Full transcript
0:01Welcome everyone to the Hermes Agent
0:03Master Class. I've dug into basically
0:06every document, every feature, every
0:11community skill and plugin and designed
0:13this master class from start to finish.
0:16It's right now designed as 11 episodes,
0:19but obviously as Hermes agent evolves,
0:22I'll be adding new episodes, new modules
0:24to keep up with the changes and
0:27evolution of the agent itself. So this
0:29master class will have everything you
0:31want to know to get the most out of
0:33Hermes agent and we're starting
0:35literally from zero completely new. This
0:39is going to be an introduction to the
0:40agent itself and full installation and
0:42then a tour of the CLI. So by the end of
0:45this we will have a Hermes agent
0:48installed and configured and we're have
0:50going to have our first task actually
0:52completed.
0:53So part one what is Hermes?
0:56Hermes agent is an open- source AI agent
0:59built by new research and it's a lab
1:02behind the Hermes model family. Most
1:05agents are built by tool companies. This
1:07one is built by the people who train the
1:09models and that matters and more on that
1:12in a second. It runs in your terminal as
1:14a CLI or as a gateway that can answer on
1:16your phone via Telegram, Discord, Slack,
1:19iMessage, WeChat. They have 16 platforms
1:22as of the latest version.
1:24It's provider agnostic. So you can use
1:26claude, you can use GBT, Grock, KI, you
1:30could use a local model via O Lama or
1:33anything else. Hermes doesn't care. You
1:35can pick exactly what you want.
1:38So why does this exist? New Research
1:41trains models. They built Hermes
1:43originally as the harness harness that
1:45generates training data. So when you use
1:46Hermes, you're using the same tool New
1:49Research used to build the next
1:51generation of their models. And that's a
1:53credibility signal you don't often get.
1:56And this is obviously open sourced MIT
1:58licensed as of this recording. Around 7
2:01thou 70,000 stars on GitHub, nearly 300
2:04contributors, and it's moving fast. We
2:06literally just got the the most re
2:08recent version 0.9.0 today.
2:12So the key bet of Hermes agent is
2:14compounding value. Most agents are
2:17stateless. So every session starts from
2:20zero. Every chat GBT session starts
2:22blank. Every new clawed conversation
2:24starts blank. Open claw usually the
2:27same. Hermes was built differently.
2:29Every session feeds the next one.
2:33So there's a four-step loop here. You
2:35give it a task. It could be anything. It
2:38records a trajectory and this is the
2:40tools used, every decision made, every
2:42result. And then it can extract that as
2:46a reusable skill based on how it worked.
2:49So the next time it's faster, more
2:51accurate, and costs less tokens. And
2:53this is compounding value as a thesis
2:55and what makes Hermes really unique. A
2:58Hermes installed that you've had for 3
2:59months is fundamentally better than what
3:01you set up yesterday.
3:04So you're going to get to see this for
3:05yourself. I'm going to be installing
3:06this on a new device that has never had
3:09a Hermes agent. And through this master
3:11class, you're going to be able to see
3:13exactly how it develops, what skills it
3:15learns, and how it becomes smarter. So
3:19how do you know Hermes agent is right
3:20for you? So if you use agents enough
3:23that compounding benefit pays off.
3:26Occasional users may not feel it. If you
3:28want model freedom, you might want to
3:29start on claude, want to start on Grock,
3:32end up using local. Hermes can roll with
3:35that pretty easily. And if privacy or
3:38cost matters to you, the local model
3:40support is really first class. So you
3:43may not want Hermes. If you just want a
3:44chatbot, you could just use chat GBT.
3:46don't need a full agent or if you need
3:50max platform integration breath. Right
3:52now, OpenClaw has uh a bigger out of the
3:54box integrations. But that is Hermes is
3:57catching up as we speak. So, by the time
3:59you see this, that may not even be true.
