is
- 7:10that you can enhance this a little bit
- 7:12later in the video. By the way, there
- 7:13are a lot of things that you can do just
- 7:15via this desktop app. It is so
- 7:16convenient. Like if you go to your
- 7:18settings over here, there's a lot of
- 7:20things that you can configure. So, your
- 7:22model, information about your chat, you
- 7:23can assign it different personalities,
- 7:25appearance. Uh there's a lot of things
- 7:28that you can customize, what you want it
- 7:29to look like. I just like leaving it as
- 7:31default. I think it looks pretty good.
- 7:32Information about workspace, safety,
- 7:34memory, and context. Yeah, there's a lot
- 7:36that you can actually do just from here.
- 7:38And with that, let's transition into the
- 7:40next section. I want to show you the
- 7:41major Hermes features so that you can
- 7:43have a very strong foundation to build
- 7:45on.
- 7:50Okay, so there are five major features
- 7:51that I think is important for you to
- 7:53understand. The first one is tools. So,
- 7:55tools are pre-configured infrastructure
- 7:57that's managed by News Research, the
- 7:59creators of Hermes. You'll be able to do
- 8:01things like search the web, generate
- 8:03images, text images, browser
- 8:05automations, etc. So, it's not something
- 8:07that you actually do yourself. For
- 8:09example, I can ask Hermes agent, "Search
- 8:11the web for research into the latest AI
- 8:16chip developments."
- 8:18So, we see that it uses a search
- 8:19function and was able to get the results
- 8:21on the latest chip developments. If you
- 8:23click on skills and tools and you look
- 8:25at tool kits, here you can see a lot of
- 8:27different tools that Hermes natively
- 8:30has. So, this is already very, very
- 8:31powerful, but Hermes can do more than
- 8:33this because it also has what are called
- 8:36integrations. Integrations allow Hermes
- 8:38to call so call upon and use other
- 8:41third-party softwares. It's what allows
- 8:42you to connect to different providers,
- 8:44for example, for different models, like
- 8:46News Portal, Anthropic, OpenAI, MiniMax,
- 8:49etc., etc. It also allows you to connect
- 8:51to messaging platforms. Like over here,
- 8:54I have Discord connected. Ignore my
- 8:561,000 tabs, but this is how I like to
- 8:59get alerts and communicate with Hermes
- 9:01as well. So, here Hermes is sending me
- 9:04my daily brief.
- 9:05>> Good morning, Tina. Here's your AI brief
- 9:08for Thursday, July 4th, 2026.
- 9:11>> That's a little taste for what you can
- 9:12build using these integrations. It's
- 9:14also where I get alerts for bills that
- 9:16are successful from Hermes, too. By the
- 9:18way, if you also want to connect to
- 9:19Discord or like Telegram or any other
- 9:21chat app, it's really easy. You can
- 9:22actually start a new session and just
- 9:24ask your Hermes agent, "Hey, can you
- 9:26help me connect to Telegram?" For
- 9:29example, or WhatsApp, or whatever.
