This is a full social media content machine.

Four AI agents that research, write, design, and publish posts for you.

Watch it work right now.

This content idea,

five signs your agent stack is overengineered,

is turning itself into a finished Instagram carousel.

Atlas found the angle

Vera is writing the slides

and in about 2 minutes,

Kite renders all of them with just a few clicks.

And here's the thing,

nothing goes online without your approval.

You check the words first

and you approve the final post.

The agents do the work

and you keep control.

Today we are building the entire thing

from the scratch.

Every prompt and template is linked below.

If you'd rather skip the 4 plus hours build,

there is also an auto installer that can set up everything for you in a matter of minutes.

Hey, Larry here and welcome to the channel.

Today we're building Cadence,

a content operations autopilot run by four specialists you just saw.

Packed with over 50 beautifully designed templates,

Atlas researches and posts your ideas.

Vera writes every slide, caption, and hashtag for you.

Then Kite does the finished designs.

Once this is done, Orion ships them to your social media platform.

And above them all is an orchestrator agent

that can take your orders in the Telegram chat

and route every job down the [music] chain.

If you post content regularly, you know the hassle.

You stare at a blank page hunting for [music] an idea.

When you finally get one,

now comes the research,

then writing slide after slide.

They're making it look good in a template

followed by the caption, the hashtag, the scul and all that.

Now, Cadence orchestrates and eats up the entire chain

in a matter of minutes.

You tell the agents once what your brand is about.

From then on, one click generates fresh post ideas for you

or you type your own rough thought

and they build the entire story for you from idea to publish post

all handled in a few clicks.

Then your post is online.

And here is where this build beats every other AI slide tool.

Normally, you would generate slides with an image model

and you would pay for every single slide.

And even fixing one word means you pay again to regenerate the whole image.

Cadence doesn't touch an [music] image model at all.

Every slide is built from a beautifully designed HTML template with real code, not with AI pixels.

So, every edit is free.

You can grab any elements, change the text, the size, the position, and it rerenders it in a matter of seconds.

Meaning you have unlimited revisions with zero cost per slide.

So your money goes into the words,

not into the pictures.

The dashboard shows the whole operation in one glance.

Every number pulled from your own agent's database

where every agent logs its name,

the task they performed,

the AI model they used,

and exactly when they performed that task.

So there are no fake metrics anywhere.

My favorite part is the studio.

You can edit every slice text or grab a single element,

move it, resize it, rewrite it,

plus even the phone preview, a template library, and some that carousel into a video rule.

We would also build a published archive where you can see all the posts that you've generated so far and published,

and also a calendar tab where every schedule content can be seen.

We would also have a configurations page which would hold your brand name, handle, voice,

and it flows into every word the agent writes.

Now, let me give you a live demo and walk through of what we are building today.

So on the dashboard you see the total number of runs that the agents have performed.

The number of ideas that we have generated in this setup.

The number of posts that we have published.

In my case I just have one because I used this to test for the tutorial

and the percentage of runs that have been successful.

Now here we have an agent activity that is pulled from the agents activity database.

So you can see currently on Friday I did a lot with the agents because I was building the setup

and that is why you see that this is basically going up

and on the right side you can also see which of the agents have currently the largest workload.

So in my case for instance these are Vera who is my content writer

and Kite who is a developer because those are the two that were doing most of the work so far.

Now when you come down here

you can also see a summary of what the agents have been doing

all pulled from the agent login database.

Now the magic happens in the ideas tab.

Here you can see a few ideas that the agents generated for me

because I'm doing stuff with AI.

So I just made them generate ideas around AI automation and stuff like that.

Now here you can put in your own idea whatever idea that you have

or when you don't even have an idea you can tell the agent hey generate some ideas for me.

So let me just click on generate some ideas

and let me see what ideas are going to be generated.

So here you can see that Atlas

who is our researcher

is the one that is going to look for ideas

around what we are building

and what we haven't posted before.

And this is going to trigger Atlas

who is our researcher

who is going to basically look for ideas

of what we can post

based on what the agents have learned so far

about what I'm building.

And then you can see that they generated five ideas.

So the first idea is automation fails at the handoff.

We also have some ideas like your agent should run the test it claims to fix

which is going to be a good one because the agents hallucinate a lot.

Now just to demonstrate an example let's say I want to write some slides on this idea.

All I have to do is to just click on promote

and then here I can put in a number of slides

that I want to put in.

So for instance I want to go for eight slides

10 or whatever number you want to go for.

So what you have to remember is that the first slide

is always going to be the hook

and then the last slide is going to be the CTA.

So the main content is going to be between the hook and the CTA.

So here if I go for 10 slides or let's say eight slides that it means that I'm going to have six content slides with one being the hook and

one being the CTA.

And then here I can decide

if the agent should do a deep research.

So for instance

if I click research with perplexity

then the agents are going to do a deep research through perplexity

to really find the right content for this job.

And then here I also built in a sort of a steroids mode

where the agents are going to write more punchy slides.

And then when you are done

you can just click on write eight slides.

Now what is going to happen is that basically Atlas is done with his job.

Atlas did the research right and now Vera is going to take over.

Vera is the AI agent that does the writing of the content.

So let's wait till Vera is done with the research and have the content ready.

All right.

So Vera is done with the research.

Remember we asked for nice slides and Vera is basically giving us an idea of what the content on each slide is going to be.

Basically you can see the content that is going to come on every slide.

If I do not like the slide and I want to refine the slide,

I can click on refine copy and then the agents would basically refine this based on what I tell them.

Now, assuming I'm fine with this and I want to promote this,

I can click on approve and pick a templates.

Remember that the contents that the agents did is basically from the idea of what the content should look like.

But here, I have a pile of beautifully designed templates that I'm going to give to you guys,

which you can just copy without you having to design everything yourself.

So, what we're going to have to do is just look for whatever templates we want to build with.

Now, because my brand is orange, let me go with this template here, 10 count.

Now what is going to happen is I'm going to click fit copy to this.

Now why do we do this step?

Because every template has the number of characters it can support.

So assuming the number of characters is way more than what a template can support

then the design basically looks ugly.

So what Vera is going to do is that Vera is going to look at the template that we have picked

and adapt the text that goes onto the slides to match that particular template

and then Vera adapted the slide text to the template that we picked.

So here I'm just going to click on render this

and then this is going to trigger another agent called kite who is basically for design.

So when I come here you can see that the slide is ready

and I can click to open this in the studio

or you can also come to the studio tab

and this will show you the slides.

So when I click open in studio you can see the slides that the agents generated.

So that you can see that I have eight slides basically tailored to exactly what we did.

Now the beauty about how these slides are generated is that typically when

people are generating Instagram slides you need to get an image generation agent or an an image generating model that is going to cost you for every slide that you generate.

And the most painful part is that

whenever you want to change any text,

then you need to pay for every API call.

The beauty about what we building here is that every slide is rendered as an HTML file.

That means that when you come here and click on fine-tune the slide and I pick on any slide,

I can basically edit and add my name Larry here.

I can do whatever I want to do without having to pay for anything.

What you can even also do that I also added is that you can also tune the slide to be left to up down.

You can reduce the slide.

You can change the font size and everything.

And then when you're done, you just click on save and this will be done.

Now assuming you're done and you're basically happy with all the contents, then you can promote this.

Now you can also convert this into a real where you can click on animate this carousel.

You can animate all the slides that are going to come one after the other like a real with a 4 is to 5 aspect ratio

or you can also create a full reel with a 9 is to 16 aspect ratio.

Now assuming I'm done and I'm happy with what I've built so far,

I can just come here click post now and then in my case I have already configured Instagram.

So I can click on post to only.

And then the agents are going to warn me, hey, are you sure you want to post this image to is

I'm going to say yes, post now.

And then the image is going to be pushed through a platform that we're going to be using, which is called Buffer

that connects to all the social platforms.

And then this will be pushed directly to Instagram for us.

And now I get a notification from Orion, which is our publisher,

that the post has been pushed to buffer.

So in a few minutes, we should see this directly on Instagram.

And right then you see it.

So this is a test account that I created on basically to test what I'm building.

And when I come here, you see that the post that we just created is live on Instagram

exactly as we rendered them.

And this exactly what I'm going to be building with you guys today.

We also do have a dedicated agents tab that gives you more insight into the agent architecture that we have.

You can see that we have every sub agents have their dedicated card

where you can see the AI model that the agents are using.

You can see the number of runs that particular agent has performed,

how many of these runs have been successful

and also some of the recent tasks that the agents performed.

On top of all these four agents,

we have the orchestrator who is more or less the boss agent.

Now below we have an activity heat map where you can see in a time frame of 24 hours which agent has been active.

So the more brighter the color here, it just means that the agent was more active within that hour window.

And darker spaces just mean that the agent was not active within that time frame.

So the entire build is going to take you about four to 5 hours if you have all the prompts and the templates.

So I also created an auto installer that can do everything for you in a matter of minutes.

If you'll be interested in getting the auto installer,

this model will also be in the description section.

Now, let's get into the build.

To get started, we're going to first install her.

He agents allows you to be able to build multiple agent infrastructure

where you can be talking to your AI agents

using all the popular applications out there

like Telegram, Discord, Slack, WhatsApp,

and all the popular messaging applications

that are out there.

Additionally, instead of being confined to using one particular AI model,

you can use any AI model that you want.

been it from OpenAI, Ben from Antropic or any other popular model that is out there.

And to install Hemisen, there are several options that are out there for you.

You can install this on a Windows PC if you have a dedicated PC that you want to run this on,

or you can also install this via a Linux operating system.

For this setup, I'm going to be using the Linux approach.

I know Linux sounds intimidating.

Don't worry, I'm going to walk you through the entire step.

And then once we are done with the setup,

which is going to take a few minutes,

everything that you're going to be doing

is going to be from the dashboard

that we're going to be building

or also from the Telegram messaging application.

