0:03 Hermes is the world's most powerful AI
0:05 agent assistant. But if you combine it
0:08 with Claude code, you can create a
0:11 powerful 24/7 AI employee solving Hermes
0:14 and Open Claw's largest problem. And in
0:16 this video, I'll show you exactly how to
0:18 set Hermes up from scratch and how to
0:20 give it a visual intelligence layer,
0:23 letting it improve itself based on how
0:26 you use clawed code and vice versa.
0:28 meaning you have a universal AI
0:30 intelligence system that will save you
0:33 time, make you more money, and will get
0:34 you light years ahead of your
0:36 competitors. And if you're new, my name
0:37 is Jack. I built and sold my last tech
0:39 startup with the gazillion customers.
0:40 Now, I'm building my own AI companies,
0:43 and I just share here the stuff that
0:44 actually works. So, if you haven't
0:46 already, grab that beautiful coffee and
0:48 let's dive straight in. Now, the Hermes
0:50 AI operating system actually connect to
0:53 your Clawude Code operating system. And
0:54 I'll be honest, this is one of the
0:56 coolest things that I have actually
0:58 built and it will blow you away when I
0:59 show you the detail. But let me just
1:01 start by explaining why we're using
1:03 Hermes or Hermes. We have that's we have
1:04 some dispute on how we actually
1:06 pronounce this and claude code and why
1:08 this is something that everybody who's
1:10 ever using Hermes actually needs to get
1:12 their hands on. So first thing about
1:14 Hermes is that it has gained 60,000
1:15 stars, okay, in the last couple of
1:18 months. It's way over six figures on
1:20 GitHub, which is just internet updates.
1:21 We like this. Basically, it's one
1:23 folder. is yours and it's got loads of
1:25 gateways. It is an agent that dreams,
1:27 that thinks, that self-improves. That's
1:29 why it's gotten so much hype. But this
1:31 leads onto the huge problem that we have
1:34 and it isn't just Hermes. It isn't just
1:36 open core. Any individual AI assistant
1:38 has no handshake. So the two ways that
1:41 we do work with AI at the moment that
1:43 are trailblazingly getting better is we
1:46 sit down and we use codeex or we use
1:48 cloud code. That is this guy over here
1:50 sat at the desk, right? He's coding.
1:51 He's building applications. He's talking
1:53 to Claude on his computer. He's talking
1:55 to anti-gravity. Awesome. Then over here
1:58 on the right hand side, we have our AI
1:59 assistant, our AI agent. So, we're
2:01 talking to on Telegram. This is Hermes.
2:04 This is Open Claw. This is any different
2:06 AI assistant. And whether you're using
2:08 Gravity Claw, Open Claw, claw claw, any
2:09 of these things, what I'm going to show
2:11 you applies to everything. Now, the
2:13 issue is there's no handshake. In other
2:15 words, the stuff we do over here never
2:17 connects to the stuff that happens over
2:19 here and vice versa. which means that
2:21 you never actually have an AI any AI
2:24 system that has a full overview of your
2:26 entire world. And that means that you
2:28 lose complete context everywhere you go.
2:30 And that's just the beginning of the the
2:32 issues that we have with this. So you
2:33 have Claude that knows about your repos,
2:35 your ND preferences, bug fixes you run
2:37 at 2 a.m. and all the software and
2:39 thoughts and conversations you have with
2:40 it. Then you got Hermes that knows your
2:42 brain dumps and Telegram your
2:44 conversation. The problem is that you
2:45 message an idea to Hermes on the walk.
2:47 You open cloud code, you type it again,
2:48 it's never seen again. Which is why the
2:51 idea of the Hermes operating system that
2:53 actually fully connects to the claude
2:56 code operating system actually solves
2:58 this problem and unlocks some
3:00 capabilities that you cannot actually
3:02 use otherwise, which is freaking
3:03 incredible. And I'll show you why that
3:05 is. So you think about this, Hermes
3:07 hears it, Claude codes knows it, and
3:09 vice versa. So Hermes itself will know
3:11 everything that you've done on clawed
3:14 code as a shared memory and vice versa
3:15 and it also unlocks some really cool
3:17 capabilities. And again the same thing
3:19 happens whether it's open claw or any
3:20 different variation. We're going to be
3:22 using Hermes in this video. Now I'm
3:23 going to show you exactly how to set up
3:26 your Hermes in three steps with this
3:27 beautiful system and all its new
3:29 capabilities. But to do that we have to
3:31 understand what this incredible
3:33 visualization layer is and what that
3:35 enables us to do that we couldn't really
3:37 do before. So this is the Hermes
3:39 operating system. It actually connects
3:41 to the Claude code operating system.