4:02Okay, enough of me yapping. Let's
4:04actually install this thing. So, will it
4:07run on my machine? If you have a Mac OS,
4:09it's native. Linux native. Windows. It
4:13only works on WSL2
4:15and you need this. These are the
4:18dependencies that it will install. Now,
4:21I am on a Windows PC, so we need this
4:25WSL2. And if you've done any Linux
4:28development on a Windows machine, you're
4:31familiar with it. But if you're not, I
4:33will show you what it means. So, this
4:35kind of gives you a breakdown, but WSL
4:37stands for Windows Subsystem for Linux,
4:40and the default one is usually Ubuntu.
4:43Uh, but there's others. I have Debian on
4:45this computer, so that's the one you'll
4:48see me use, but usually the default will
4:50be Ubuntu. It's basically the same
4:51thing. And this lets us develop with
4:54Linux applications and tools uh directly
4:57onto Windows. So, it's pretty nice and
4:59it's pretty easy. Uh, you need to be
5:01running on a Windows 10 version or
5:03higher. Um, and it's just this one
5:06thing, WSL install. You need to be in
5:09PowerShell in administrator mode.
5:12And then just that one command and
5:14that'll install Auntu
5:17distribution of Linux uh as the default.
5:22And once you have that installed, you're
5:24ready. This is PowerShell.
5:27Uh so just do WSL
5:30and it will launch your uh WSL.
5:34There we go. Base in Tombi at Compass.
5:37You'll see something like this.
5:40That's the name of my my PC. But it
5:43doesn't matter where you install it.
5:45It's going to ignore whatever directory
5:46you're in and just install it to the
5:48Hermes, which is going to be home slash
5:51uh whatever your ID is and then Hermes.
5:53Hermes. So the quick install here is
5:56this is the uh you see on the top here,
5:59this is the new research/ Hermes agent
6:02and this will give you the quick install
6:04mode here.
6:08And it's just one command
6:14and that'll open up the installer here.
6:17It'll install all the dependencies that
6:19you need. You see python uh git node.js
6:28and everything will compile here.
6:33And the configuration is pretty quick.
6:36You'll need to know your pseudo password
6:39um for the build tools.
6:42See, all dependencies have been
6:44installed here.
6:48Okay. And once the installation is done,
6:51you're going to get choice between full
6:52setup or just a quick setup.
6:56You can do quick, it's very quick, but
6:57for this we're going to do full setup.
6:59So, you can get to see everything here.
7:02So, first you can see the providers and
7:04these are your AI providers.
7:06New portal is new researches
7:08subscription service for AI models. Open
7:12router um you can see anthropic which
7:14has the cloud models you can use with an
7:17API key or through cloud code. Um you
7:20have open AAI codeex you have Quen uh
7:24GitHub Copilot Hugging Face Inference
7:27Models and there's more providers uh
7:29that you could
7:31use. For now, we're just going to use
7:33open router. Uh, since I have an API key
7:37for this one,
7:39uh, no open API key. So, let's here I'll
7:42show you how to do that. So, if you just
7:43go to open router.com, you're going to
7:45have to put in a credit card at least
7:47get a couple credits. You can see
7:51and you just get an API key. You can
7:53create it there and then just insert it.
7:55It'll be something like this SK
7:57something. So, once you get your API
8:00key, you just add it here.
8:03It saved it. I'll show you how to use
8:04other models for now, but for now, we're
8:06going to use Open Router.
8:08Um, you can see all the different models
8:09that it has. It has Opus 4.6 as
8:13currently in use as the default, but
8:15let's try to do something a little bit
8:16cheaper.
8:19See, has a lot of different options.
8:24You can enter a custom model name.
8:27Uh let's do
8:30for now let's do this one the MIMO V2
8:33Pro.
8:37So then it will offer a fallback and
8:39rotation. So Hermes can keep multiple
8:41credentials for one provider and rotate
8:43between them when one of the credentials
8:46is exhausted or rate limited. And this
8:48preserves your primary provider while
8:49reducing interruptions. So this is a
8:51pretty nice function.
8:54So you just add a a second API key here
8:58and then you could have that whenever
9:00the first one runs out of credits or you
9:02hit some type of uh rate limit. And then
9:06we have the the max tool calling. Just
9:10set that to 90.
9:13So, that will
9:15determine how often your agent is
9:18allowed to
9:19uh call tools.