So you wouldn't have to come into the terminal

to do any stuff with quotes and stuff like that.

But I'm going to walk you through the entire process

from A to Z

so that you get a breast with how this whole flow would work.

Now, if you have a PC that you can run this on,

hey, you can use this as well.

But in my case, for instance,

I do not have a dedicated PC.

So you can rent a server somewhere

that you used to run this on.

I personally prefer this option,

but hey, it's up to you.

With using a private server is that

you rent a server from a different company

where you pay about $4 a month

and then this server is going to be on 24/7.

You do not have to worry about maintenance.

You do not have to worry about internet connectivity

and stuff like that.

But of course, this comes for a price

and that is why you have to decide

if you want to run this on your personal PC

that you have at home

or if you want to rent a server to run this on.

And for the VPS provider,

my personal picks are typically Racknet and Conabo.

Now, with Rocknet, for instance,

you can get a server with two CPU cores,

35 GB of this space,

2 GB of RAM

for about $35 a year,

which amounts to about $2.9 a month,

which is quite affordable for a server,

considering that you do not have to worry about

keeping it online 24/7,

internet activity,

maintenance,

and stuff like that.

With Rak, you pay once a year

and you're basically done.

You don't need to worry about monthly payments

and stuff like that.

The second option that you can also go for

is from Contabo.

They also do have quite reliable servers

and the price are also quite affordable

for the specifications of the VPS that you get.

So with Contabo for instance,

you can get a server that has four CPU cores,

8 GB of RAM

and about 100 GB of SSD disc space

for about $8 a month.

This one here is going to be way more sufficient

for what we're going to be building.

So you can decide on either going with Raknet

or going with Contabo.

So I'm going to put a referral link in the description section that you can get one of these.

And these videos are not sponsored by any of these companies.

I'm just giving you what I typically use when I'm setting up these infrastructures.

If you want to go with Raknet, just come here.

Like I said, my recommended option would be this one here.

Click on order now.

And then you're going to be brought to a page

where you would see the summary of the hardware specifications

of your VPS that you selected.

And then when you scroll down as well,

you see the server label,

which you can change later if you want.

Now, in terms of the configuration options,

I would recommend that you keep this as a default.

Operating system,

we're going to be using Ubuntu 2012.04.

For the RAM, leave everything as a default.

immedately you increase the RAM it's going to in an additional cost

and that is why I would recommend that you go with the default options

that is also here in terms of the location you can

choose whatever location that is closest to you

but what we're going to be building latency is not much of an issue

so I wouldn't worry too much about having it as close to you cost

leave it as a default if you take something

higher you would basically pay more

after that click on continue

you're going to get an email that would contain all the credentials

you're going to need to connect to the VPS ser

and I'm going to walk you through that process as well

if you want to go with a contabo option like I said

this option that we have here cloudvp4

or the lowest option that they have is going to be way more powerful

for what we're going to be building.

So here click on get started here.

Be careful the default is sometimes 24 years.

That means that you're going to be locked with this VPS for 2 years.

And of course it gets cheaper

the longer the contract is going to be.

But just be careful

because if you go for 2 years

then you are locked to pay this amount of money

for 2 years.

If you want to have the flexibility of cancelelling this every month

then you can go for the one month option

where you're going to be paying about 5.5 in my case a month.

Now in terms of the region you can leave this as European Union which is the default because Contabo is in European Union

with Contabo immediately you choose a different location then it would incur an additional cost

but like I said earlier with what we're going to be building latency is not going to be that much of an issue.

So you can leave this completely as European Union.

When you scroll down for the storage you can choose SSD 100 GB which is going to be free

and way more sufficient for what we want to build.

For the operating system leave this as YUbuntu which is the default.

For data protection, I typically recommend that you go for the auto backup option.

Sometimes I'm working with my agents and something completely gets screwed up.

And then with the auto backup, it basically takes a snapshot of your VPS every day.

So that means that if something goes wrong, you can always go back to about 10 days back

and then you can always restore whatever image you had on that VPS for that particular day.

Now this optional but like I said it's highly recommended

because if you imagine if you've been building your content management system for over one year

and by mistake something screws up you completely lose this.

Now to save this as well you can also configure your own auto backup

which is going to take a snapshot of your VPS every day.

This for instance you can build with cloth

but if you don't want to do this and you want to go with contabo's default auto backup option

you can choose this which would about 1.65 a month.

Now additional features you can leave everything as a default.

In this section here, you're going to have to put in a password

where you're going to connect to the VPS server.

Here you can generate a new password or you can put in any password that you want.

This is going to be a password that you're going to have to remember

because this is a password that you're going to use to connect to the VPS server.

If you forget this password, you can go into your conso control panel and reset the password.

But why go through the hassle after that?

Just click on next, make the payment, and I'm going to walk you through the next steps.

If you went with a racknet option after making the payments,

you're going to get an email like this from Racknet.

With this email, we basically going to need three things.

The first thing that you're going to need is the IP address of your VPS server.

Every VPS server has a unique IP address and they're going to provide this IP address to you.

The second thing that we're going to need is the username of the account that you use to connect to the VPS server.

With Raknet, most of the time name is always going to be root.

And then thirdly, you're going to need a password which they're also going to provide to you here called root password.

And here's what they're going to provide the password to you.

All these three credentials are going to be in your email.

With the SSH port, you do not have to worry about this

because with most of the applications that you use to connect to the VPS server,

they always default to an SSH port of three.

Now, with the IP address, with username, and with a password,

you would be able to connect to the VPS server,

which I would walk you through in the next step.

If you went for the Contabo option, you're going to get an email like this from Contabo.

Mine is in German, but the layout is going to be the same

irrespective of which country or which language that you have.

In the email from Contabo, we're going to need two things.

We're going to need on this left side here the IP address

which is the unique IP address of your VPS server

and we're also going to need the username with Contabo as well.

The username is in most of the cases root

the password they wouldn't provide the password to you in the email

but this is going to be the password that you put in when you were purchasing the VPS server.

So with the IP address with the username and with the password

that you put in when you were purchasing to the VPS server

we would be able to connect to the VPS server.

And to connect to the VPS server there are several applications that are out there.

The good news is that whether you're using a Windows operating system

or you're using a Mac operating system,

both do come with inbuilt applications that you can use to connect to the VPS server.

Using a Windows operating system,

just click on the search bar

and look for an application that is called Windows PowerShell.

And then when you click on this,

this is how the Windows PowerShell application looks.

If you're using a macrop system,

the application that you're going to be using is called terminal.

The commands that we're going to be using here in PowerShell

are identical to what you would be using in PowerShell.

So you do not have to worry about differences in terms of commands.

To connect to the VPS server,

just put in the command ssh

space and then put in root.

Root is going to be the username of the VPS server,

which is the same for Contabo

and for racknet.

And after that,

you're going to put in the IP address of your VPS server.

The trick here with PowerShell is that

whenever you have something in the clipboard

and you right click on your mouse,

it basically paste whatever you have in the clipboard,

which can be quite useful,

especially when you are typing in passwords.

And after that, just press enter.

If this is the first time that you're connecting to the server,

you're going to be asked whether you want to add this fingerprint of the server to your PC.

If you're sure you're connecting to the right server,

just type in yes and it should be fine.

And after that,

you're going to be prompted to put in the password of your VPS server.

And here,

if you have a complex password,

just type this password somewhere.

Come in here,

just right click on your mouse.

It would paste the password into the clipboard

and just press enter.

And then you can see that in my case,

I'm connected to my VPS server.

So,

I'm going to clear the screen by just typing clear.

And then we can start with the configuration process.

So now we connected the VPS server using the router name

and root as a user is typically the highest authority on a Linux server.

Now to install Hermes

my tip for you would be not to install Hermes on the root user

but create a new profile or a new user on which we're going to run this on.

Now this is quite helpful

because then you do not have Hermes being the highest authority on your VPS server.

In my earlier videos, we always did this on root, which is absolutely fine.

But I always try to improve and bring you guys the best safety tips out there.

And that is why we're not going to be installing Hemis this time around on root.

We're going to be creating a dedicated user on which we're going to install the whole Hemis application on.

Now, to create a new user on Linux, just use the command user space.

And after that, just put in the name of the user that you want to create.

In my case, I'm going to call this Larry.

So, add Larry.

And after that, you're going to be prompted to put in a password for this user.

So, I'm going to put in my password.

And after that, just press enter.

You're going to be prompted to retype the password.

And then after that, just press enter.

And then we're going to be asked if you want to put in a full name.

You can just skip this by pressing enter.

Room number.

Enter.

Phone.

Enter.

Just skip this.

Home phone.

You can also skip this.

You can also skip this.

And then you're going to be prompted

if the information that you have put in is correct.

Just press Y for yes.

And after that, just press enter.

Now, you can see that then Larry is added as a user.

Now, we're going to give this Larry user some pseudo privileges

so that the user can be able to run some commands

and install the Hemis application.

And to do that, we're going to use this command here.

User mod space minus ag p sudo larry.

So here, just replace this Larry with your username.

And after that, just press enter.

By the way, I'm also going to provide you these commands.

So you do not need to type this or you do not need to remember them.

After that, just press enter.

And then we're going to get Larry

now being a user with pseudo privileges.

I'm going to clear the screen just to make it a bit cleaner.

And now we're going to switch from the root user into the Larry user

for us to be able to login into the Larry user that we just created.

And for that just use the command su space minus or hyphen space and then just type in Larry.

And you can see that the syntax change from root to Larry.

So that means that now I'm not on the root application which has the highest authority.

I'm now logged in into the Larry path which has this authority.

Now when we're connecting to the VPS server, we use the command ss space root the IP address.

But here as well

if you want to connect to the VPS server

instead of root you can use the new user that we created

which is in my case Larry

and then you can just put in the IP address of the VPS server

then press enter

put in the password that you

created and then you see that I'm also connected to the VPS server

so you can connect to the VPS server either using root

and after that switching to the larger using su space minus Larry

or you can also directly go into that using the

SSH path for Larry.