3:42 I'll put a link down below so you can
3:44 check this one out. The Claude Code
3:45 operating system shows your AI spend
3:48 across every single model. DeepSeek
3:50 Chatgbt Claude gives you a breakdown of
3:52 your usage. It actually dreams for you
3:54 overnight. So based on every message
3:56 you've ever had with Claude, it will
3:58 give you improvements about, hey Jack,
4:00 you're rooting 87% of the work to Opus
4:03 4.7. Most of it's hiker territory. So, I
4:04 actually built a system here where
4:06 called dream for you overnight based on
4:09 all this information that happened on
4:10 your computer. And you'll see why this
4:12 is so important for her in a second
4:13 because not only does this have every
4:15 model you're using, all your chat logs
4:17 from any AI using your computer, it has
4:19 all that usage data. It has these
4:22 beautiful independent memory systems,
4:23 okay, that are all connected together,
4:25 all your connections and all this data
4:27 and your usage. Now, what's cool is we
4:29 can actually leverage that in Hermes. So
4:30 if I come over to Hermes agent on the
4:32 left hand side here, you can see it
4:34 shows me all my connections. It tells me
4:36 the version I'm on. It tells me what
4:38 model I'm currently using, memory, my
4:40 weekly streak, and I have these
4:42 wonderful chat windows. So for example,
4:44 here I can just chat with Hermes if I
4:45 want to in the browser. But this is
4:47 where it gets crazy is you have this
4:49 thing here called the Pantheon. Now the
4:51 Pantheon is essentially these custom AI
4:52 personas that I actually built with
4:54 Hermes. It's really freaking cool. And
4:56 we'll get more into what these do in the
4:58 video, but effectively what they let us
5:01 do is assign models and roles and
5:03 personalities. For example, let's say
5:05 that you want to do some deep research,
5:07 right? And you want to use a specific
5:08 model for that. You may want to use
5:10 Labyrinth or if you want to do some
5:12 autopilot and crown stuff or maybe you
5:13 want to philosophize and do some deep
5:15 reasoning and you want to pull up a
5:16 specific model. Well, we can add these
5:18 and create these here. give it a name,
5:20 effectively a description, a system
5:22 prompt, tell them which model you want
5:24 to use, and then give it a name in in
5:26 brand basically. And I can even add
5:28 these personas by clicking this button
5:29 here, selecting who I'd want. So let's
5:31 say for example, I want to go with the
5:33 Oracle. I can describe the name, the
5:35 description, a system prompt, a model,
5:36 and you know, let's say that I want to
5:37 do loads and loads and loads of deep
5:38 research over the night. I may want to
5:41 use Deepseek model or a free model. I
5:43 can actually specify this. And what I
5:45 can do is share this then with Hermes
5:48 agent. And effectively then if I say hey
5:50 go and do X or use a philosopher or use
5:52 Mercury or Labyrinth it will do that for
5:54 me using the models the systems and the
5:56 prompts without me touching a single
5:58 thing. So we build out this pantheon of
6:00 different characters and models it can
6:02 use. And again this is a real simple
6:03 system. I'm going to show you how we run
6:04 this all the way through in the video.
6:06 Now what's really cool here is all the
6:08 memory from Hermes is reflected here.
6:10 This is an example one. We're going to
6:11 set it up completely for fresh together.
6:14 We have a user profile. What the agent
6:15 knows about us and its soul. We can
6:18 connect Obsidian to Hermes as well,
6:19 which is fantastic. And then we've got
6:22 this Claude OS bridge. And effectively
6:23 what this does is gives all of the
6:25 information that I showed you earlier in
6:27 this section here on your computer from
6:29 all the models using all your knowledge
6:30 systems, all this beautiful stuff, and
6:32 brings it over to Hermes, which is
6:34 fantastic. So we can bridge them
6:35 together by using this real simple
6:37 install prompt. And at the bottom, you
6:39 can see effectively all of your skills
6:41 that exist within Hermes and what that
6:42 looks like. And then a quick little
6:44 hacks at the bottom. This is the idea of
6:47 visual intelligence, a operating system
6:50 that combines your entire world of
6:52 connected intelligence. And that sounds
6:54 super fancy, but what it just basically
6:56 means is you have everything now in one
6:58 place. Now, you can build this with all
6:59 the knowledge that sits within your
7:01 computer and Hermes. I've also put a
7:03 link to all of this inside the
7:04 community, the claw code operating
7:06 system, the Hermes agent. So, you can
7:07 literally uh just go ahead grab it and
7:09 then you're like running with it
7:10 basically. It even has a really cool
7:11 setup wizard that runs you through
7:13 everything, gets your name, gets your
7:15 profile photo, finds all of the apps
7:16 that you have. So, for example, it will
7:17 find for you all the things that you've
7:19 got. Uh, and then you can go through the
7:20 whole setup, and it's all kind of like
7:22 synced up. So, it's like plug-and-play.
7:23 I actually took me 7 hours on the
7:24 onboarding alone just to get this set
7:26 up. Just to give you an idea of like
7:28 what goes into building this, but you
7:30 can do all this stuff with everything on
7:31 your computer. But the bottom line is
7:33 you want some kind of beautiful visual
7:34 innovation dashboard to connect your
7:36 worlds together. So, now we understand
7:38 the visual intelligence layer. The next
7:40 step is to actually install Hermes onto
7:43 our computer so we can begin connecting
7:44 everything together. Now, I'm going to
7:46 come over here and grab on Hermes. Now,
7:47 the way we get this done, let me just
7:49 jump out and exit the demo so I can show
7:50 you this. Now, I did a video explaining
7:52 that you can essentially connect
7:53 everything together where we would
7:55 leverage bits of Hermes, bits of open
7:58 claw, and I do believe that is the best
8:00 way to get a custom AI agent model. The
8:01 reality though is that this is the
8:04 fastest way to go from zero to one to
8:06 get a working agent in the system using
8:08 Hermes and it's very good. But if you
8:09 want the best of the best, you're going
8:10 to want to leverage different aspects
8:12 from different softwares and build it as
8:13 you go. I'm going to do something in the
8:16 community on c building out those custom
8:17 beautiful agents. But just so you
8:19 understand how this fits inside the
8:20 matrix. So what we're going to do is
8:22 come down and grab this code here. It's
8:24 also available on the Hermes website, of
8:25 course. And you can even just say to
8:27 Claude, "Hey, install this, please." So
8:28 we're going to head over to
8:29 anti-gravity. And if this sounds like
8:31 I'm speaking Spanish, check out this
8:32 master class on screen. I run through
8:34 everything and I won't fear when you
8:35 come back. Welcome back. So now we got
8:36 the terminal. All we're going to do is
8:37 come down here and paste in the code,
8:39 which is fantastic. Of course, in
8:40 anti-gravity or claude or whatever
8:42 terminal top terminal, it's going
8:44 through and testing everything, which is
8:46 great. And then literally from this, we
8:48 can just begin the full setup journey
8:50 inside this terminal window. Okay,
8:51 beautiful. So once we've entered in the
8:53 code, we can pick everything we want to.