- 9:37And if you click on messaging, you can
- 9:39see all the different types of message
- 9:41integrations, messaging app integrations
- 9:43that you can connect to if you wish. I
- 9:45personally like using Discord uh because
- 9:47it allows me to be able to separate like
- 9:49different channels and then get like
- 9:50alerts and communicate based on
- 9:52different channels. Also, I can use it
- 9:53on my phone. But, I'm pretty sure that
- 9:55they're probably going to roll out with
- 9:57a mobile version of the desktop app soon
- 9:59as well because Open Cloud did. So, I
- 10:01wouldn't be surprised if you can just
- 10:03use the desktop app like with a mobile
- 10:04version very shortly too. Great. So,
- 10:06News Research and Hermes supports a lot
- 10:08of these integrations already and
- 10:09they're super easy to use. I'm going to
- 10:11put on screen now some of my favorite
- 10:12integrations that I like to use. And if
- 10:15it so happens that Hermes doesn't have a
- 10:17native integration already, you can use
- 10:19what is called MCP in order to provide
- 10:21your agent with additional third-party
- 10:23tools. Notebook LM, for example, is one
- 10:25of my favorite MCPs in order to do
- 10:26research and to create a podcast like
- 10:28you just saw earlier for my daily
- 10:29briefings. And the way you get it to
- 10:31connect is also, as you probably would
- 10:32guess, really easy. You can just say,
- 10:34"Hey, help me connect to the Notebook LM
- 10:38MCP, [clears throat] for example." And
- 10:39then you would just follow the
- 10:40instructions. I'm also going to put on
- 10:41screen now some of my favorite MCPs and
- 10:43what I like to do with them. I'm not
- 10:44going to go into too much more details
- 10:46about MCPs and any additional plugins. I
- 10:48feel like that's a little bit more
- 10:49advanced, maybe for a next video. Now,
- 10:50let's move on to the second category of
- 10:52features. Also, the feature that made
- 10:54Hermes famous to begin with, which is
- 10:57skills. You see, Hermes agent advertises
- 10:59itself as the agent that grows with you.
- 11:02And what that means is that it's able to
- 11:04learn as you work with it, aka develop
- 11:07skills with you over time. So, what are
- 11:09skills? Well, you can think about skills
- 11:11as an instruction manual for completing
- 11:13a task. If we go to Skills and Tools,
- 11:15you can actually see some of the
- 11:16pre-built skills that Hermes already
- 11:17has. And there's like different
- 11:19categories for this. For example,
- 11:21there's Apple Notes that can manage
- 11:22Apple Notes, Apple Reminders, iMessage
- 11:24that can send and receive iMessages.
- 11:26Let's see, if we go to Research, there's
- 11:28one that allows you to search different
- 11:29papers, Creative lets you generate ASCII
- 11:32video videos, etc. etc. There are a lot
- 11:34of them over here. So, you can actually
- 11:35check them out. But, what makes Hermes
- 11:37famous is that as you work along with
- 11:39it, it's able to learn these skills both
- 11:40by itself and you can specifically teach
- 11:42it new skills to do, too. Here, for
- 11:44example, in this conversation, I was
- 11:45asking Hermes to help me research some
- 11:48business ideas and see how valid it is.
- 11:50And it tells me this is ideal for
- 11:52creating AI-generated micro-dramas,
- 11:54which is something that's already very
- 11:55dominant in production in China. So, it
- 11:58you know, did a lot of research, told me
- 12:00information about what's the landscape
- 12:01looking like, what the competitors are
- 12:03looking like, what's the blueprint, and
- 12:05where's the opportunity here. And it
- 12:06also gave me a plan for what to build,
- 12:08including phase one, phase two, and
- 12:09phase three, blah blah blah, etc. I then
- 12:11asked you to give it a final score after
- 12:13scores across different dimensions based
- 12:15upon these additional criteria. So, I
- 12:17gave it more criteria about what I cared
- 12:19about. Uh so, I don't want to raise VC
- 12:21capital, I want it to be bootstrap
- 12:22lifestyle business, etc. etc. And then
- 12:23finally, it was able to evaluate how
- 12:25good this is for a business idea that I
- 12:27might want to pursue. So, it tells me
- 12:29that it has a score of 7.5 out of 10,
- 12:31and more details over here as well.
- 12:32Here's where I want to draw your
- 12:33attention to. I asked it, "Make this
- 12:35into a skill for evaluating business
- 12:37ideas." And that's when I said, "Sure."
- 12:39And it was able to make this skill. You
- 12:42can see the description of the skill.