So I'm just going to be sticking to with this because I already created a set with this here.

I'm just going to type in clear to clear the screen.

And then now we can start with the Hemis installation.

To install Hemis, we're just going to come to the original website from Hemsus.

And then when you're on this page, scroll down here and then come to the install via terminal.

We're going to be going with the native application.

And here the command always starts with curl

so that you are sure that you are on the right one.

Now click here to copy the oneline installer.

The advantage here is that with this oneline installer,

it's going to install all the dependent that required to install or to run hemis

without you having to install application A, application B and stuff like that.

After that, just press enter and then the installation is going to run.

So, it detected that I'm running Linux

and it's also going to run all the other things that are missing on my server.

Now, here I'm being asked,

do you want to install RIP for faster file search for TTS voice message?

This would be good.

So, we can just type in yes that we want to install this and press enter.

and then I'm going to put in a password and then press enter and then the installation is going to proceed.

So I'm going to wait for the installation to be done and then we can continue with the build.

All right, so the prerequisite installation is done.

So now we can start with the setup.

Now for here we're going to have to choose how we want to set up Hemis.

I would recommend that we go with a quick setup

because a full setup is going to ask us so many things

that we're going to have to configure which we do not need.

And this is also the recommended part.

So just make sure that this option is selected.

If not, just press up and down on your keyboard.

Make sure that the pointer is on this

and then press the space bar or enter key.

Now in the next stage, we're going to have to create an account with news research.

If you do have an account, then you do not have to create a new one.

And basically, this is a step where we're going to bind this VPS

or this server or this Hemis installation to our news research account.

Now, here you can go with their free subscription.

Nose research does have some paid plans

where you can use a pool of AI models that they have,

but in my case, I'm going to be coming with my own AI model.

So, I'm going to go with a free option.

So just press control on the keyboard and click on this URL here.

And if you do not have an account, you're going to get an option to create an account.

But in my case, because I already have an account and I'm logged into my account,

it directly brings me to a screen where I can bind my account to the setup.

So here, just click on connect.

And when this is successfully, you're going to get a notification that this has been connected.

And then when you come back to the Hemis installer screen,

you see that it brings you to the model selection process.

Like I said, in my case, I'm not going to be using any of the models that no research provides.

I'm going to bring in my own model.

So here you can just press enter on the keyboard to skip the model selection process.

In the next stage we're going to pick a few tools that you can use with her agent.

For instance, you can give your agents the options to be able to search on the internet

and script data from the internet on particular websites.

And here for instance you can use a tool that is called fire crawl and the same applies.

There are other tools that you can use for image generation

for text to speech for speech to text and so on.

So here you can just press enter to use all of these

or you can also uncheck them if you do not need them.

In the next stage, we're going to select the terminal back end.

Here you can keep this as the default, which is local by just pressing on the space bar.

In the next stage, we're going to connect a messaging platform with which we're going to use to talk to all our AI agents.

And here, it's recommended that you set this up

because we don't want to be talking to the AI agents

every time through the terminal here on the VPS.

We want to have a simple, more friendly user interface

that we can talk to the AI agents.

So, here we're going to select set up messaging,

which is recommended anyway.

So press space bar on the keyboard

and then you're going to list all the messaging platforms

that are currently configured with Hemis.

And as you can see there are a pool of them that you can use.

But in this setup we're going to be going with Telegram.

So just press down on your keyboard and then here don't press enter.

If you press enter it would skip this part.

Press spacear on your keyboard to make sure that is checked.

And after that you can just press enter.

And then here we're going to create a telegram bot.

And this is going to be the bot

with which we will talk to the agents

that we're going to create.

For the Telegram B setup,

you can either go with the automatic setup

or you can also go with a manual.

In my case, I'm going to go with the middle option

just to show you how you can do this

if later you would need this for another application.

So for manual, just press two and then press enter.

And then HS is going to ask us what is the Telegram bot token.

So to do that, just come to Telegram and look for a bot called BotFather.

So you can see that the right botfather bot has over 7 million monthly users.

So that you are pretty sure that you are using the right bot.

Be careful here.

As you can see there a lot of botfather here.

But the one that you need is this one here with a check mark here and also with over 7 million monthly users.

To create a new ground, use the command forward slash and then type in new bot.

And then you can press enter or you can click on the suggestion that we have here.

Now here we're going to have to give our bot a name.

I'm going to call this km in my case for computer mechanic content creation OS.

So when you're done with the name just press enter and then you're going to be asked to put in the user of the bot.

Now very important here the username has to end with the name but

and that is why I copied this because in my case I just want to call it the same name but just add the word but to it

and after that just press enter

and then the bot is going to be created for you.

Now when the bot is created the botfather is going to give you a unique key

that means that you are the creator of that bot or you are basically the owner of the bot.

So I'm going to click here to copy the key

because that is what the hemis installation is asking

and I'm going to paste in the key here

and after that just press enter.

Now this step is a highly recommended step.

Hermes is asking you,

do you want to put some users

who are the only people that are allowed to talk to the bot?

Now this is very very important

because if someone has the username of your bot,

you don't want them to be able to talk to your bot.

And that is why we're going to tell her,

hey, let me be the only one that can talk to this bot.

And the only way Hermes can know that it's you

is by giving Hermes what your user ID for Telegram is.

And to get your Telegram user ID,

we're also going to use a bot

that is called user info bot

which also has over 400,000 users.

There are a lot of other bots out there.

So make sure that you have the right one here.

And to get your user ID,

you can just send the bot any message like hi.

And the bot is going to respond with your user ID.

So I'm going to click here to copy my user ID.

I'm going to come back to the terminal in Hermes installation.

I'm going to paste in the user ID and I'm going to press enter.

And then Hermes is going to ask you if you want to set this user ID as the home channel.

That means that in the messaging chat that you are talking to the bots,

whenever the Hemis gateway is restarting or any system things are going on,

you're just going to get a notification in this chat window.

You can type in a Y for yes and then press enter.

And then you're going to be asked if you want to install the Hemis gateway as a system service.

This very important because whenever your VPS server restarted due to a system update or anything,

the Hemis gateway is going to start automatic which is quite useful so that you do not have to start this manually.

Press Y for yes and then press enter.

And then here you're going to be asked how do you want to run the gateway?

In our case, we're going to go with the system service.

So press enter.

And then I'm going to be prompted again, do you want to run and install the gateway as a background system service?

I'm going to press Y for yes

and I'm going to press enter.

And then the Hemis installation is completed.

Now very important here,

we're going to have to reload our shell

so that we can just type in Hemus

and this will launch the Hemis application.

And to do that,

just copy the command that they have provided for you here.

Paste this command back in the terminal

and just press enter.

It's basically going to restart the shell for you.

Now remember,

we are basically done with the Hemis installation.

But in my case,

I didn't select any AI model

because I want to bring in my own AI model.

So here we're going to configure our own AI model.

And I'm also doing this in this way

so that you guys can learn

how you can add your own AI models later.

Now, if in future you want to add some AI models,

you can basically ask your AI agents to do this for you.

But I typically do not recommend this

because in most of the cases the AI model is not very smart,

they basically mess up with your configuration.

And that is why for stuffs like that

I prefer to do this on my own.

Now for which AI model to use

this is absolutely up to you.

But what I typically use most of the time

which a lot of people do not also know

is that if you do have a charge GPT subscription

if you the plus account or the pro account

you would be able to use an AI model from open AAI

that is called Codex

on her agent basically for free.

If you have the $20 plan you can use

you can use this.

If you have the pro plan you can also use this.

I'm I'm not very sure if this is available for the Go plan.

In my case, most of the time I'm using the pro plan.

And the usage limits that you have for the pro plan is more than you would get for the $20 plan, obviously.

So, if you're not sure which one you need, you can go with the plus account.

If you hit your limit, you can basically always upgrade to the pro plan.

Now, the is also a bit generous in terms of the token limits.

So, you would be able to do a lot even if you are using the $20 subscription.

The advantage here is that it's not basically a pay as you go.

Whenever your limits are done, you are done.

You do not get any surprise bill from OPDI

that hey, you've used more than your subscription allowed,

so you have to pay extra.

That does not exist.

And that is why I personally go with these options

because it's you don't have to worry about having a surprise bill

or anything like that.

Personally,

I'm on the pro plan

because I have a lot of Hem setups

and that is why my token usage is also very high.

And also for the tutorials that I create for you guys,

I typically try this over 10 times

and that is why I typically need a higher limit in terms of the tokens that they make available to you.

when I had one Hemis account

or when I had one open

CL account,

this one was way more sufficient for me.

Now, what people also not aware of is that if you also do have an X account,

you can also use the Crog model on Hermes agents for free.

So whether you have a chat GPT paid account or you have a premium or paid subscription,

you can use in both cases their AI models on her agent for free.

If you do not want any of these and probably you want a particular model like from Antropic,

oo, sonet or whatever, we can also come to open router.

Open router has quite a pool of a models like all the popular AI models that are out there.

You can get them on Open Router.

The advantage with Open Router is that they're going to give you one API key

and with that API key, you can get access to all the models that are out there.

The tricky thing that you have to take note of the fact that with Open Router

is basically pay as you go.

It's not a case where you paid a month and then when you hit your limit, you hit your limits.

You basically pay as you go.

The more you use, the more you pay.

So, I mean with whatever model that you want to use is completely up to you.

just go for the option that makes the most sense to you.

Now, in my case, I'm going to be going with my OpenAI because I have a CHBT subscription.

And to use that, I'm going to clear the screen here to make it a bit cleaner.

Just put in a command Heis space model and then press enter.

And [clears throat] this is going to load all the models that are currently available on Hemis agent.

If you do have an paid premium X account, then your option would be here, the X AI Grock.

The process is going to be the same irrespective of which option that you go for.

So, in my case, I'm going to select this by pressing the space bar.