8:55 I've chosen OpenAI codeex. If you don't
8:57 have that, by the way, it will just open
8:58 it up and run through them. We're happy
8:59 with this one. So, we're going to head
9:00 and click enter on that, which is
9:02 fantastic. This is good because it's
9:04 free with your chat GPT subscription.
9:05 So, that's decent. The other one that we
9:07 want to have in there is open rout
9:09 obviously, but we'll start with CEX for
9:10 now. Hit on enter, which is fantastic.
9:12 Then, we decide we're going to use our
9:14 existing credentials, which is great, or
9:15 we can just reauthenticate. So, for us,
9:17 our existing credentials are fine. GBT
9:19 5.5 is awesome. Our text to speech
9:21 provider, we're happy using the current
9:23 one for now. Terminal background, we're
9:24 going to keep it here and run it on our
9:26 beautiful home computer. We're not going
9:28 to run this one on a VPS. This is going
9:29 to be fine on our home MacBook which is
9:31 running all the time. Click on current.
9:34 Max iterations is 60. That's fantastic.
9:35 Then we click on all for this one.
9:38 Compression threshold is fine. 0.8 is
9:40 good. It just means the closer it is to
9:42 0.95. Basically, the less often it
9:44 compresses. If it compresses too much,
9:46 it can be a little bit more forgetful.
9:47 So, we're happy with the slightly higher
9:49 rating there. And then for session reset
9:50 mode, we're going to ahead and keep
9:51 current settings for that. And then
9:53 platforms to configure. We're going to
9:54 select Telegram. I'm going to click on
9:55 enter. And it's going to ask for
9:58 Telegram bot token. Now, we use Telegram
10:00 because it is a gorgeous API. It's so
10:01 easy to use. To do this very quickly,
10:03 just open up Telegram. Download the app.
10:05 Get the bot father. I'm going to do for
10:06 slash and we're going to bring him over
10:07 here so you can have a good beautiful
10:09 look. And we're going to say new bot.
10:10 Click on this guy over here. Great.
10:12 We're a new bot. We're going to call it
10:13 Hermes bot. Great. Let's choose a
10:15 username. So, go ahead and pick a random
10:16 name for that bot. So, I could call it
10:18 something like this. And you see guys,
10:19 you'll get a token. All you're going to
10:20 do is come back over here and just enter
10:22 in that token. Hit enter. And then it's
10:23 going to ask you for something really
10:25 crucial, which is allowed user IDs. What
10:28 this effectively means is that only the
10:30 IDs that you give it can message your
10:32 Hermes bot, which is fantastic, which
10:33 means unless they have your mobile
10:34 phone, they're signed in, they can't
10:36 message your Hermes bot. So to get that,
10:37 come back over to Telegram and you're
10:39 going to search for account called user
10:41 space info. Drop it any message you want
10:43 to or for/start and it will give you all
10:44 the information and you're looking for
10:46 your ID and all you're going to do is
10:48 literally copy that ID, come back over
10:50 and throw it into the terminal. Then of
10:51 course you're going to go ahead and
10:53 confirm your user ID by pressing Y. Once
10:55 you've done that, it will ask you if you
10:57 want to launch it as a service. Just say
10:58 yes. Then you're going to come all the
11:00 way down here for example, hit on done.
11:02 And then we are finished. Then guys, you
11:03 can actually come back to the bot
11:05 father. Click on the link to open a new
11:06 tab. And you should see your Hermes bot
11:08 here. So I can click on start. So now we
11:09 can actually begin our conversation. I'm
11:11 going to say hey there. Like so. Drop
11:12 this guy a message and see if he comes
11:14 back to us. He's typing. He's having a
11:16 think. And we can have a little a little
11:17 play around with what he's going to say.
11:19 And this should be Hermes bot. And look
11:20 at this guys. He's come back. Hey there.
11:21 I'm Hermes. I can help you with
11:23 anything. you are fully ready to rock
11:25 and roll at Hermes is officially
11:26 downloaded on your computer which means
11:27 that you can use this on your phone
11:29 wherever you want to. So we land into
11:31 the dashboard which is awesome. We can
11:33 see all of our connections here in our
11:36 global universal AI intelligence.