- 12:43So, what it's supposed to do. So,
- 12:44business idea evaluator is supposed to
- 12:46score the framework for evaluating
- 12:47business ideas against bootstrap
- 12:49lifestyle criteria and the different
- 12:51dimensions. And it also explains how
- 12:53Hermes in the future should be using the
- 12:55skill. So, how to use. User describes a
- 12:56business idea. Um it would score each
- 12:58dimension 1 to 10, assign weights,
- 13:01calculate weighted total, and then
- 13:02present the total score out of 10,
- 13:03individual scores, and key strengths and
- 13:05weaknesses, and then finally give it a
- 13:07clear recommendation, which is go,
- 13:08caution, or no go. And so now I have
- 13:10that skill available to me, and I can do
- 13:12{slash} it's called business idea
- 13:14evaluator, and I can say evaluate
- 13:16business idea. Uh let's see. To evaluate
- 13:19business idea of a personalized learning
- 13:23app that is able to generate a study
- 13:27plan and lessons specific to what you
- 13:30want to learn that uses AI to be able to
- 13:33generate this. Okay, cool. And we can
- 13:35see that the it says here the user has
- 13:36invoked a business idea evaluator skill.
- 13:38Okay, great. It tells me lots of things,
- 13:41blah blah blah blah blah, and it tells
- 13:42me, "Wow, 9.2 out of 10." Great. I
- 13:45should do this. And it asked me, "Want
- 13:47me to start building the phase one MVP?"
- 13:49So, say I really like this idea and I'm
- 13:51going to use another skill and say,
- 13:53"Save this research about the AI
- 13:56learning app into my Obsidian vault."
- 13:59Great. So, it uses the Obsidian skill.
- 14:01If you want to look under thinking, it
- 14:02says, "I'll use the write file tool as
- 14:04recommended by the Obsidian skill." And
- 14:07under projects, we have another project
- 14:09called AI learning app where it has a
- 14:11lot of information here. Don't worry so
- 14:13much about this Obsidian setup. I'm
- 14:15actually show you later how to make this
- 14:17in the memory section of the video. But
- 14:18first, I want to talk about one more
- 14:19major feature of Hermes agent, which is
- 14:22cron jobs or scheduled task. A cron job
- 14:25is a scheduled task that runs
- 14:26automatically at certain times or
- 14:28specific intervals without having to
- 14:30manually trigger it. You see a lot of
- 14:32things in life are actually reoccurring.
- 14:34Like every day, you got to get up and
- 14:36like exist and eat food, you know? Yeah,
- 14:38like computer, when we're doing AI
- 14:40stuff, it's very similar. You generally
- 14:42want to do very similar things every
- 14:43single day or every single week where
- 14:45you know some type of cadence. That's
- 14:46why cron jobs are really useful for
- 14:47that. One example of this is a daily
- 14:49scheduled brief that I have. At 11:00
- 14:51p.m. the previous day, it creates this
- 14:54brief for me of all the daily news that
- 14:56I'm interested in and also makes it into
- 14:58an audio file that I can listen to in
- 15:00the morning. I also have cron jobs that
- 15:02summarize my Apple notes because I
- 15:03really like like, you know, writing a
- 15:04bunch of stuff into my Apple notes um
- 15:06and then compile that together into
- 15:08Obsidian as well. So, I can like kind of
- 15:10query my own brain if that makes sense.
- 15:12Another one that's very important is
- 15:13that I have a cron job that every single
- 15:16night looks through to make sure that
- 15:18everything is running properly cuz I'm
- 15:19hosting a lot of the models. I also have
- 15:21like multi-agent systems going on
- 15:22building software all at the same time.
- 15:24So, I want to make sure that everything
- 15:26is being handled the way that's supposed
- 15:27to be handled. I will put on screen some
- 15:28of the cron jobs that I have and the
- 15:30ones that I'm building as well, some of
- 15:31my favorite ones, which hopefully can
- 15:33give you inspo for things that you can
- 15:34build as well. So, take a screenshot or
- 15:35not. I also do have a free guide in the
- 15:38description where you can sign up and
- 15:40then I will send you the free video
- 15:41companion to your email, if you want.