And then here we're going to have the option to choose between open open AI API.

If you go with API then you pay as you go.

So the more you use the more you pay and then you would have to create a dedicated API key to collect this.

But in our setup we're going to be using our charging subscription.

So I'm going to go with the open AAI codeex.

And to do that we're going to press control and we're going to click on this URL that is provided here.

And in the process we're going to have to enter this code that is over here.

So just highlight this code.

Very important.

Press control shift C.

Do not press Ctrl C.

Ctrl C in Linux cancel an application

and this would quit this whole screen

and that is why you press control shift C

and then the link is going to bring you to this page.

All you're going to have to do is to just log into your OpenAI account.

So I'm going to click here and then click on continue

and then just put in the code that we copied from the PowerShell

from the Hermise installation and click continue.

And then our CEX account or our OpenAI account is now bound to our Hemis setup

and then when you come here you see that the authentication is done

and then here we are being asked to select an AI model.

And then here you can use any of the models that you have here.

Remember the higher the number like 5.6 takes way more tokens than 5.4

4 than 5.5, which also takes higher tokens than 5.4.

In my case, I go with 5.5 most of the time

because it's still a very super powerful model and I always want more tokens.

But hey, if you are on the $100 plan and you have a lot of tokens,

you can also go with a 5.6.

That's absolutely up to you.

I'm going to go with 5.5 here.

I'm going to press spacebar.

And then you can see that it says that my default model has been set to GPT 5.5.

Now before we go into the telegram to test to see if everything works from the Hemis installation and

from the model configuration just type in Hemus and

this is going to open what is called a Hermes terminal user interface.

Now it's going to be a chat box

where you can basically talk to your default AI agents

to make sure that everything works.

But a lot of people like me do not use this part

because then you always have to log into your VPS

come to this weird page to talk to the AI agent.

We just want to make sure that the setup works

and that is why I'm coming here.

So here I'm just going to send in a message like hi

and if the agent responds then it means that the hemis installation works.

It also means that the AI model configuration worked

and here you can see that the agent response hi what can I help you with?

So that means that our installation worked

the AI model configuration works.

Now to leave this screen you can press Ctrl C on your keyboard

or you can also just type forward slash and type in exit and press enter

and this is going to kick you out of the Hemis terminal user interface.

A quick word from today's sponsor, Sudalis AI,

and it's worth paying attention to this one.

Problem they are solving is that enterprises are currently drowning in disconnected AI.

Most companies are using AI tools that do not even talk to each other.

One tool does this, another does that, but nothing connects.

So nothing really works together.

Sudalis AI built a system that connects everything into one stack.

Check them out at sudalis.ai with all the links in the description section.

Thanks Sudalis AI for sponsoring this video.

Now we're going to come back to botfather.

When bot created them for us, he also gives us this URL with which we can start talking to the bots.

So here just click on this URL here.

It's going to open a new chat session with the bot that was created.

You can just click start.

Now this is going to send a message like start to the bot.

But the bot doesn't respond but it doesn't understand what start means.

So here we're just going to send a message to the bot like hi.

And if everything works the bot is going to start typing it.

So if you see this typing symbol then it means that the bot is configured and the bot is going to respond to you.

So the bot responded hi what can I help you with today.

Now we're not going to be talking to the bot in this session here.

What we want to create is we're going to create a telegram group for our bots.

The reason why we're going to create a group is that we're going to have basically three topics.

The first topic is going to be where we're going to be talking to the bots

which is going to be our orchestrator.

And then we're going to have a second topic or a second channel

which we're going to use to talk to our developer agent

who is going to build the whole dashboard for us.

And then we're going to have a third channel

which we're going to call studio.

And that is where we're going to brainstorm ideas and stuff like that

with our agents.

When we are creating the group,

a prerequisite to create a group is that you have to get at least two people in the group.

So we're going to need the username of our bots

so that we can add this bot to the group that we're going to create.

I'm going to click on the name of the bot here.

And then here we can click to copy the username of the bot.

And after that we can close this dialogue box

and then we can create this new group.

Now to create a new group,

click on the hamburger menu that you have here on the left section here

and then click on new group.

And then we're going to name the group.

In my case, I'm going to call this KM Content OS

and then click on next.

And then we're going to have to add at least one person to the group.

And that is why we copy the username of our bots.

So I paste in the username

and I'm going to click to select the bots

and I'm going to click on create.

And then now we have our group created.

But like I said, we want to create different channels in this group.

Now, we're going to have to make the bots that we added an administrator

so that the bots can talk in this group.

Click on the three dots that we have here.

Come to group info.

Click on manage.

Come down here and click on administrators.

Add administrator.

We're going to select the board that we created because the board is already in that group.

And just review the permissions that you want to give to the board and just click on save.

And then now you see that our bot has an admin privileges.

After that, click on close.

Now, whilst we are here, we also going to enable this aspect that I call channels.

So by default, Telegram doesn't enable these channels or what is called in Telegram topics for you.

So to enable this, just click on the topics section here, which is in my case by default off.

Click on this and then you can enable this.

And then here you can decide the view that you want to go for.

In this case, I'm going to go with the tabs option here.

And then just click on save.

And I'm going to click on save again.

And then we can now close this.

And then immediately we do that when we enable topics.

You see that now we have different channels.

We have all channel where we going to see all the messages from all the channels.

And also we have the general channel.

So first let's rename this general channel.

Click on the three dots here.

Come to edit.

And then we can call this channel orchestrator

because this is the channel where we're going to talk to our main bot.

I'm going to click on save.

And then we're going to start talking to orchestrator here.

So let me just send a message in the chat here.

When I send hi and our bot is typing that means that orchestrator is in this chat.

So we can basically start talking to orchestrator.

We are going to create other two channels.

But we are not there yet.

Let's just build the foundation and then when we get there,

we're going to go through that process as well.

All the prompts that I'm going to be using in this build

can be made available if you interested.

You can get this for a token of support on my website

or if you are also a member of my community,

you can also get this.

If you guys want to be a member of my community,

it's $10 a month

and you can get basically access to all the prompts

and all the templates

and all the tutorials prompts that I have on my channel.

And this goes to also support the work that I do.

A lot of effort goes into creating all these tutorials,

creating all these forms that I give to you.

All the AI models that I also use in building these are also quite expensive

and that is why I would need to get this with a token of support

to help me to be able to keep bringing these tutorials to you as well.

So if you want to join my membership group,

I'll be very happy to have you there.

We also do have a telegram channel

where you can basically get access to me or all the people in the community

ask any questions that you want

and everyone there would be happy to support you with the questions that you have.

And I also do create auto installers.

For instance, for the setup that I'm creating,

I'm building an auto installer that would build the entire thing for you in less than 5 minutes.

And this is going to be available to people in my community or my membership as well.

Said, hey, if you want to join,

I'm very happy to have you on board.

So, what I did now is that I moved Telegram to the left

and it basically had all the prompts on the right

so that we can basically copy this

and it will be very easy.

Now, whenever you set up Hemis for the first time,

the bot has no idea who you are.

The bot doesn't have a name.

the bot has no idea how the bots can help you with with whatever you want to do.

So the first thing that I always recommend to users is hey let the bot understand who you are.

Give the bot a name.

Tell the bot your name

and give the B a gist of how he or she can help you for this.

My first point that I'm going to give to my bot is this.

Your name is going to be orchestrator.

You are going to be my top level coordinator for my multi- aent setup

and I am the owner and I have the highest authority.

And then here you're going to fill in your name to the agents.

You're going to also give the agents your brand name.

You're going to give the agents your social media handle

because remember we are creating a social media content creation machine.

You're going to tell the agents what your brand teaches.

Now, this is very important

because we're going to have this dashboard that is going to create ideas for us.

And for the agents to be able to create these content creation ideas

or these post ideas for you,

they need to understand what you're building.

And that is why this part here is non-negotiable.

You can also tell the agent what your audience is and stuff like that.

So, I'm going to copy this.

I'm going to paste this in here.

And I'm going to fill in the sections for my build.

So here I bought,

hey,

my name is Larry.

My brand is computer mechanic.

My social handle is computer mechanic.

And here I teach it technical builders to build AI agents and automation,

who my audience is,

my voice,

and my working time zone.

Now the working time zone,

for instance,

is important because when the agent is creating like Chrome jobs,

then the agents will basically know when to reach you and stuff like that.

Now I'm going to send this to orchestrator to get him breast with who I am and also what we're going to be building today.

And then orchestrator responded,

"Hey, got it.

I'm orchestrator

and you are Larry.

Your brand is computer mechanic and

your content is plain direct content teaching builders to build agents and

automation." Now,

the next two prompts here are what I would recommend that you do.

The second prompt here is basically where you're telling orchestrator to interview you.

Now, what you give to orchestrator here is basically the backbone, the foundation.

But there are so much that you can let your agents learn about you.

So just take time and give this prompt to Kasha

is going to ask you some questions to understand who you are

of the setup that you want to build your audience and all those things.

But I'm going to skip this in the tutorial

because otherwise the tutorial will get lengthy

because for my testing or would ask you about 10 15 questions.

So I would skip this just not to get the tutorial too lengthy.

The second thing that I would also suggest to you is this permanent operating rules.

Now when you are working with the agents

it's always advisable to let the agents always give you a plan before they do anything they want to do.

And also when the agents are working,

you always want the agents to keep you updated in the process of what they are building.

Otherwise, sometimes you just sit in the dark and you do not know if the task is completed or the task is is done.

And that is why I prepare always these permanent operating rules which you can also give to your agent.

It's optional for you

if you want to have this

this there ready for you.

I'm going to skip this part as well

otherwise in my build orchestrator is always going to ask me

give me a plan that I have to type a proof before he does it

and this would slow down the build.

Now we're going to start with the main hat of the build.

We're going to tell orchestrator that

hey we're going to create different agents.

We're going to create an agent like Atlas

who is going to take care of content

creates carousel ideas.