11:37 Obviously if I click on Hermes I can see
11:39 all of Hermes integrations which are
11:40 looking very lonely right now to be
11:42 fair. We've got the model we've got the
11:43 active model that we're using right now
11:46 which is GBT 5.5 and then we've got what
11:48 we used this week. Now here we've got a
11:50 chat window which is very very cool and
11:51 effectively this shows us all the
11:52 conversations that we have. So this
11:54 shows hey there's a conversation on
11:56 telegram you can just talk to it here if
11:57 you want to obviously it's part of the
11:59 operating system or you can talk to it
12:01 directly within basically um you can use
12:02 it within telegram if you want to. So
12:04 you see if I bring telegram up for
12:05 example and I give it a prompt something
12:06 like this. Hey there I would like to
12:08 connect to GitHub um let me know
12:10 anything you need from me in order to
12:11 connect to GitHub just so we can start
12:13 doing some very cool things. Paste that
12:15 one off like so. Send that off. And what
12:16 you'll see now is this will now appear
12:19 for us actually within our own our own
12:20 desktop environment and also of Hermes
12:22 intelligence system. And interesting
12:23 guys, it's come back and it's done a
12:25 skill view git or it's checked out
12:26 Hermes agent. It's checked all the
12:28 different versions. And because I have
12:30 this connected part of the CLI, our
12:32 command line interface, it's already
12:33 connected. So I'm actually going to
12:35 update the Hermes bar here to include
12:37 all your CLIs as we speak. So now we got
12:38 this. I might say something like, "Hey
12:41 dude, quote me a brand new repo called
12:43 Tango Blast, please." and let me know
12:44 the link to that repo once that is
12:46 complete. Send that one off just to
12:47 validate it's got the skills. Send that
12:48 one off and you can see him is working
12:50 in the background. And dude, look at
12:51 this. It's created the repo for us
12:52 literally within the chat. And would you
12:54 believe most importantly, we now see
12:56 GitHub in the terminal which is in our
12:57 command center which is great. Then we
12:58 click this guy right here and open him
13:01 up. And we can see this is Tango Blast
13:02 just been created. Fantastic. So now we
13:03 got that locked down. The next thing
13:05 that I want us to do now is take a look
13:07 at building out persona. This is really
13:08 important because if you're doing
13:11 certain tasks like we don't need Albert
13:13 Einstein to wash off to mop our floors,
13:14 right? We want Albert Einstein
13:16 scribbling not making much sense on a
13:18 whiteboard in a very esoteric area of
13:19 practice that only five people can
13:20 understand. So to do that we're going to
13:22 come down here and install the Pantheon
13:23 which is very cool by clicking on this
13:25 guy right here. Then it should pop up
13:26 and we can begin it. So we've got
13:28 Labyrinth. These are a couple ones I
13:29 thought were really cool just to get
13:30 started if you just want to get an idea
13:32 of what it's like. All you do let's say
13:33 we've got the philosopher here which is
13:35 fantastic. We got a description for
13:36 wrestling with ambiguous problems, pulls
13:38 and threads, questions, premises, etc.,
13:39 etc. You are the philosopher. Treat
13:41 every question as a starting point, not
13:43 an instruction before answering surface
13:45 meta question behind the question. I
13:46 think that's very very freaking cool. We
13:48 can add a persona here if we want to. Of
13:50 course, say we like the, you know,
13:52 Orpheus or whatever. We can add various
13:53 different things that we like. We're
13:55 going to go ahead and cancel that. Now,
13:56 what we want to do here as well is back
13:58 up her. So, to once we've built out all
14:00 the different personas we want to see,
14:02 maybe one for research, one for tool
14:04 calling. So, anything that's doing deep
14:06 research, like for your morning briefs,
14:08 I want you to delegate that to a free
14:10 model, like an absolute like you don't
14:12 need your strongest models doing that,
14:14 which is why, for example, autopilot and
14:16 cron, I like to use Mercury because it
14:18 just does all that stuff for me in the
14:19 background. So, what I'm going to do now
14:20 is I've got the stuff. The first thing
14:22 that we need to do to activate the
14:24 Pantheon is take Hermes anywhere. So,
14:25 look at this. Mirror your Hermes for
14:27 private GitHub repo. So, your config and
14:30 personas survive a machine swap. Every
14:31 edit is version and you can roll it back
14:33 if something starts misbehaving just in
14:35 case it goes wrong. Two prompts. Paste
14:37 each one into your Hermes Telegram chat
14:38 or any Hermes session and she'll walk
14:40 you through the rest asking for what she
14:41 needs. Beautiful. So to connect the
14:43 Hermes to GitHub again, just copy paste
14:44 this one. We can throw it in the chat
14:46 window above or we got Telegram here if
14:48 you want to come over here. You can come
14:50 down and literally drop that bud in and
14:52 let it run wild and do its own thing.
14:53 Then it'll start to think and
14:54 effectively just correct this GitHub
14:55 repo for us. And it's really important
14:56 because one of the things that can
14:58 happen when you're doing all these
15:00 different features is it can start to
15:01 you know forget things and just like
15:02 that it's come back. So absolutely I'll
15:03 wait for your confirmation before
15:05 creating repos and copying files. What I
15:07 need from you it's got the GitHub CLI
15:09 already. Git username it's got that repo
15:11 name is awesome backup. I'm going to say
15:12 sounds great. Go ahead and create all
15:13 that for me please and let me know when
15:15 that's done. And then I want you to set
15:17 up a chron schedule. Let's say I don't
15:19 know 11 p.m. every day. You're going to
15:20 do a full update for me please.
15:22 Beautiful. So now we have this thing.
15:24 We've created a mirror of this in GitHub
15:26 which we can use anytime we want to. The
15:27 next thing that we need to do then is
15:29 basically go ahead and recreate
15:30 everything that we just made in the
15:31 pantheon. So all we're going to do here
15:34 is to paste this into Hermes. Um after
15:35 we've connected the repamp she'll push
15:37 the latest Persona Labyrinth mercury
15:38 philosopher and anything that we've
15:40 added and we're basically explaining
15:42 where this is what it looks like and how
15:44 everything sort of connected. So going
15:45 to copy and paste this like so. I'm
15:47 going to copy that. Come over drop that
15:48 one in. And then we're ready to rock and
15:49 roll. Beautiful. And now it's come back.
15:51 It's actually found everything there.