- 15:43Great. So, we've covered the major
- 15:44features for Hermes Agent and you can
- 15:46build so much just with these features
- 15:49here. Now, there is one more concept
- 15:50that I have separated into its own
- 15:52section because it is so important,
- 15:54which I'm going to talk about now, which
- 15:55is memory.
- 16:00For those of you who have used Open Claw
- 16:02or like Co-work or other types of agents
- 16:04previously, you probably got really
- 16:06frustrated at some point because of the
- 16:09memory. Like it started like forgetting
- 16:10things. Um it can't like search up what
- 16:12it's supposed to start to getting really
- 16:13slow as well and these things started
- 16:15breaking as well. Very annoying. All of
- 16:17this has to do with memory management.
- 16:18Now, Hermes already has a pretty good
- 16:20memory system and actually like for most
- 16:23people, if you just use his default
- 16:24memory system, you would be okay. But,
- 16:26if you are someone who really wants to
- 16:28get a lot out of Hermes and just use it
- 16:30to the best of its abilities, there are
- 16:32also two customizations that you can do,
- 16:34which I find that has been amazing and
- 16:36really, really helpful. But, first let
- 16:37me actually quickly cover what is Hermes
- 16:39Agent's default memory setup. Okay, we
- 16:42are back in Obsidian where I'm showing
- 16:44you the entire Hermes folder, everything
- 16:47that's contained in here. The memory
- 16:48system is a two-tier system. And
- 16:50actually, you've already seen uh the
- 16:52first tier when I showed you earlier.
- 16:54Pop quiz, write in the comments, do you
- 16:56know what it is? It's okay if you don't.
- 16:58I'll show you again. It is the
- 16:59combination of three different files.
- 17:01The memory.md file stores really
- 17:04important information about the setup,
- 17:07like where things live, what are the
- 17:08major plugins and integrations that are
- 17:10there, what's in progress, etc. The
- 17:13user.md file that has information about
- 17:15the user, you, and the is sold on the
- 17:18file, which is information about the
- 17:19agent. So, it knows who it is. So, this
- 17:21first here gives a general overview for
- 17:23your Hermes agent to know high-level
- 17:25information. But, each of these files is
- 17:27actually kept deliberately brief.
- 17:28Otherwise, they just get really, really
- 17:30bloated and slows down Hermes in the
- 17:32long run. But, what happens if you want
- 17:33to find information that is more
- 17:34specific? Well, luckily, Hermes has a
- 17:36second tier memory system where it can
- 17:38use a tool. Remember what a tool is?
- 17:40Called session search where it can
- 17:41search through logs of archived
- 17:43information. This is stored in a sequel
- 17:45light database. I don't have to go into
- 17:46too much detail about this, but Hermes
- 17:48actually does store all the
- 17:50conversations that you have about it. It
- 17:51just wouldn't actually go and look at it
- 17:53unless it needs to retrieve the
- 17:54information. Let me actually show you
- 17:55this in action. So, start a new session
- 17:57and I will write, "Remind me, what were
- 18:00we talking about earlier about business
- 18:02ideas?" And you can see that it's using
- 18:05the session search tool, searching past
- 18:08sessions. And it's able to tell you,
- 18:10"Okay, we talked about two business
- 18:11ideas recently, a vertical micro drama
- 18:14condensation engine and AI personalized
- 18:16learning app." So, that's how it's able
- 18:18to retrieve that information, even
- 18:19though it's not in its core memory.md
- 18:22file. So, I like I said, these two
- 18:23tiers, I guess like tiers, levels,
- 18:25whatever, levels of memory are actually
- 18:27really pretty good. But, if you really
- 18:28want to enhance the system, there are
- 18:30two things that you can do to push you
- 18:32up two notches that can make it even
- 18:35better. So, that is tier number three.
- 18:37You can install a external provider
- 18:39memory plugin. My favorite one to use is
- 18:41called Honcho's. Honcho's doesn't
- 18:42actually replace tier one and tier two
- 18:45of your memory. It just adds on to it.