9:22Tool progress. We're just going to set
9:24that to all so you can see every single
9:27tool call for now. But you can change
9:29this. You just turn this off or only
9:32show the the new tools when they happen.
9:35You see here in Telegram, this is my
9:38uh existing Hermes agent. You can see
9:41the tools that we have. Like these are
9:43all tool calls, listing, searching
9:45files, executing code. So it'll list out
9:47all these so you can see them, which I
9:49kind of like. So we'll just do all for
9:52now. But you can set this for off or set
9:54this for only new tools, however you
9:56like. Context compression.
10:01So for this, the higher the threshold
10:03means the later you take to compress.
10:06Let's do 0. Let's put the difference.
10:080.75.
10:10depends on which model I'm going to be
10:12using because some models can handle
10:15longer context windows better and others
10:17need compression quickly. Uh session
10:20reset mode, inactivity and daily reset.
10:24Recommended reset whichever comes first,
10:26inactivity only, daily only, never auto
10:30reset.
10:32Um let's do inactivity and daily reset.
10:37Inactivity timeout.
10:40So this is how long inactivity is.
10:43Each message adds to the conversation
10:45history which means growing API cost. So
10:48this is also about context management.
10:51If you have very long uh context, if you
10:55continue to do work there, you're going
10:56to have higher token costs, higher IP
10:59API costs.
11:02Um,
11:04yeah, we'll just skip that. Daily reset
11:08do midnight. That makes sense. Select
11:10platforms to configure. So, we're going
11:13to do this
11:15in the next video when we get into the
11:18the um the gateway. We're going to set
11:20up a couple of these. So, I'll save
11:21those for next time.
11:27Uh choose provider local browser
11:30browser use fire crawl.
11:34Uh we'll configure that later. Keep
11:36defaults.
11:39Flux 2 pro. You can get an API key here.
11:42I guess
11:45this is for im image generation. Let me
11:47see. Is this a paid thing? Okay. Okay.
11:50So, I don't know if I'll use this FAL
11:52for image generation or something else,
11:55but we can add a key for now. You just
11:57signed in. There's no you signed in with
12:00your GitHub
12:02and it'll create it for API, I guess.
12:04API. Sure. Hermes agent. Okay. So, I
12:07just paste that in.
12:09Uh, right. Microsoft Edge search
12:14provider
12:15fire crawl cloud.
12:18Uh, Tabi is good as well. AI native
12:21search.
12:23You can also do this self-hosted or you
12:25can do it locally. Uh, let's keep the
12:27defaults and I can configure that later.
12:30Uh, so here we go. That was all the
12:31configuration you need to do. Some of
12:33the stuff we'll set up later like the
12:35Telegram that'll be in the next episode.
12:36But should we launch Hermes chat now?
12:39Sure.
12:43There we go. Hermes agent.
12:47Sure you all wanted to see that. So this
12:48has all the available tools. You can see
12:51all the available skills and these come
12:54with it. None of these are custom. These
12:55just come with the with the Hermes
12:58agent.
13:01You can see I'm using MIMO V2 Pro
13:03through Open Router right now. So there
13:05you go. That was the installation and
13:07the setup. Now we're going to go through
13:08a couple of the CLI commands that you'll
13:10use. So here are some of the top ones
13:13you will use. So it's important to go
13:16through these very quickly. Hermes just
13:20starts this chat UI uh TUI terminal user
13:24interface. If you just do Hermes, you'll
13:27do this Hermes. Hermes gateway starts
13:29the messaging gateway which you'll need
13:30to connect to Telegram. Hermes doctors
13:33to check for any issues that you may
13:35have.
13:37So I exited um it's just slashexit to
13:40get out because let's try a couple of
13:42these these CLI commands. So Hermes
13:45model
13:47uh just shows what you're using.
13:51Okay, you have to reload your shell to
13:52use the Hermes command. So let's do
13:54that. So you can see Hermes model will
13:56open up the configure and you can select
13:58which provider to easily switch through
14:00them. So we're going to stay with stay
14:02with two.
14:04Uh so we'll just get out of that sets
14:06for the model
14:08per me status status.