15:53 Now, here's the cool thing is the way
15:54 that I've set this up, guys, is that
15:55 when you add these, and if you build
15:57 yourself, I recommend you do it the same
15:59 way as well, is that effectively what
16:00 this means is when you build these
16:01 different personas, as and when you
16:03 change them in your operating system,
16:05 it's automatically changed within
16:07 Hermes. Because the way this works is
16:09 these all live in a folder. So, if I add
16:11 new files to that folder, Hermes knows
16:13 where to go in the folder. So, if I
16:15 change the job of the philosopher, he
16:16 does a great job, but he already changed
16:18 his job. Same with Mercury over here. Or
16:19 I add new ones, that's all going to be
16:21 updated in the same location. So it just
16:23 works fine and perfect for us anyway.
16:25 And what this saves us doing as well is
16:26 repeating the same thing a gazillion
16:28 times. So going to say that all sounds
16:29 great. Go ahead and do that for me,
16:31 please. Beautiful. And just like that,
16:32 we're all ready to rock and roll. And
16:34 now when we add new different um
16:36 basically characters and players to our
16:37 pantheon and gods to our pantheon, it'll
16:39 be ready to rock and roll. And then for
16:40 example, guys, let's say I want to
16:42 invoke Labyrinth. I might say something
16:43 like, "Hey there, go ahead and use
16:46 Labyrinth and do some deep research on
16:49 the best strategies for email marketing
16:50 and I would like five of the best
16:53 headlines for a roofing company, please,
16:54 when trying to get new clients." Send
16:56 that one straight out. And as you can
16:57 see, guys, as it's now delegated the
16:59 task, and you can see now our basically
17:01 Labyrinth, Mercury, and Philosopher are
17:02 fully synced, which means it's linked
17:04 and connected now to our new agent.
17:05 Beautiful. And we've summoned Labyrinth,
17:06 Labyrinth has gone ahead and done this.
17:09 And of course if we were using Opus 4.7
17:10 it would have gone ahead and used a
17:12 different model to do that for us. And
17:13 look at this catch storm water weather
17:15 demand offer a low friction next step
17:17 segment by home in a situation lead with
17:19 trust proof. It's given us good details
17:20 and it's gone ahead and created that for
17:22 us which is fantastic. And now we've
17:23 covered exactly how to connect Hermes to
17:25 anything. That's great. But now we need
17:27 to take this to a completely new level.
17:29 And if your system doesn't have this
17:30 information if your Hermes agent doesn't
17:32 get those it is not performing to its
17:34 potential. We're going to give it an
17:37 incredible memory system to obsidian and
17:38 then I'm going to show you how you can
17:40 supercharge it with an incredible skill
17:42 that most people haven't even heard of.
17:43 Beautiful. So, the first thing I want to
17:45 do now is connect is aic memory. I
17:46 covered this in the full course I did
17:48 inside my community that goes foundation
17:50 setup all the way down to power features
17:52 memory systems this system and loads of
17:54 other cool stuff like monetization. I'll
17:56 put a link down below if you'd love to
17:57 check that one out. But the cool thing
17:59 here that we talk about a lot is this
18:01 Obsidian system right now. If you have
18:02 in Obsidian installed, it looks a little
18:04 bit something like this. You can see
18:05 I've got my memory core here and it
18:07 shows me everything. I've got my desktop
18:09 Obsidian. Really freaking cool. I just
18:11 get to visualize all my data, which is
18:12 fantastic. Now, what we can do first of
18:13 all, let's come into Hermes. And as you
18:15 can see, these are the memories that
18:16 it's starting to gain for me. I wanted
18:18 to install this fresh view on a desktop.
18:20 So, you can see user has her pantheon.
18:21 As you can see, this is the agent
18:23 memory. This is the user profile. And
18:25 then over here is the soul, which we can
18:26 build out shortly. Now, what you want to
18:28 do is just clarify that you have a
18:29 vault. to come down here and click I've
18:31 I've got a vault. Then what you're going
18:32 to do is literally come down here and
18:33 you're going to copy this code right
18:35 here. And if you don't know the exact
18:37 destination of your vault, you can just
18:38 ask cla code by copying this and
18:40 throwing it into cloud code. So here for
18:41 example, I've got it here. I'm just
18:42 going to paste that in. And I'll
18:44 probably say something like, hey, I was
18:46 doing a deep YouTube strategy. Just find
18:48 me the fault location, please. Just like
18:49 that. And we can send this one off. And
18:50 then we've got it here. So we just copy
18:52 this. It comes straight back over. Throw
18:54 in the path like that, which is
18:55 fantastic. And say I run it now.
18:57 Confirm. And said Obsidian is connected.
18:59 Just to double check this, we can
19:00 literally come back up and drop a
19:01 message over to Hermes agent if we want
19:03 to. And as you can see here, for
19:04 example, I said, "Hey, I've added it.
19:05 I'd like to reference this Obsidian
19:07 vault when answering questions, please.
19:08 Here is a file location. Save this as a
19:10 memory." I gave it to it and it said
19:11 saved. I'll use this wiki as your
19:13 Obsidian Vault and reference searcher
19:15 where my questions may benefit from your
19:16 notes. So now we've connected Obsidian
19:18 to Hermes. Awesome. Next thing we want
19:20 to go ahead and do then, guys, is shoot
19:22 over to the home section and you come to
19:24 the homepage. You can see now that all
19:25 the conversations you have with Hermes
19:28 are now fed into your main operating
19:29 system. So if you're using a clawed code
19:31 operating system, you can actually have
19:33 all these knowledges when it's giving
19:35 you these really cool recommendations.