- 18:47It's looking at all your interactions
- 18:48with your Hermes agent and collecting
- 18:50what is called implicit understanding,
- 18:52patterns across your interactions and
- 18:54things you may not even be aware of
- 18:55about yourself, the projects that you
- 18:56do, the way that you behave, and it
- 18:58supplements the Hermes agent with this
- 18:59information so it has better memory. And
- 19:02to do this is also very easy. Cookie for
- 19:04anybody that can guess. Yes, you just go
- 19:07into a new session. So, say, "Help me
- 19:09set up Honcho." And it will walk you
- 19:11through the exact steps of setting this
- 19:13up, including going on the Honcho
- 19:14terminal. I think you do get quite a lot
- 19:16of free credits right now, at least at
- 19:18the time of this filming, so you can
- 19:19take advantage of this. Then, after you
- 19:21have it installed, it's just going to
- 19:22silently be working and collecting data,
- 19:24and you can have some conversations with
- 19:25it. You want to see what's happening,
- 19:27you can look at the web app on Honcho,
- 19:28or you could just go to your terminal.
- 19:31Okay. If you do know how to use
- 19:32terminal, this is very handy. Okay, so I
- 19:34have a three different terminals running
- 19:36at the same time, just like ignore
- 19:37what's going on. Let's focus just on
- 19:39this one. Ignore all of that. So, we can
- 19:41just write like Hermes Honcho status,
- 19:44and we can see, great, after it's
- 19:46installed, it's true, blah blah blah,
- 19:48it's it's done correctly, and then it's
- 19:50telling you me information about myself.
- 19:52Like, so here's my user Picard, it has
- 19:54information about this explicit
- 19:56observations, it also says some of the
- 19:58information that's observed. So, great,
- 19:59we know that it's working. You will just
- 20:01find that as you're talking to your
- 20:02Hermes agent, sometimes it would start
- 20:04pulling information from Honcho when it
- 20:06needs to have specific information. It's
- 20:08like a subtle thing, but you'll notice
- 20:09that it just makes your experience so
- 20:12much smoother. Would recommend this if
- 20:13you intend to use Hermes a lot. Now,
- 20:15finally, the fourth tier. You really
- 20:17want to make your memory system super
- 20:19duper robust by setting up what is
- 20:21called a second brain. That is the
- 20:23Obsidian thing that I was showing you
- 20:24earlier. So, this is one of my Obsidian
- 20:27second brains. And the way that you do
- 20:28this, just go to Obsidian, and then
- 20:31click download, and you would be able to
- 20:33have it as a desktop app. And once you
- 20:35open it, you'll be greeted with
- 20:36something that looks like this. Click
- 20:38open folder as vault, and you can just
- 20:40pick it wherever you want, just create a
- 20:41file, call it like Hermes vault, or
- 20:44whatever, and then just click open, and
- 20:46then you have your vault. That's
- 20:48currently empty. What you need to do is
- 20:49you have to connect your Hermes to your
- 20:51vault, which, again, all you got to do
- 20:53is just ask your Hermes to help you set
- 20:55up an Obsidian vault. But basically,
- 20:56after you do this, as I showed you
- 20:59earlier, when you're talking to your
- 21:00Hermes agent and you're doing research
- 21:01or whatever, you can ask it to use the
- 21:03Obsidian skill in order to save certain
- 21:05things into your Obsidian vault, and it