14:12Um and this shows you everything the
14:14whole configuration that you have here,
14:17where your project is, what model you're
14:20using, what provider you're using, your
14:22API keys. Um you can see which ones are
14:25set and which ones are not set. So this
14:27is very helpful. Hermes status you could
14:30see the O providers
14:33now uh new portal open AI codeex quen
14:37oath and it shows everything that's
14:39configured and most stuff obviously is
14:41not configured because I just set set
14:42this up um Hermes insights this is
14:46probably not going to have much but
14:48let's see Hermes insights since I just
14:51pulled it up yeah no sessions found
14:52because I didn't I just loaded it um so
14:55these are other things you can see
14:57Hermes sessions browse Browse uh the
14:59cursor pick uh curses picker to resume.
15:04So you can browse which session you want
15:06to resume. Hermes skill browse. You can
15:08discover and install skills. Let's do
15:10that.
15:12Hermes
15:14skills browse.
15:17There we go. We got a lot of skills.
15:20Uh 50 skills loaded.
15:23Have all these different ones. Canvas.
15:25You could see duck.go Go search
15:29honcho.
15:31They're all official, so there's no no
15:33concern about that.
15:36So you can actually search for them
15:38specifically. Let's say Hermes skills
15:40search. Say you want something about
15:42with email.
15:44Uh these have all email related and you
15:46can see the trust. Some are official.
15:48This agent mail is official. Then you
15:50have a bunch of community. Some are
15:52trusted which I guess are like this is
15:54anthropics
15:56skills which is obviously a trusted
15:59provider. You can see they have all
16:01these skills cold emailing react email
16:04business email.
16:06So that's a very useful skill if you're
16:07looking for a specific skill with
16:09Hermes. Hermes skill search and then
16:10whatever your query is
16:14Hermes config show which is going to
16:15show your configuration. Like I said
16:18Hermes doctor and Hermes update which
16:20pulls the latest. Let me see if we're
16:21actually updated yet. Burmese update. I
16:24think they may have had an update
16:26recently.
16:31It's updating some of the dependencies.
16:34Found new commits. Yeah, they just had a
16:36a new update actually while I was
16:39filming this.
16:41There you go. We're updated. Pretty
16:44easy. So, these are the top 10 slash
16:46commands and these are in chat commands.
16:48So, let's go back to the chat. And
16:50that's just Hermes chat, right?
16:53There we go. Back to the the terminal
16:57still with Mimo V2.
16:59So new, if you've used like claude code,
17:02you're familiar with this as well. You
17:04could think of this terminal interface
17:06as like a claude code type of thing or
17:08codeex or whatever you're using.
17:13um new starts a new fresh conversation a
17:17new session slashmodel you can swap the
17:20model mid session so that's a really
17:23nice feature let me see slashmodel
17:28um
17:30I'm going to switch to
17:33zi
17:35slashgm
17:37by one point
17:40model switched Okay.
17:44No, that's not the right one. Slashmodel
17:50ZIL
17:51SLGM
17:555.1. There we go. So, there we go. We
17:58switched. Now we're in GLM51.
18:00And I don't have to restart the session.
18:02Um, I can just continue. I don't have to
18:06re reboot this configuration.
18:09Here we go. Hello, I'm Hermes, your CLI
18:12AI agent. I'm running on GLM 5.1 model.
18:16There we go. Fast, which is a a brand
18:19new one that actually came out in
18:20today's release. I believe this is
18:22priority routing through OpenAI and
18:24Anthropic.
18:26I don't have those configured, but if
18:28you want like something quick, going
18:30through those is going to be faster.
18:34Background prompt, so you can do a
18:36background task and keep chatting.
18:39slash by the btw and then you can ask
18:42ask a question. So that's a side
18:44question while Hermes is working. Uh
18:48slash Q with the prompt which adds like
18:50a next term Q. This is also like when
18:52Hermes is working you can do that. You
18:54can try some of these a little bit later
18:56when I once I give it a task. Slash
18:57compress. So this manually compacts the
19:00context. If you have a long context
19:01window and you want to compress
19:03everything, you can do that. Skills, you
19:05can browse or invoke a skill. uh slash
19:08yolo toggle dangerous command approvals.