19:37 So it has that full level context when
19:38 it's making decisions, which is really
19:40 cool. Then if you come back over to
19:41 Hermes over here and then you scroll
19:43 down, what you can actually see at the
19:45 bottom is this really cool thing. Let me
19:46 go ahead and grab it for you. It's this.
19:48 It's Claude OS bridge. So this lets
19:51 Hermes read the dashboard on request and
19:52 ask what did my dream say or what's
19:55 inside my cloud operating system it
19:57 basically helps it understand cuz if you
19:58 when you use cloud code and different
20:00 models it actually saves your chat logs.
20:02 So what this lets it do is query that
20:04 and think about that and dream and think
20:06 about well actually Jack's been talking
20:08 to cloud code a lot about these kinds of
20:10 topics maybe I should bring this in or
20:12 uh there's a really key piece of context
20:14 over there that are never mentioned to
20:15 me. So we effectively connect the
20:17 intelligence together which is awesome.
20:19 We've got a full prompt. Basically, tell
20:21 the you can kind of explain it like what
20:22 you're trying to do if you want to or
20:24 copy a prompt completely your call. It
20:25 just makes it a lot easier to do. Then
20:26 effectively come over to the chat. I'm
20:28 going to drop this bad boy in like that.
20:29 Send it off and let it do its work or
20:31 basically explain, hey, I want you to go
20:33 through my entire computer. I find all
20:34 the stuff that's relevant that could be
20:36 helpful in doing this. And I want you to
20:38 set up a cron job that every night
20:40 you're going to think proactively about
20:42 ways in which I can improve based on all
20:43 these different conversations and things
20:44 like that. That's effectively the next
20:46 prompt which is going to be like this.
20:47 Hey though based on all the information
20:49 that you have in our conversations and
20:52 also that you glean from the computer
20:54 based on these files I want you to
20:57 delegate a cheap model to reflect or use
20:59 GBT 5.5 on ways in which I could
21:01 potentially improve and be better and
21:03 come back in the morning with a couple
21:05 suggestions of that I could potentially
21:07 action two to three max do that at let's
21:10 say 8:00 a.m. Dubai time alongside a
21:11 beautiful morning brief for me and just
21:13 like that it's come back and done it.
21:14 It's that crown job for us. It's going
21:16 to use the metal llama model. I don't
21:17 mind it using the chat GBT model
21:19 actually because we've got it all in our
21:20 subscription. Hey dude, just do me a
21:23 favor. Actually, use my chat GBT 5.5
21:24 subscription, please. I think that
21:26 should be fine for this particular use
21:28 case. Awesome. Send that one off. And
21:29 just like that, now this is every
21:31 morning going to be thinking proactively
21:32 based on all the conversations across
21:34 everything what we want to do. And this
21:36 very nicely takes us on to some
21:38 beautiful use cases. So, we know loads
21:40 of the classic ones like, hey, why don't
21:42 you go ahead and set me brief? We get
21:43 that. But let's talk about something
21:45 that's going to grow your business or
21:46 grow your client's business. Let's say
21:47 that we're out and about. We're working
21:49 with Telegram. Now we have this
21:51 fantastic connected internet ecosystem
21:52 of things. Why don't we look about
21:54 getting leads? And before I double click
21:56 on what that would look like, I just
21:57 want to confirm what we've done here.
21:58 Right, we've got Hermes, which is great.
22:00 We've downloaded that. Second thing we
22:02 did is we got the Hermes personas. We
22:04 connected it. We did loads of beautiful
22:06 things. And the third thing we did is we
22:08 built a beautiful skill together. And
22:10 one other thing that we can do before we
22:11 actually build the skills out is to come
22:13 back over to the Hemy's agent and
22:15 effectively we can say, "Hey there, I
22:18 would like you to build a soul.md." I'm
22:19 correct in assuming that your soul.md
22:21 basically is all the key information
22:23 about me. So I'm going to brain dump
22:24 some information and I want you to
22:26 retain that in our conversation so you
22:28 can give me better results. Is there
22:29 anything you'd like to know from me? And
22:31 then literally just brain dump
22:32 everything to it and this will then
22:34 appear in your sold MD and you can track
22:36 that. So, one of the skills I wanted to
22:37 take a look at here is going to be a
22:38 lead scraper. And we're going to go
22:41 ahead and use Apollo. Apollo is pretty
22:43 much the gold standard when it comes to
22:45 actually finding people. So, one of the
22:47 best ways to grow your business is to
22:49 find people that can help you. This has
22:51 got a gargantillion features that we can
22:53 just go and use. But what I'm most
22:55 interested in here is actually the API
22:58 because sometimes we just have ideas of,
22:59 you know, you'll find it yourself when
23:00 you're having coffee. I I want to target
23:02 that person, but maybe it's not
23:04 convenient to have your desktop out and
23:06 start building campaigns. Maybe we just
23:07 want to check to Hermes and say, "Dude,
23:08 go and find these email addresses, find
23:11 out who they are and like set create a
23:12 campaign for me and get this, this, and
23:14 this." And we can now do that. That's
23:15 one of the coolest features cuz the
23:17 delta between idea and average should be
23:19 small. And when you know how to connect
23:20 things like Apollo, you can connect
23:22 anything. So to grab this, we're going
23:24 to grab your API key from Apollo. Great.
23:25 So to find this, come down here to admin
23:27 settings right at the bottom and then
23:29 click on integrations. When you're on
23:30 integrations, we should be able to find
23:32 API key, which you can come down here
23:34 and type in API key. Apollo API.