- 21:07can do this for you. Like I said, this
- 21:09is just one of my vaults. I have like
- 21:10daily briefs that are here. I also like
- 21:12to save projects that are here. I'm
- 21:13going to put on screen some other things
- 21:15I like to save into my vault, which
- 21:16could be useful for you. So, I like to
- 21:18save like certain projects. So, in this
- 21:20way, like say at some point I actually
- 21:22want to build like this AI learning app,
- 21:24right? Like all this information is
- 21:25here. So, why is this useful? So, first
- 21:27of all, you can like read the things in
- 21:28your vault, which is very nice. Uh maybe
- 21:30you're like, "Wow, I wonder what it says
- 21:31over here." But on on top of that, your
- 21:33Hermes agent is also able to access this
- 21:35information that you stored into the
- 21:37vault. So, you can continue working on
- 21:39things from there. For example, I have
- 21:41like this food tracker PRD, uh which I
- 21:43made in which I wanted to track like the
- 21:46food that I'm eating because I'm too
- 21:47lazy to track my food, but I really
- 21:48should cuz I'm trying to get more in
- 21:50shape, you know? So, I wrote this PRD
- 21:52for this like app that I wanted to
- 21:53build, but I didn't actually like ever
- 21:54build it. So, what I can say here is,
- 21:56"Use Obsidian skill." Probably don't you
- 21:58actually need to like specify Obsidian
- 21:59skill, but I'm just going to do that
- 22:01just to make sure it's like very
- 22:02specific, "to grab the PRD for the food
- 22:04tracker and come up with a step-by-step
- 22:08plan for how I can implement the MVP
- 22:12version in 1 hour." Great. So, you can
- 22:14see that it's using the skill view and
- 22:16it's able to grab it. Okay, great. So,
- 22:18it came back and now it has a plan. So,
- 22:21it gives me the key MVP scope for 60
- 22:24minutes. I should get like a photo of my
- 22:25food GPT-4 vision and Discord replies
- 22:28with macros, Obsidian JSON storage with
- 22:30the daily totals, and having specific
- 22:32commands like today log and goals I'm
- 22:35able to see what it is that I have been
- 22:37eating and logging. And I should not
- 22:39have weekly chronic corrected flow
- 22:41because I want to do this in like 60
- 22:42minutes, right? It tells me what the
- 22:43tasks are and blah blah blah blah blah,
- 22:44etc. So, if I actually want to do this,
- 22:46then I can start implementing it.
- 22:47Anyways, not going to do that now, but I
- 22:48hope you can see why having this fourth
- 22:50level of an Obsidian second brain is
- 22:52really, really helpful. Especially as
- 22:54you do more work, you save more stuff,
- 22:56and there's more things that you would
- 22:57want your agent to be able to help you
- 22:59with. Okay, so there is one more topic
- 23:01left for your Hermes setup that will set
- 23:04the foundations so you can build so many
- 23:06amazing things. And that is open-source
- 23:08AI.
- 23:13Now, I want to make a caveat here. This
- 23:14is optional. If you are super baller and
- 23:17you're like totally cool just paying for
- 23:19like Anthropic Claude models or like
- 23:20other models through news for whatever
- 23:22and you're like willing to spend like
- 23:23hundreds of hundreds dollars, that's
- 23:25totally cool, okay? That's totally cool.
- 23:27I FELT THE STING AND RAN. But for me, I
- 23:29am not willing to spend that much money.