We're going to have Vera

who is going to be our writer.

We're going to have Kite

who is going to be our designer

and also Kite is going to be our developer.

And we're also going to have origin

who is going to be our publisher

publishing things to whatever channel

that we want to use.

And here we are basically giving orchestrator a prep or a gist of what we want to build.

Right?

And we are telling orchestrator that hey you are already one of these five agents that we're going to create.

You are going to be the main agent and you are the one I'm talking to right now.

And also important I'm telling him when he is creating these profiles he shouldn't create one for himself

because I did test with some AI models that created a profile for themsel as well which doesn't make sense

and that is why I also tell it here that

hey

you do not need to create one for yourself.

So I'm going to copy this prompt

and I'm going to give this to orchestrator

to prep him what we're going to be building.

I'm also telling him,

hey,

confirm that you understand the plan,

the five rows and the profile name mapping

before he even goes to change or build anything.

And what I like about Hermes is that

whenever you send Hermes something

that if he finds useful long-term,

he commits this to the memory.

And that is why you always see this message like memory.

It means that something has been written to the memory.

And by the way,

cadence is the name that I give to the content creating OS

that we're going to be working on today.

And then he's just saying hey

the five cadence rules are orchestrator

which is me and then the other profiles that we're going to be creating.

And he's saying that you are Larry

the owner and the final authority.

Now in the next prompt this is going to be the most important prompt.

This is going to be the prompt that is going to create the AI agents that we're going to need.

I'm going to send the prompt to orchestrator.

So I'm telling orchestrator that

hey create five persistent hemis profiles.

Now the reason why I use persistent is that hemis can also

create agents that are basically called in to do a task and

then after that they basically dead.

So what I want to create is that I be created permanently.

So that means that whenever we need them,

they are going to be there.

Now here whenever you are creating an agent

always with Hermes there is always one default agent.

There's always one agent who is going to be at the top

and that is typically the default agent.

And below the default agent you can have whatever number of agents that you want to have

but there's always going to be one agent at the top.

And that is why in my setup I typically make the default agent who is at the top orchestrator

and then he builds the other agents that would work under him.

So someone for instance asked if you can have two orchestrators.

In principle you can have two orchestrators that

but it means that one orchestrator would have to be under the default orchestrator

which you can also do.

This is basically up to you how you want to configure this.

Now in creating a profile for an agent you basically need to tell the agent

hey what am I as an agent?

So for all the agents

we are telling orchestrator to add this information

into a document

that would be called their soul.md.

So a soul.md is a document that tells an agent hey my name is Larry.

This is what I do.

This is the person that I'm working with.

This is the system that we are working with and stuff like that.

And that is why in this prompt for all the agents I tell orchestrator what he is supposed to put in their description.

So for instance for scout I'm saying hey your name is scout you are a content scout for candid studio.

Your job is to find fresh scroll stopping Instagram Tik Tok carousel ideas and to research facts behind them.

So for all the agents I'm telling them what they are going to be doing.

And for instance Kite Kite is also going to be our designer.

He's also going to be our engineer.

So he has what I'm describing for Skype for him to understand what he does.

Now whenever you start a new chat session with an agent, they always load in their soul.md.

And that is why it's very advisable that we describe in detail what the agents are going to do.

Now he can change the text to whatever you want, whatever fits what you are doing.

For instance, in my case, I said Instagram and Tik Tok, but maybe you want to have both, LinkedIn and stuff like that.

You can basically add this here as well.

So, orchestrator responded.

He says done.

I have created a persistent herm profiles each with their own soul.MD.

And then you can see here we have scout who is Atlas.

We have Scribe who is Vera.

We have Dev who is Kite.

We have Rich who is Orin.

And he also confirmed that the profiles indeed exist.

Now if you want to also confirm yourself

because sometimes the agents do hallucinate.

It can be that the agent orchestrator told you

especially if you're not using a very good model

or can tell you hey I created the agents

but basically they were not created.

So you can double check this yourself.

You can come into your terminal and just type in a command hemis profile list

and then press enter

and then this is going to list all the profiles that you have on your hemis setup.

So you can see that in my case

I have default which is orchestrator

and then we also have the four agents that we created.

Now here you can see that the gateways are not running.

Their gateways are stopped which is fine because we want the gateway only to be activated when the agents are working.

So we verify that indeed orchestrator created all these agents.

Now this is going to be a prompt that is also very very important.

Now in this prompt I'm telling orchestrator to define this for each agent.

Now I want every agent to have their own dedicated memory.

Imagine you have five agents.

Every agent has their own particular role.

It's going to be a mess if all these agents are sharing one file for their memory.

So agents always write long-term stuffs into a file that is called memory.mmd.

So if all the five agents are using one file, it's going to be a mess.

Now that means that orchestrator is going to dump his long-term memory in there

very it's going to dump her long-term memory stuff in there

and it's going to be a mess.

And additionally, this is going to make the memory file very very heavy.

And remember, the bigger the memory MD file, the more tokens that you use.

And that is why I typically like to go with a dedicated memory for every agent

so that every agent writes own stuff into their own lane.

And every agent is also going to have their unique identity,

their dedicated folders and role boundaries.

So here we are telling every agent to only do stuffs that are within their role.

So for instance, if I go to my agent who does my research like Scout

and I tell Scout, hey build me a dashboard,

and Scout will tell me, hey, no, probably you should talk to Jeff because he's a developer.

Now this is quite important because for instance imagine that you have an agent who is your finance

and this agent maybe has access to your PayPal has access to your stripe and stuff like that

and you give your finance agent the particular rules

that he's supposed to work with maybe don't withdraw money from my account and stuff like that.

So finance agents knows these rules

but if another agent comes in and does that particular thing

and the agent doesn't even know what the rules that you give to finance is

then the agent can basically mess up your setup

and that is why I always have this

sort of role boundaries where every agent understands what their particular role is

and then here orchestrator is working on that as well for us

and here orchestrator is done as well

all the agents have their dedicated memory and

he also created dedicated folders for all the agents

and he also tested the role boundaries

for instance he asked Vera, Vera, can you write a code for me?

And then Vera responded, hey, that is Kite's department.

And he also asked Kite's ver research and kite responded that is Atlas department.

So that means that the role boundaries are working.

Now in the next step, we're going to create a project folder that is going to host our project.

And very important, we're going to create the agents login database system.

Now, whenever the agents are performing a task, we want them to log whatever they do.

So we're going to build sort of an agent login database.

The idea here is that before an agent sends you a message that, hey, I'm done with this task.

The agent is going to log their name into the list,

they're going to log exactly a description of what they did.

They also going to log the time stamp that they did this

and they're also going to write the AI model that they use for that particular task.

And then we're going to do this using what is called an SQL lightsaber.

We're going to build a small database server on our server

where the agents are going to log all the activities.

For instance, when we come to the dashboard I'm going to be building.

When you come to the agents tab, you can see that we have this heat map

where we see when every agent was active.

And here for instance, the agents also post the work that they did,

the most recent work that they did.

So this gives you full transparency of what the agents are doing.

So that if an agent is stuck somewhere, then you know that the agent

for instance didn't perform that task.

And by the way, this is the dashboard that we're going to be building.

And for that, I'm going to give this prompt to orchestrator

to build this agent login database system for us.

So orchestrator is done with the agent login database.

And he also did a smoke test where we can see

dev built the cadence store the AI model that he used

and the state that he's completed and when he performed this task.

Now remember we have created a base but the agents do not know.

So currently it's only orchestrator that knows that such a database exists.

So we're going to have to roll this out to all the agents

and then we are telling the orchestrator

hey save this operating rule in your own memory

and then first distribute it to Atlas Vera and all the other agents.

And this is the instruction that we are giving to them.

Before sending any response,

log what you did into the kid into the cadence runs table

by running this command.

So that's going to be the command that you're going to run

to basically insert what they did into the database.

I'm going to copy this

and I'm going to send this to install this

to make sure that all the agents are aware of the setup that we did.

And after that I'm also going to tell him

hey have every agent run a smoke test

so that they lock something into the database.

Now this would help us

for us to know that

all the agents have access to

this database that we created.

Now when the agents are working

sometimes you get a prompt like this

where you're going to have to allow them to be able to run some commands.

In my case you can either go with allow once

allow only for this particular session

or you can also allow this always

so that they don't ask you this every time

and then that is also done.

So he also did a smoke test

where all the five agents that we have performed the smoke test

where they just completed the AI model that they used

and the time that he performed

and the date and time that he performed this

task.

Now we're going to move in to one of the major parts.

So currently you can see that we are sending all our messages in the orchestrator channel

because so far we've been working with orchestrator

but now we're going to move into the aspect

where we're going to start building the dashboard

and for that

we're not going to be using orchestrator.

We're going to have a dedicated agent kite

who is our designer

and our developer

who is going to be creating the dashboard for us

because he's going to be our developer.

We're also going to have a third channel

which we're going to call studio

and that is going to be the channel we're going to use

to brainstorm ideas on new content to create

and stuff like that.

Now, by default, the issue is that Telegram doesn't allow you to use one bot token to talk to several bots.

So, with Telegram, one bot, one token.

So, if we want to create another dedicated Telegram channel

where we're going to talk to Kite,

then we basically can't use orchestrators B token.

We have to create a dedicated bot token for kite.

Now, there are workarounds around this

like I did in my earlier video

where we basically created a router plugin.

So what the router basically does is that

whenever you send a message in a channel

it first goes through that single B token

and the router determines

hey this is a message that is coming from orchestrator's channel

so let me send

this to orchestrator

the same thing that also happens

for instance if you have an agent that is called

a message comes from that scribe telegram channel

then it goes to your same bin

and then the router determines

hey this is coming

from thread 4

because every channel has their dedicated ID

so with that ID the router is able to determine where the message is coming from

and then it routes it to that particular agent.

Now, this is quite useful if you have different channels that you want to create.