19:11Be careful with that. And of course
19:13slashhelp with everything else. And that
19:16is probably the most helpful. You can
19:17see it had a lot of stuff. So this has
19:20all the slash commands you could ever
19:21want. All the tools and skills.
19:24So if you ever need some type of skill
19:28or some type of command, just the
19:30slashhelp will tell you.
19:33There are a lot. So these are all skill
19:35commands as well. These are the ones
19:36that are installed so you can invoke
19:38them in this session
19:40if you need them. So this is where stuff
19:43lives. And just quickly everything is
19:46stored in this uh slash. Hermes file
19:50or directory I should say. Sorry, this
19:53is small. But if you go down, if you're
19:54in a Windows PC, you go down here to
19:57Linux and mine says de Debian, but yours
20:00might say Ubuntu and you go to home and
20:04then whatever your username is. And then
20:06you can do this hermes. And this is
20:08where all the files live. The soul file.
20:12Um, this has your history and this
20:14config. The YAML file is the main
20:18configuration and this is your settings
20:20and you can edit that. You can edit the
20:23ENV file because that's going to be your
20:25secrets API keys. This one, this ENV
20:28file, API keys, any kind of passwords,
20:31stuff like that, it's going to be stored
20:33there. Sessions has every past
20:35conversation.
20:37Skills, you can see sessions is over
20:39here. We've only had one, so there's
20:42only one there.
20:45skills folder here
20:50has all of these. These are the ones
20:51that come with the Hermes agent
20:53basically. Sorry, this is small. Let me
20:55see.
20:57You can see a little bit better here.
20:59And you can see here logs has your all
21:01of your agent logs and the error logs
21:03memories here. So, it's important at
21:05least to be aware of this Hermes
21:08directory in case you ever need to
21:10change anything here. And you can edit
21:13uh these top three config.yamel.env
21:17and memories. You don't want to touch uh
21:19the sessions or chrome by hand. Ask your
21:22agent to do it cuz you could break it.
21:25So your first real task live demo. So
21:28this was this video was teaching you how
21:30to install Hermes agent from scratch,
21:33how to set it up. We just did a basic
21:35setup. There's obviously a lot of
21:36settings we're going to be fooling with
21:38in the coming videos. Okay, so it's time
21:41for our first real task, and it's going
21:42to be to do some competitor research. I
21:45have what I think is a really great idea
21:47for a business, but I need to my Hermes
21:49agent to research the competitors and
21:51see if there's any market gaps here. So,
21:54in order to do this, the one tool it's
21:55going to need is web searchability.
21:57So, what we're going to do is tab. They
21:59have a really good free monthly credits.
22:01You can also, they have duck.go go
22:04available for free. But just to show you
22:06another option here is uh taby.com
22:10and then you just sign up. You don't
22:12need to put in a credit card or anything
22:13and you get a,000 free credits which is
22:15pretty good. But we'll look in more deep
22:18into web search options later on.
22:22But here you get your API key here. Save
22:25that. Then back in your WSL you do
22:29Hermes config or Hermes setup actually
22:33quick setup configure missing items only
22:36and these are all the items that are
22:38missing but Tavly API Okay.
22:59API key. Uh, space to toggle that. Enter
23:03to confirm.
23:05And then you put in your API key. Just
23:07paste it there.
23:09set up.
23:11Not going to do that right now.
23:13Okay, so that should be set up. And I
23:16checked here. I did Hermes status. You
23:18can see there we go. Tavi is set up. So
23:21we should be able to use web search now.
23:23So let's get into Hermes then. Hermes
23:25chat
23:29uh using Mimo V2 Pro still. So this is
23:32my idea. Here we go. There we go. I had
23:36an idea to start a business that uses AI
23:38to make custom children's books. Instead
23:41of just the name, it can use AI to
23:42create custom illustrations in the style
23:44the customer requests as well as story
23:46themes, etc. Please do a thorough
23:48research into other companies and
23:50competitors who may provide this service
23:52and report back to me with their names
23:54and try to identify any market gaps that
23:56exist. So you can see it's initializing
23:58the agent here
24:03and it's preparing the web search to
24:05find
24:08got some good leads. You see these are
24:10all the tool calls it's it has
24:13I think this is a good business idea.