- 23:31So that's why I started using local AI
- 23:33models. Models that I have downloaded
- 23:35into my Mac Studio, runs completely
- 23:37locally, it's private, and everything is
- 23:39like just is contained within this Mac
- 23:41Studio, which is actually kind of crazy,
- 23:43right? That you can just like be able to
- 23:44own and use this like very powerful
- 23:47model. By the way, that is actually what
- 23:49I've been using this entire time
- 23:50throughout all of these like demos and
- 23:51stuff that we've been doing. This is a
- 23:53completely local model. Like when I use
- 23:54it, it's just on my machine. I don't
- 23:55need to pay for anything. It's free. So
- 23:57yeah, the one I'm using is a Qwen 3.6
- 23:5935B model. And I'm actually just going
- 24:00to put on screen some of the other local
- 24:02models that I do use for different
- 24:04things. Not going to go into too much
- 24:05more detail about that. The general
- 24:06consensus is that if you want to save
- 24:07money like running Hermes and agents,
- 24:09using local AI models is going to help
- 24:11reduce that cost substantially. And
- 24:13there are some really, really good local
- 24:15AI models now that honestly like we were
- 24:17seeing me do stuff, right? And it's it's
- 24:19like just like really, really good. So
- 24:21if you have the hardware and the ability
- 24:23of doing this, then I would recommend
- 24:24that you go for it. Now, the easiest way
- 24:26to use an open-source local model is to
- 24:28download Ollama, which is a model
- 24:30package manager. So you go to ollama.com
- 24:33and then click download. Then when you
- 24:34open it and you open the launch tab,
- 24:36there actually is Hermes agent here. So
- 24:38you can just copy-paste this terminal
- 24:40command. Go to your terminal, paste it,
- 24:42press enter, and you can select one of
- 24:44these to get started. It actually gives
- 24:45you recommendations. Or you can go to
- 24:47the tried and true path now and just
- 24:49say, "Hey Hermes, help me use Ollama to
- 24:53download open-source models and use them
- 24:56with you." Click enter and just follow
- 24:59the instructions. However, what I prefer
- 25:01to do is use something called llama.cpp.
- 25:03Without going into too much detail,
- 25:05llama.cpp is actually what Ollama uses.
- 25:08It just like wraps around it, so it's
- 25:09easier for you to use. But if you use
- 25:11the underlying thing itself, which is
- 25:13llama.cpp, it gives you direct control
- 25:15with the model, so you can tweak it to
- 25:16the way that you like it, and it's
- 25:17faster as well. Took me about like 30
- 25:19minutes to set up, you know, again,
- 25:21using Hermes to help me like set it up.
- 25:23Got to do like some terminal stuff, but
- 25:24you know, after 30 minutes, I was able
- 25:26to have my model up and running, and
- 25:28it's been great. Would recommend. All
- 25:29right. Honestly, congratulations for
- 25:31making it until this part of the video.
- 25:33I know we've covered a lot today, but
- 25:34trust me, if you have that patience to
- 25:36set it up like this, and just like do it
- 25:38properly, you're going to get so much
- 25:39more out of your Hermes agent. It is
- 25:41amazing.
- 25:43>> [music]
- 25:45>> So, for the last section, I'm actually
- 25:47going to show you what you can build
- 25:49once you lay this foundation. I'm going
- 25:50to show you my multi-agent framework.
- 25:52This is my Discord, where I primarily
- 25:54build workflows and softwares with
- 25:55Hermes. So, let's actually build
- 25:57something together to show you how it
- 25:58works. Recently, I've been really into
- 26:00building these little desktop
- 26:02productivity learning apps. Like this
- 26:05one, for example, it allows me to just
- 26:07input a study plan or whatever it is
- 26:09that I'm doing, and then I can like
- 26:10check them off, and you know, yay, it
- 26:12tells me that it's complete. And most
- 26:14importantly, it saves all of this to my
- 26:16Obsidian vault. So, after I finish these
- 26:18plans, it would automatically save it,
- 26:19and I can actually see everything that
- 26:21I'm doing every day. So, let's build
- 26:22another one of these little companion
- 26:23apps today. I'm going to go here and
- 26:25say, "@Hermes, draft the specs for a
- 26:28simple Pomodoro app that has the same
- 26:31style as the roadmap tracker app. I want
- 26:35to be able to optionally
- 26:37input [music] what I'm working on, and
- 26:40then set the amount of time. I want the
- 26:42amount of time to be visual, so I can
- 26:45see
- 26:46it decreasing, like a visual timer. I
- 26:49also want things saved to Obsidian Vault
- 26:54as well. Click enter.
- 26:56Cool. So it looks like Hermes has picked
- 26:58it up.
- 27:01Okay, great. It gave this PRD and it
- 27:05looks good to me. So let's say yes.
- 27:08Queue it, please.