But in our setup, we're going to create only three channels.

So, I'm going to show you another workar around

because with every tutorial, I try to bring in a new flavor

so you guys can learn also new things as well.

So, what we're going to do in this setup is that

we're going to create another bot token exclusively for kites.

So, that means that every message that goes through this particular gateway or this particular telegram bot token would go directly to kite.

But then we have another challenge.

So by default whenever you create channels in telegram

all the messages that we create in these channels

are going to go through one bot.

So which is now our case

every message is going to go through orchestrator

but which doesn't make sense right

we want orchestrator to only listen in a particular channel

and we also want kite to only listen in the channel that we'll create for him.

So what we're going to do is we're going to tell orchestrator to build sort of a plug-in

where orchestrator would only listen in the channels that are dedicated to him.

So orchestrator would never listen in the channel that we would create for Kite.

And the same thing also applies to Kite.

Kite would only listen to his channel that we'll create for him.

This would all make sense soon.

Trust me.

Because this is very very complex.

I typically like to brief the agents

whenever we want to create something complex

for them to know what is coming.

So here I'm sending this message

to prep him on what we want to do.

So I'm telling Oka,

hey, we're going to build something together.

The goal is that kite by default is going to build our studios dashboard

and I want to talk to this engineer directly in his own group

where he would have his own bot and his own identity

and then I'm explaining to why we need a second bot

and you can also read this to understand basically the reason why we need the second bot.

So I'm basically telling him hey this is what we're going to build so that can understand what we going to do.

Now, whilst orchestrator is working on this,

we're going to have to create a second bot token for Kite

because remember that Kite would need his own dedicated bot token.

Orchestrator responded that he understands

the goal is that Kite has his own Telegram presence

where we can speak to him directly.

But we first going to have to create this bot for Kite.

And to do so, we're going to come back to our bot fer that we used to create the first B that we have forestrator.

And we're going to create a second bot.

So, just type in new bot just like we did for the first time for orchestrator.

And we're going to call this bot km kite content.

Let me call this content kites.

Right?

And I'm going to copy the username.

We press enter.

And just like we did earlier, I'm going to use the same name that I g.

But I'm just going to end with a bot because every username with telegram for a bot has to end with the name bot.

And I'm going to press enter.

And then we get a dedicated token for bot.

Remember this was the token that we had for orchestrator.

And we have now create a dedicated token for kite.

So we're going to have to add this bottle for kites also to our hemis setup.

And to do that I created this command for you here.

Come here.

Copy this command here.

Come back into your terminal instance.

Let me clear the screen.

Very important.

Make sure that you're not on root.

You are on Larry or the new user that you created and just paste in this command.

Now this command is going to ask you to put in the bot token.

So I'm just going to run the command here and then it's going to tell you hey paste your kite token and press enter.

So I'm going to come back to Telegram.

I'm going to click here.

Make sure that you are copying the one for kite.

I'm going to click to copy this.

I'm going to minimize this.

I'm going to come back to the and I'm going to right click to paste this in.

And then the key is pasted here.

And after that, I'm just going to press enter.

And this is going to save the keys into our dashboard.

So, let's go back into our setup.

And then let's go into our group.

And then I'm going to tell the I have created this B token for Kite

and I have also added the B token myself in this space here.

So, I'm telling him to verify that the token that I have added exist

and also to do a few sanity checks to make sure that it would work for Kite.

I'm going to copy this and I'm going to send this to orchestrator.

And also what is quite useful and what I always recommend to you guys is that

whenever you're giving your agent a complex tax that you're not sure if the agents would be able to pull this off,

you can always give the agents reference

to the hemis.

So for instance, if I'm telling you, hey, if you are unsure, please look into the hemis in this part here

so that the agents would basically understand how he's supposed to build this.

Then I get this command approval from orchestrator.

I'm going to click on always.

And then orchestrator said, hey, step one is completed.

He's verified that the token that we added exists and he's also saying that hey the bot is live.

We can basically send stats to him to directly message Kite.

Now what we're going to do now is that we're going to create the channels that we need.

So we're going to create one channel for Kite.

We're going to create one channel as well for studio

which we're going to use for the brainstorming

before we even bring Kite into that chat.

So to do that we're going to click on the three dots here.

You wouldn't see the option to create a group.

So we're going to come back to all

and then you're going to click on the three dots here.

Then you're going to create a topic.

Now you're going to have to give the topic a name.

In our case, I'm going to call this kite, right?

And because kite is going to be a developer,

let me just give him a laptop icon.

And after that, just click on create.

Now, whenever you are creating a new topic,

before topic can be activated,

you're going to have to send a message in that topic.

For this topic that we are creating,

we're going to need the thread ID for this topic

because it is with this ID that orchestrator would bind this topic exclusively to kite.

So to kick off this group,

I'm just going to copy this prompt here

and I'm going to send this immediately to orchestrate that

hey, what is this topic's thread ID?

You can send it any message.

We could have sent in hi just to create the channel

and then we could follow up with a message like

what is the topic thread of this ID.

I'm also going to show you that other approach as well.

Now kite is created.

We have the telegram topic ID.

We have the thread ID.

Come back to all

and then here click on create topic.

And then we're going to create the next topic

which is going to be the studio.

Now remember the studio is where we're going to be brainstorming ideas

on new content, promos, ideas and stuff like that.

So I'm going to call this idea with this icon here

and I'm going to click on create

and to trigger and create this as well.

I'm just going to use a welcome message like

hey what is the topic ID of this group

and then here as well we're going to get a response.

Now remember we have created kites

but still it's orchestrator that we are talking to

because kite hasn't even been added to this group yet.

Right?

So all the messages that you are seeing here

the thread the response to the thread ID for kite

the response ID for studio

all of this is still coming from orchestrator.

So now we have the thread ID.

So we're going to send a message to orchestrator that hey now we have the thread ids.

So we want you to wire our telegram group.

I'm going to give orchestrator telegram group ID of our current group.

And I'm also going to give him the thread ids of all the channels.

So he would have the thread ID of where I'm talking to him.

He would have the thread ID for kite.

He would also have the thread ID for the studio group.

And then only with this he would be able to bind the channels to the particular agent.

So for orchestrator would have access to this channel that we are talking to him

and orchestrator would also have access to the studio channel

because that is what we going to he's the boss

and basically we will be brainstorming the ideas with him on the studio channel

but kite would only listen in this channel

and kite would have no access to the orchestra channel

and also to the kite channel.

I'm going to open notepad

and I'm going to paste in this prompt

because we're going to change a few things in there.

The first thing that we're going to need is the telegram group ID.

Now to get your telegram group ID.

We're going to use the same bot that we used to fetch our telegram ID called user infobot.

So when this opens, click on forum down here

and select the group that you just created.

In my case, this is KM content.

And then when you click on this, just press send.

And then the bot will send you the ID of the Telegram group.

You can click on here to copy this.

And I'm just going to replace this with this section here called Telegram ID.

In the next stage, we're going to fill in the thread ID for the three channels.

And for that, I'm going to come back into our Telegram group.

But in most of the cases, the first channel that you create always has a thread ID one.

So most likely orchestrator would reply that is one.

But just to be sure, let me just ask him this as well.

Okay, she responded that his thread ID is one because that is that was created.

So I'm going to put in one here.

I'm going to go to kite.

Kite was 36.

I'm going to put in 30 and I'm going to come to studio and studio is 40.

And then basically we're done.

Only these are the four parameters that you're going to have to fill in in this prompt.

And after that,

copy the message.

Very important.

Don't do this in studio or kite.

do this in orchestrator's channel

and then paste this message

and send this message to orchestrator

to basically implement the topic lane agents binding

and he orchestrator responds

hey I've saved the

four lane members to memory and completed the save part

but there is an issue that he's having

he's not able to run this command to restart the gateway

so he's telling us if we can run this for him

so I'm just going to copy the

command here

I'm going to come back to my terminal

let me just clear the screen to make it a bit cleaner

and I'm going to paste in this command here

very very important again

make sure that if you created a new user you are on the new

user not your root account

and and press enter.

And then it happened as well when I was testing this that you get this response failed to connect to bus.

So just copy everything here and then come to telegram and then tell okay I got this error and just paste it.

Oops, I think I didn't copy it.

Well, let me come back into PowerShell control shift C to copy this.

I'm going to come back to Telegram and I'm going to paste this.

And then he would basically correct this and give me another command that I can use to fix this issue.

And I should have found this.

He says the error means your shell didn't use a bus environment.

So he's basically giving me another command to run.

So I'm going to copy this and I'm going to clear the screen again.

I'm going to paste in this new command.

I'm going to press enter.

And then you see that immediately I get a notification that my gateway is restarting.

So that means that the command is

so it took a few seconds for everything to complete.

So just be a bit more patient when you see this.

So I'm going to copy this here.

Control shift C and I'm going to send this orchestrator it worked out.

This is the message I got.

So you see that when I sent it to immediately orchestrator,

he didn't respond because the gateway I think was still booting all the agents.

So sometimes when you restart a gateway immediately

it takes a while for all the gateways for the agents to get restarted.

So when you send this immediately

and you do not see a response

just be a bit more patient.

And here just as a tip

when you come back into your terminal

and you type in hemis profile list and enter

you would see that now we have two gateways running.

Initially we only had the default gateway that was running

and the gateway for dev was stopped.

But now you can see that we have another gateway that we have built for kite which is running.

So what we're going to do now is we're going to have to add kite into this group that we have created.

And to do that we're going to come to botfather click on the URL that he gave to us to be able to fetch the ID for kite.

Click on the name and then we're going to click to copy the of kite and then we're going to come back into our group.

I'm going to go to all.

I'm going to click on the three dots here

and go to view group info.

And then here I'm going to click on manage

and I'm going to come to members down here

and I'm going to click on add members

and I'm going to put in the username of kitespot.

And I'm going to select this

and I'm going to click on add.

And then now Kite is added to this group.