24:15It's a real one I had. Hopefully we can
24:18find some kind of market gap.
24:24Uh okay. And here we go. We got our
24:26report back pretty quick as you just
24:28saw. Um here's a thorough breakdown of
24:30the competitive landscape. Competitive
24:32landscape market falls into three tier
24:35tiers.
24:36Uh the traditional personalized book
24:38companies which were pre- AI
24:41but let me see the AI powered ones. Here
24:43we go. Direct competitors. These are the
24:45companies doing what you describe using
24:47AI for both story and illustration
24:48generation.
24:50Uh oh, there's a lot. This might be a
24:53saturated market.
24:55might not be
24:59may not be workable.
25:01There's a lot doing this. Like this
25:03first one, one of the earliest AI
25:04children books platform template based
25:06stories AI illustrations.
25:09H
25:12okay. Tier three general AI tools
25:14indirect. Yeah, these are just visual
25:15ones. You can see here just that quick
25:17search it had found a lot of
25:19competitors, market gaps and
25:21opportunities. There are some uh based
25:24on competitive analys here are the
25:25underserved areas. Custom art style
25:27selection. So most competitors let you
25:29customize the character but lock you
25:32into one illustration style. Few let the
25:34customer chooser to find the actual art
25:36style.
25:38Uh the user said in the style the
25:39customer request this is the real gap.
25:42That could be something right. Older age
25:44groups almost everyone targets ages 2 to
25:48seven.
25:50Uh, I really wanted children's like baby
25:52books. Those are the ones everyone's
25:54buying as like gifts. Bilingual.
26:00It found a lot of uh gaps here,
26:03but I don't know if any of these are
26:05workable for me. Audio first, culturally
26:08specific
26:09author creator tools. That's more on the
26:11B2B side. Um, so here we go. Wonderbly.
26:17I see me.
26:20So, there's actually a lot. So, I gave
26:21it my my big take. The biggest white
26:24space is custom art style control
26:26combined with therapeutic niche story
26:28themes and older age groups. So, there
26:32we go. There's your kind of market
26:34report, competitor report. Just for this
26:36first video, we did something simple,
26:37but
26:39I think this is a a task that a lot of
26:42people have, right? We all have random
26:43business ideas. Maybe it's just me. And
26:46you can use your Hermes agent to uh
26:48check on them, check on the competition,
26:50see what kind of market gaps exist and
26:52try to plan when you want to start a
26:55business. And there you have it. That's
26:57going to be the end of this first video.
26:59Uh right now you should have Hermes
27:00installed and a provider configured. You
27:03know the top 10 CLI commands and top 10
27:05slash commands. Got our real first
27:08artifact com the competitor report and a
27:11memory file that just started learning
27:13you. So, this has already started
27:15learning me just with this first task.
27:18You could see.
27:21So, we just started this blank right
27:23here. And throughout this master class
27:25series, you're going to be seeing how
27:27this Hermes agent that just started
27:29brand new, how it develops over time as
27:31we continue to use it and work on it
27:33with different tasks and actually try to
27:36start some businesses. I'm not sure
27:37about this AI children's book one, but
27:40maybe other ones that has left less
27:42competition. That's going to be it for
27:44session one. I hope you enjoyed it. Next
27:46session, we're going to talk about
27:47different deployment options and
27:49messaging. We're going to hook it up to
27:50Telegram, maybe to some other apps so
27:53that you can see your options with that.
27:55And you don't have to just chat with it
27:56in the um in the terminal like this.
28:01Uh but thank you for watching. All these
28:03videos will be available on my school
28:05community once it launches which should
28:07be in June. But I'll be I'll be um
28:10presenting some of these on YouTube as
28:12well. So you may be watching it there if
28:13you so if you want to watch the full
28:14series from day one in the school
28:16community. It'll be available all of the
28:18episodes, but I'll be posting some on
28:20YouTube. So if you're interested in this
28:22content, please subscribe. Please leave
28:24a like, leave a comment. Let me know if
28:26you have any questions about setting up
28:28Hermes like this. and I will see you in
28:30the next