23:36 Fantastic. Come down and click on
23:37 connect. That'll open up a brand new
23:40 page for developer.apollo.io. Fantastic.
23:42 And on the left hand side, click on API
23:43 keys. Cool. So, I've deleted my old key
23:45 so I can show you a brand new fresh one.
23:46 So, click on create new key here. I'm
23:47 going to call this one something like
23:49 Hermes. I always like to name the thing
23:51 that it is. I'll say for use when using
23:53 Hermes, whatever the thing is. Come
23:55 down, select API keys. We're just going
23:57 to set this as a master key. come down
23:58 and click on create API key. Beautiful.
24:00 That's done. I'm going to copy this one.
24:02 Superb. And then what we can do is
24:03 literally come back over now and connect
24:05 that with Hermes. So if I bring up
24:06 Telegram, for example, I might come down
24:08 here and say, "Hey there, dude. I would
24:10 like to do some automations with
24:12 Apollo." So I've got the API key. Let me
24:14 know everything you need to do to add
24:16 this as a connector and a skill, please.
24:17 And so the easiest way, but this
24:19 suggests to do it, which is freaking
24:21 cool, is you can literally do a bash
24:23 command, which is this space your API
24:25 key. So, let's go ahead and run that one
24:27 together by opening anti-gravity. I'm
24:28 going to come down here and get a new
24:30 terminal by click on terminal. Click on
24:31 new terminal. And you can literally just
24:33 ask Hermes for step-by-step instructions
24:35 of how to add your API key in. And it's
24:36 good to add it to environmental
24:39 variables like this because it will be
24:41 stored in the chat logs which will be in
24:43 GitHub and also other places. And
24:45 obviously the model reads it. You can
24:46 throw it in, of course you can. It's up
24:48 to you if you want to. But if you just
24:50 want to be absolutely irclad, you can
24:51 add it in using environmental variables
24:53 if you like. Beautiful. So once you've
24:54 gone through that process of connecting
24:56 it together, we can now start to do some
24:58 really cool bits and pieces. But first
24:59 of all, why are we using Apollo? Why
25:01 does that make sense? How does this all
25:02 work together? Well, a couple things
25:04 that are good to know about this. Um why
25:05 it's worth taking a look at. They have
25:07 the biggest B2B database, which is cool.
25:08 They have emails that will get
25:10 delivered. Mobile numbers are pretty
25:12 decent. Uh intent signals and scoring,
25:14 so we can see whether or not they want
25:16 to buy. Of course, you can go and scrape
25:18 things using different services, but if
25:20 there's a reliability factor, there is a
25:22 size factor, there is actually hit
25:23 rates. What people don't appreciate
25:25 sometimes is that when you scrape a lot
25:27 of data that your hit rate can be quite
25:28 low, like you can only hit sometimes
25:30 like, you know, x% of them rather than
25:32 as many as you can. So, we want good
25:34 reliable databases and if you know you
25:35 got a great product, it's really cool.
25:37 We can do sequences and cadences all
25:38 within a platform, which is really cool.
25:40 So, you don't have to buy a second um
25:42 subscription, which is decent. uh and it
25:44 can play nice with your uh stack which
25:46 is good. So with this it's continually
25:48 refreshed. It's one API one quotota uh
25:50 compliance is baked into it. Intent and
25:52 scoring fire happen automatically and we
25:54 get the sequence and the dialer in the
25:55 same screen if you want to take this to
25:57 a new level on the platform. And then
25:58 scraping your own obviously has things
26:00 like GDPR risk, zero intent signal. You
26:02 may still need to get sequences and that
26:04 kind of stuff. But let's go ahead and
26:06 test this together with a good use case
26:07 cuz you've seen exactly how to connect
26:09 to Hermes. Now let's put this puppy to
26:10 the test. So let's try this one for
26:11 example. Hey there, dude. I've been
26:13 thinking recently that I'd quite like to
26:16 target roofers because roofers are high
26:17 margin and it's a bit of an unsexy
26:19 business. So, I'd like you to do for me
26:22 a bit of a prospecting brief on roofers
26:24 using the Apollo skill and then on top
26:25 of that, see if you can find for me, I
26:28 don't know, 20 different roofing
26:30 companies in Austin, Texas that I might
26:32 want to reach out for. Just give me like
26:34 the business name, a little bit of
26:35 detail about them, and then I can start
26:36 thinking about whether or not I want to
26:38 get contact info and what that looks
26:39 like. Thank you. I'm just going to send
26:40 that off. That could have been the
26:42 voice, that could have been anything.
26:43 And then we can come back and get these
26:45 beautiful uh just this beautiful
26:46 information to help us grow our business
26:47 basically. And we've got great pain
26:49 points. Roof is here. Past customers are
26:51 not being reactivated after storms.
26:52 Leads come inbound but they don't get
26:54 follow up fast enough. Really cool. And
26:56 we've got a strong offer. Freaking
26:58 awesome. Then we've got 20 Austin
26:59 roofing companies down here. We've got
27:01 Kid Roof. We've got Texas Fifth Wall.
27:02 All these different guys. I've not
27:03 included the email address down here,
27:05 but you can see we've got all these
27:07 different messages now using the Apollo
27:09 API. And look at this. My top average
27:10 picks. If I were prioritizing first, I'd
27:13 go with ABCD to EFG. And guys, we can
27:14 connect this now to Gmail to draft
27:16 emails and send emails to these people,
27:18 which is amazing. One of the best ways
27:19 to do this, by the way, which is just so
27:22 easy, is something called Zapium. It's
27:23 like a absolute cheat code for
27:24 connecting things. So, you're just going
27:25 to come over to this website on this
27:27 MCP, which is great. Now, I would
27:30 recommend this. Never give OpenClaw or
27:33 Hermes the ability to send emails, only
27:35 to draft. I'm still at the point where I
27:36 would never let them run riot. A couple
27:38 of my buddies have done that and they've
27:39 um been a little bit sorry if they did.