We also have to make Kite an administrator for him to be able to work here.

Still on the all tab, I'm going to click on the three dots.

I'm going to come to manage group.

Then I'm going to come to administrator

and I'm going to click on add administrator.

I'm going to add Kite there.

Review the permissions.

If you're fine with that,

just click on save.

And then we can close this.

So now we have added Kite to the group.

So what we're going to do now,

we're going to come into Kite's channel

and we're going to send a message like who are you?

And this is the part that always gets me nervous

because that is the point that you see

if the agent to channel binding works.

If everything works, we are supposed to see a message from Kite that I am Kite.

So let's see if this is going to work.

Perfect.

I'm a bit relieved.

So here Kite responded.

I'm Kite Laris Cadence designer and engineer.

Now just to make sure that the other parts are working, I'm also going to send this to orchestrator.

And if this should respond that he is orchestrator.

So if Kite is responding in the shadow, then we are screwed up.

Let me do the same thing for studio just to be on the safe side that we have the right agents behind these channels.

Great.

So in orchestrator's channel he still responded I'm orchestrator which means that this is working very well.

Kite if you remember also responded he is kite.

And then now when I go to studio orchestrator also responded orchestrator is there.

So that means that the telegram channel to agent binding work.

Kite is never going to listen in orchestrator's channel

and orchestrator would also not listen to ka's channel

and that's exactly what we want.

So now that we have all the core aspects of the framework that we need ready,

her message agent is ready,

orchestrator is fully set up

and kite our developer is also set up,

group is ready,

we can now start with the build

of the dashboard.

Now with the entire build of the dashboard,

we're going to be using Kite

who is basically playing the role of a designer

and also our developer.

And before we even let Kite start building the dashboard,

we're going to have to brief Kite on the whole thing

that we want to build from the beginning to the end

and a concept that we have for what we're building.

And for that I'm going to give Kite this prompt here

as a briefing to know what we want to do.

So I just sent this to Kite

and in the process I'll be explaining to you guys

what I just sent to Kite.

I'm telling Kite,

hey before we even write any single line of code,

here is the whole picture of what we building.

So I'm telling Kite

we're building a premium content automation studio

with a beautiful web dashboard.

I'm telling Kite how the whole process is going to work.

So we're going to have Atlas who is going to be our researcher.

basically go and propose some ideas and then we will promote one as a user

and then Vera would write the content of the slides

and after this is done Kite who is also

our developer and our designer would fit this content into one of the templates that we would pick

and then we would render that template from an HTML document into an image with about 10 1080 to 1350p

and then Orin who is going

to be our publisher would publish this to any network that we want to build this on

and I'm also telling Sky that we're going to set this on a small Python server

which is going to be on dashboard

and also very important I'm telling Kai that we're going to give him or a template

that we're going to be working with

and I want this template to be the design source of truth

because I don't want him to change anything that we're going to be putting in the template

and also a few things that I'm giving him

that I want to have a backup protocol

and also some principles of how the whole build is going to be.

So Kite responded that hey understood here is the plan in my own words.

Candice is a complete content automation studio with a polished web dashboard as a control interface

and he's basically just repeating all the things that we told Kite in the sequence

how the whole flow is going to work.

So Kite is basically ready to fill in to build the dashboard with us.

So in the next stage

we're going to tell Kite to build the Python back end

on which this web dashboard is going to be running on.

I'm going to give this prompt here to get that ready.

I'm telling Kai that hey we're going to build this dashboard in a way that is going to be very secured.

So for the first part

we're going to be setting this thing

on what is called SSA tunneling.

So that means that whenever you want to connect to the dashboard

you're going to use a command like SS minus N minus L this port here

and at the end you're going to put in your username and your IP address.

But I'm going to show you

how this would work.

Now the reason why we are using this in the first place

is that the web dashboard is not going to be available to the public internet.

It's only someone who knows the password of your VPS server that would be able to connect to the dashboard.

And that is why it makes it so secured.

But even after we are done with the build, we're going to be installing tails skill that will save the dashboard for us.

Now tail skill is one of the most solid security setups that we can use to get our dashboard online.

So for the first time, we're going to use SSH tunneling to access the dashboard.

But when we are done,

we're going to be installing tail skill

which is going to host the whole thing for us.

And the best thing about tail skill

is that we would then be able to access the dashboard

on any PC that is running our tail skill account

or even on your mobile phone.

All right.

So kite is done.

He set up the back end that we used to run the dashboard

and he also gave us the SSH tunnel command

that we're going to use to access the dashboard.

So what we're going to have to do

is to just copy this command

that Kite has provided for us here.

We're going to go back into PowerShell

and then in PowerShell just put in this command

ssh minus L.

This is the port

that we use to save the dashboard.

This is the security end

and also this is the your username

which in my case I created a new username

if you remember called Larry

and this is the IP address of the VPS server.

Now here I'm going to put in my password and then just press enter.

Now it's going to bring you to this point here.

It's going to be a ca just blinking.

That means that you are connected to the back end of the VPS server.

So what we're going to do now is we're going to open a browser

and then just as Kite gave to us, he says that we should open this URL on our browser.

We're going to copy this URL here.

We're going to paste this in here and press enter.

And then you see that we are connected to the dashboard.

But because we do not have anything there yet, that is why we get this error response.

But in principle,

we are connected to the back end of the Python server.

Now what we're going to do is in the next thing

we're going to have to put something on the dashboard for us to see

basically what we're building.

And for that we're going to tell Kite to create a web upload page.

Now in this web upload page

we're going to upload all the files

that is going to save the template

of what we want to build.

I'm going to copy this prompt to Kite

and I'm going to send this to Kite

to build the upload point

where we can upload all the dashboard stuffs

that we need.

And this would contain the HTML template to the dashboard

and all the the slides

that we're going to be using

when we building these carousel slides.

Whilst Kite is building that, let me show you how the template looks.

So this is the empty template that we're going to give to Kite to upload on the dashboard for us.

So we see that we have the content tab which is a bit populated.

The idea tab looks empty.

The studio tab also looks empty.

Everything looks empty.

But once he's done, he's going to basically add this to the back end

and then we're going to populate this dashboard template as we build.

All right.

So Kite is done.

He's built the back end which we're going to use to upload all the files

that he would need to build the dashboard.

And he gave us this URL here for that.

So I'm going to copy the URL.

I'm going to open the browser and

I'm going to paste in this URL.

Hopefully, we have a web upload upload dashboard.

Okay, so this is working quite well.

So, he built the cadence upload back end

and he's telling us that he expects four files.

So, the template,

so the HTML templates

and three JSON files

which are going to be the templates of the carousel slides

and for the reuse.

So, I uploaded all the four files that are required

and also this is confirmed on the dashboard

that these four files are saved.

So the next point

what I'm going to tell Kite is that

hey I have uploaded all the four files

that are required to build the templates

to build the dashboard.

So I want him to serve these files on the Python temp

so that we can actually see that this is working.

So what we're basically telling Kite in this prompt

is that the file that we upload the template is called cadence dashboard template.html

but I want him to rename this to index.html

into the dedicated folder that we have created

for the dashboard

that we're going to be building.

And also very very important at least in my case

if you want to change the templates then you don't need to give Kite this instruction

but what I'm telling Kite is that hey this is going to be the design source of truth.

So whenever he's building anything he should always check for the template

to make sure that he hasn't made any design changes to the temp

so that whatever he's going to be building is going to be identical to what we have in the

templates.

on telling him that hey the template contains basically mock values like hardcoded stuff

so that nothing basically looks empty

but those numbers that we have on the templates are going to be replaced with real metrics

once we

start building the dashboard and this one of the nice things not that I like about Hermes

so what you can see here for instance is that he's done with basically the work that we give to him

and you can see that he's using an inbuilt browser to check to see if everything that is there is correct

so he's clicking and taking browser snapshots

he's clicking different sections and taking snapshots of that aspect

just to be sure that whatever he did is identical to what is in the

template.

And this is one of the main reasons why I really like working with Hemis

because I build dashboards a lot.

And if the agents or Hemis allows the agent to be able to work with a headless browser

to be able to see what it's building,

this is really cool.

All right.

So Kite says he's done.

He says the template is logged in but is partially complete because there's one file that is missing.

So when I look here, he says that he didn't find the templates packed because it's false zero.

Let me see why this is the case.

So it could be that our upload didn't work.

So he's saying that we should re-upload this file through the upload point.

So I'm going to put in this here again and I'm going to re-upload the templates pack.

So I'm going to tell Kite I have re-upload templates pack.json.

Can you confirm?

Perfect.

So it appears that something happened the first time that I uploaded this.

So he says that yes,

he can see it now.

He checked the file

and all the uploaded source files are now present in the folder

that we require them to be

which is perfect.

So now let's go to the dashboard

and see if the dashboard is actually rendering

on the Python ser.

So to do that

we're going to use local host to do that

we're going to use local host col8892.

The 8892 is the part of the server

and the local host is basically the local area on our server

that we're going to use to access the dashboard.

And then when I open this hopefully beautiful.

So you can see that now we have the dashboard loaded just as we had in the template.

So let me open the template for us to do a one is to one copy.

This what we had in the template and this is what is being saved on our dashboard.

So all the pages are identical to what we had in the template.

So kite is doing a pretty good job.

Now, in the next stage,

we're going to,

this is one of the things that I always recommend

when you're building stuffs with the agents.

Let the agents always create a copy of the dashboard

before they make any changes.

I always say this in every video

that I worked on the dashboard for several days

and my agent just made a mistake

and basically I lost everything.

And that is why it's very important that you tell your agent that,

hey, before you make any changes to the dashboard,

please create a copy of the dashboard and save it somewhere.

And also on the dashboard, we have this section here that is the version number.

And I want him to update a version number anytime he makes a change to the dashboard.

So I'm going to copy this prompt

and I'm going to send it to Kite to implement this

and Kite is done with that as well.