- 27:11This says that it is queued and it will
- 27:12let me know when it's ready or needs an
- 27:14input. Sounds good. And hey, look at
- 27:16that. Yay! Under the builds channel, we
- 27:20can see that the build is successful.
- 27:21All right, moment of truth. Okay, so it
- 27:25has been built. Let's try it out. Let's
- 27:26just say 1 minute and for the label,
- 27:29we'll just put test. [music] Click
- 27:31enter. Okay. So just to see that it's
- 27:33actually floating on top. So if I go to
- 27:36like Hermes for example or some other
- 27:38apps. Okay, so we know that it's always
- 27:40going to be on top, which is good. Uh
- 27:43yep, and we can also move it around to
- 27:44the side as well.
- 27:53Okay, that was so obnoxious. Let's go to
- 27:56Obsidian and there you go. 1 minute it's
- 27:58called label test and it's completed.
- 28:00Yay!
- 28:01So yeah, that is a full build uh using
- 28:04Discord. Since it is a multi-agent
- 28:06system, you can also come back and ask
- 28:08it to do other things as well. Like hey
- 28:10Hermes bot, draft the spec for some
- 28:13ideas of other desktop widget apps
- 28:17>> [music]
- 28:17>> that I can build.
- 28:19Et cetera. At the same time, you can
- 28:21also ask it, Hermes bot, uh draft the
- 28:23spec to change the colors
- 28:28of the Pomodoro app to more pastel, for
- 28:31example. Yep, and they're going to be
- 28:33working in parallel. Yeah, isn't that
- 28:35just kind of like mind-blowing? Like
- 28:36I've built so many different types of
- 28:38software and workflows using this
- 28:39multi-agent system. Once you get to this
- 28:41stage, Hermes also has a native
- 28:42orchestration via this Kanban board,
- 28:44which you you actually like orchestrate
- 28:46all of your different agents and
- 28:47coordinate them and make them like just
- 28:49super super optimized. So yeah, I've
- 28:51been having so much fun building with
- 28:53Hermes. So if you guys want to see
- 28:54another video where I go more in depth
- 28:57about this multi-agent system and other
- 28:59projects that you can be building, let
- 29:01me know in the comments if you're
- 29:02interested in that. If you are, I'm very
- 29:04happy to go more in depth into this. I'm
- 29:06going to put on screen now some of the
- 29:07things that I have my multi-agent system
- 29:09set up, more advanced things and skills
- 29:10and just like things that I'm doing
- 29:12right now that you can do, you can
- 29:14unlock set up your Hermes agent
- 29:16properly. What I'm particularly like
- 29:18really into right now is actually using
- 29:19hardware and connecting hardware with my
- 29:21Hermes. So yes, super exciting. But for
- 29:23now, I think this is way enough to get
- 29:25you started. Thank you so much for
- 29:26watching until the end of this video. I
- 29:28hope this is helpful for you. I'm going
- 29:29to put on screen now a little quiz that
- 29:31will help you retain all of this
- 29:32information that we've covered. And let
- 29:34me know in the comments what you are
- 29:35planning to build with your Hermes. I
- 29:37will see you guys in the next video or
- 29:38live stream.
About this transcript
This page contains the full transcript of Hermes Agent Fundamentals In 29 Minutes by Tina Huang, generated from the public captions YouTube serves with the video. The transcript has 6,571 words across 935 segments, with the original timestamps preserved so you can click any line to jump to that moment in the embedded player.
What you can do with it
Use the transcript to take notes, quote the speaker, build a study guide, generate a summary with ChatGPT or Claude via the YouTube Summary tool, or export it as a timed subtitle file with YouTube to SRT. You can also re-open it in the transcriber to choose a caption track available on the video.
Free YouTube transcript tool
YouTube2Text is a free YouTube transcript generator — no signup, no daily limit. Paste any YouTube link and get the full transcript instantly, with timestamps, click-to-jump, caption language selection, AI prompts for ChatGPT, Claude, and Gemini, and exports to TXT, SRT, VTT, or Markdown.