27:42 We're just not quite there yet with
27:44 writing but drafting is fine and you can
27:46 approve it which I think is awesome. So
27:47 you can pick the popular AI agent that
27:48 you want to. You can come down here to
27:50 see all of them. It's really freaking
27:51 cool. We're going to come down here to
27:52 new MCP server and what you're going to
27:54 come down clipping guys is open claw.
27:55 Don't worry about the fact it says open
27:56 claw. We're just going to substitute
27:58 that for something different and instead
28:00 you can copy the prompt here and then
28:02 come back over here. Paste this one in.
28:03 And then instead of open call, just come
28:05 down and change this one over here to
28:07 Hermes. And then come down here and just
28:08 make sure it's fine. And then you're
28:09 ready to rock and roll. And just send
28:10 that one off. And then when you click on
28:12 this on the left hand side, you can see
28:13 we can now add tools. So what we're
28:14 going to do is add a tool. I'm going to
28:15 come down here, find Gmail. And
28:17 crucially, I'm not going to give it the
28:18 ability to send emails. We want to
28:20 follow something called the principle of
28:22 least access, which is that any tool or
28:24 connection has the fewest possible tools
28:27 available to actually do the thing. So
28:29 find emails is fine. Get attachments is
28:31 fine, add labels is also fine. It's even
28:33 doing it right now as we speak, which is
28:34 cool. So, we're going to say, "Hey, that
28:35 sounds freaking cool." We're going to
28:37 allow that. That's absolutely fantastic.
28:38 We're already authorized. Thank you very
28:40 much. We appreciate your hard work.
28:42 Archive emails. Delete. We don't need it
28:44 to have delete, but I think create draft
28:46 and draft reply are fine. Create labels
28:49 is okay. Send emails is a no right now.
28:51 Remove labels is also okay, but again,
28:52 we're not going to do too much of that.
28:53 Going to come down and just click on
28:55 connect. Now, this only has access to
28:57 the stuff that I'm okay with. I say add
28:59 seven tools. The other one that we're
29:00 going to want to add is calendar access
29:02 as well. So that's cool. Let's come down
29:03 and give my calendar access. Also, I'm
29:04 going to give it the ability to find
29:07 events, retrieve events, find, you know,
29:08 busy periods in the calendar. That's
29:11 fine. Find calendars not required. Add
29:13 attendee to event. That's also okay. And
29:14 I'm going to give it the ability to
29:16 create a calendar event since that's not
29:18 necessarily too messy. And I'm kind of
29:20 okay with it doing that. And also, um,
29:22 deleting an event is also cool. So,
29:23 basically, to your own preference,
29:25 basically build this one out. This one's
29:26 on my computer at home. So unless
29:27 someone comes through the store right
29:29 here and they're not going to, they're
29:30 not going to be able to actually
29:31 physically get the stuff. And so just
29:32 give it the access that you're happy
29:33 with. And when you're happy, you come
29:35 down, click on connect, and Zapia is
29:36 pretty much like the Willy Wonkers
29:38 Emporium for connectors and stuff like
29:39 this. It's very cool. I'm going to come
29:40 down and select my account, which is
29:42 cool, which is this one. Then add my
29:43 eight tools. And when that's done, I
29:45 click off. And then basically you'll see
29:46 if I come down here on to connect and
29:47 then you'll see on apps that is all
29:49 physically connected together, which is
29:50 wonderful. And then just like this guys,
29:51 it's now confirming that everything's
29:53 connected. And that's as easy as it was.
29:54 I can say something like, "Hey there,
29:56 dude. Could you tell me, for example,
29:59 the title of the next calendar invite
30:00 that I have, the next appointment on my
30:02 calendar." Now, Hermes has got this.
30:03 We'll just see what has to say. This is
30:05 the easiest way to connect to Google. It
30:06 that I'm aware of. It is just super
30:08 duper easy. It's fantastic. And look at
30:10 this. We're even seeing a pre-flight
30:12 compression. So, these are tokens. This
30:14 is a threshold. That's going to compress
30:15 a little bit and give us a wonderful
30:16 answer. And whilst we wait for Hermes to
30:18 come back with the answer for that, I
30:19 just want to touch on some pricing stuff
30:20 on the Apollo side. If you do want to go
30:22 at that one, it is a paid subscription.
30:24 There are free alternatives out there
30:26 like you can use ampify. This one is
30:28 paid for a reason because of the quality
30:30 obviously. So all the way down from $0
30:33 up to $119 if your organization
30:34 obviously you can pick the one that you
30:37 like. You can use this API from free but
30:38 they basically there'll be certain
30:40 things you can't get like emails. So
30:41 just be aware of that when you're
30:42 actually going ahead and using it. All
30:44 right guys and just like that it tells
30:46 me my next calendar is weigh in. It does
30:48 not know about the beautiful Nando I had
30:50 tonight. But now we've connected all
30:51 this stuff together. It just brings us
30:53 on to a very interesting question and
30:55 that's how do we exactly solve the
30:57 second part of this puzzle. We get the
30:59 fact that we can use Hermes on our phone
31:01 which is fantastic but it's only half of
31:02 the knowledge system. And so the next
31:03 thing we got to do is learn how to build
31:05 this agentic system for Claude, which