So when we come to the dashboard

and we reload the template.

So we should now see that he has changed.

So we used to have editorial.

Let me open the template for you to see what this looked like before.

So in the template in this bottom section here

we used to have version 0.1 editorial

and you can see that now this is updated in our version

to version 0.1

which means that kite is going to keep the version update always

updated as we move on.

So currently we're going to start with the build.

So all the parameters that you see here in the content tab are basically hardcoded from the template that I built

and also the all the stuffs that we have in the agents tab, the heat map, everything is hardcoded.

So we're going to tell Kite, we're going to give Kite a prompt to basically update the dashboard metrics that we have here

and also the agents tab so that we have real data, real data from our hemis setup that we have

but not from the hardcoded stuff that we have in the template.

And I'm telling him that hey

when you are done

confirm that both tabs show the real data

and the heat map matches the run table.

Now for the heat map

all the information is going to be pulled from the agents login database that we built.

If you remember

we built something where every agent is going to log the activity

whenever they do something

whenever they perform any action.

All right.

So Kite is done with that as well.

Let's look at what Kite built.

Let's reload the dashboard.

So here you can see that he's now pulling the real metrics from our data agent database that we built.

So you can see that now kite is trying to pull the real metrics from the agents login database.

So you can see that currently in the agent workload

because none of the agents have done something today

that is why everything stays at zero for atlas vera

and origin and kite is the only one that'll be working with today.

So kite is basically at 100%.

And the same thing when you come to the agent database

you can see that everything is almost zero for all the agents

because they haven't done anything today.

So I'm recording today on the second day.

So basically I would expect that there would be some metrics from the previous day.

So I'm going to tell Kai that hey it appears that the metrics that are being pulled are only being pulled from today

but there should be some activity from the agents the previous day.

So he should make sure that the dashboard doesn't only reflect today

but basically reflects all the data that we have in the agent login database.

So I'm telling Kite hey this looks awesome.

Thanks.

However it appears that the agent metrics being displayed are only from today.

I do not see any data for instance from yesterday.

even though the database should contain this.

So he should basically investigate and see what is wrong

and make sure that we are serving all the data that is currently in our agent login database.

All right.

So Kai says he's done.

It appears there was an issue.

So he says that the real problem was there was this issue with the test server Python server

that was basically deleting stuff from the production database.

So he basically fixed this as well.

So he says that the fixes are made.

The test no longer touch the production database

and he's basically put in some metrics

some implementations to make sure that this does not happen.

But what I would like Kai to do here before we even proceed further,

I want to be sure that this problem is fixed.

I'm going to tell Kai that hey

perform a quick end to end smoke test

to all the agent

trigger a few test activities

and verify that the resulting data are saved in the database

and also correctly reflected on the dashboard.

If there is anything that is not there,

he should fix it as well

because the agent login database is going to be an integral part

of what we are going to be building.

And that is why I want to be thousand% sure that the agents are logging

and also that all the things that they do also reflect here on the dashboard.

And by the way, all these deep prompts that I'm adding,

I'm going to make these available to you.

So that should you have the same situation,

you can just copy paste the prompt and give it to the agents or Kite to fix this for you.

All right.

So Kite says he's done with the smoke test.

He repeated this across all the agents.

And currently you can see that all the agents completed.

When we come back to the dashboard, you can see that now we don't have the 00 that we had.

We basically have metrics from all the agents that is being pulled correctly from the agent login database.

And the same thing also applies to the heat map.

And when we come to the dashboard, we also see that the agents total runs are also correct.

There is one more thing that is not correct here

which has to do with the agent workload distribution.

This would always have to sum up to 100

so that we can always see which agent has the most workload.

I'm also going to tell Kite to fix this

and after that we can continue with the build.

And for that I'm telling Kite,

hey on the dashboard homepage,

the agent workload share distribution does not make sense.

So basically I'm telling him

that everything should add up to 100%

but currently everything adds up to 350.

The idea is that

whenever I look at this

I'm supposed to see which agent has been more active.

So he should basically correct this for us as well.

And Kite has done with that as well.

So basically now

when I look on the dashboard

you can see that the distribution now makes sense.

It all adds up to 100%

and I can see clearly that Kite is the most active agent

because he is the one building the dashboard anyway.

So now the agent login database is working correctly.

all the metrics that have been put on the agent tab are also making sense

so we can start with the build of the other components of the dashboard.

I'm going to explain the concept of what all the prompts are going to do

and I'm going to put in all these prompts in one go

so that you don't have to sit and watch me put in one after the other.

Now in the next prompt we're going to build an exporter

and let me explain why we need this.

So initially, previously whenever you wanted to build a carousel image like a slide or something like that,

you always had to design everything in a photo editor, right?

And then save it, export it as a picture

and then you had your carousel image.

But the trick what we're going to be using here is

we're going to render every slide.

We're going to build every slide as an HTML.

Now, if you build a slide as an image, you pay for every image that you generate.

So if you want to change any text in the image, then it means that you always have to pay for API calls,

which makes it not fun anymore to work on slides.

So, we're going to build this in a smarter way.

What we're going to do is we're going to build every carousel as an HTML file.

Now, because it's going to be an HTML file, it will be very easy for us to edit any text that we want to edit.

And let me just go into my setup

just to explain for you guys

to get a concept of how we're going to build a slice.

So, let me take this one here for instance.

Everything that you are seeing here looks like an image,

but it's basically not an image.

It's an HTML document that is rendering the file like this.

So whenever I want to export this as an image

then I'm going to deploy what is called a headless browser.

So a headless browser doesn't need a desktop to run.

It's basically going to render this file here

which is basically an HTML file

and then take a snapshot of what it sees.

So eventually all the things here that we are seeing

which is rendered as a code

would be then exported as a JPEG

and then with that you can upload this to Instagram,

Tik Tok or whatever

wherever you want to build this on.

Now the reason why we do it in this smart way

is that originally whenever you want to build any carousel slide

you have to use an image generator.

Maybe you can use nano banana

or all the AI models that are out there to generate images.

But when you even want to change one text

then you have to tell the AI model again

hey I want to change the text.

Maybe for instance I want to change this from most people

to most users

and I have to tell my a model again

hey I want to change the text from most people to most users.

And then you pay for every edit that you do

which is probably not the smartest way.

So how I build this

and what most users are using

is that you build the whole slide as an HTML file.

And by so doing

when you can change everything that you want to do in the file

without having to regenerate the prompts

or regenerate some API keys.

So here for instance this is what our dashboard is also going to do.

When I click on fine-tune

and I pick this element

I can just change the text here from people to most builders, right?

And you see that it does

I can save and rerender

and then I have my new slide

without me having to do any API call.

the same thing.

I can change all the contents.

This is why I really like what we're building here.

I can change the layout.

I can move every content here,

change the size,

put it in every format that I want to do it

without me having to pay for any API calls.

So, what we build,

we're building every carousel slide as an HTML file

and then when we are ready,

we deploy a headless browser

that looks at that HTML file we have created

and creates a snapshot of as a picture

that we can then upload to more

any social media platform.

So we avoiding the aspect of complicating the whole carousel image design

where you have to make an API call for every edit

and pay for every edit.

We are rendering everything as an HTML file that will be completely edited.

Quite crazy the things that you can do with standard HTML.

Like you see here all these templates that you are seeing here,

none of it is a picture.

Everything is built with standls that someone might not even believe you that this built with HTML.

So basically with this prompt here,

what the prompt is doing is that we're going to build this setup of this headless browser

that would render whatever that we build from an HTML file into an image

that can be uploaded on any social media platform

that we have.

So we are not building images straight ahead.

We build an HTML file which contains the structure that we want.

Then once we are satisfied with it, we export it into an image.

And for this we are using a headless browser that is called Chromium.

And by default, most of the Hemis setups come with Chromium.

So we might not even need to install this on our own.

We can just leverage what Hemis already has and then use it to build what we want to build.

If you remember when we were building the dashboard, there was a aspect that I told you that the agents were basically.

So here for instance,

you can see that Kite was rendering a browser

where he was going to the dashboard

and making sure that everything was working.

That is the same concept I'm going to use.

We're going to let the designer in this case

as well is going to be Kite

use the same headless browser.

But right now he's not going to use it to check the dashboard,

but he's going to use it lots of what the HTML file contains.

It makes it so easy to be able to build slides without having to pay for API calls every time.

And Kite is done with that as well.

He did a smoke test

and he verified that the images were indeed being rendered

from the browser

into for instance here a PNG file.

Now in the next stage we're going to build all the things that are going to run from the back end.

So I'm going to explain all these backend prompts and then I'm going to give these to kites

but I'm going to explain all these prompts to the point where we get to a prompt that would trigger a front-end design.

So when you are using the prompt

everything that I say

backend back end

I try to add this

so that you can see what to expect.

If it's a backend prompt

then it wouldn't show anything on the dashboard.

It's basically building the foundation

on which we would build

to show something on the dashboard.

When we get to the aspect here where I write design

then it means that what you going to do is going to show a design.

But if I get to an aspect where I realize that it's quite useful to explain

I'm going to explain this to you guys as well

so that you can also understand what these prompts look like.

And in every case

I always include sort of a description

to basically explain what the prompt is going to do

and what you should expect.

So now we are done with the exporter prompt.

The next prompt is going to be the template packs auto fit.

Now whenever we are generating an idea, let me use my setup to give you a concept of what I'm doing.

So when I click on generate ideas, let's see.

I already have some ideas that are already generated here.

So let's look at for instance building a one person AI operation team.

Now when I like this idea

and I click on promote

basically I can decide on the number of slides

that I want to generate.

So let's say for instance

I want to generate six slides

and I say write slides.

What is going to happen is that Vera

who is going to be our right

is going to write the slides for the carousel

or the content or the text for the carousel

that we want to generate.

Now like I said all these things that Vera is going to give to us is just going to be raw text

but we have to be able to insert these text into the template.

This prompt that we're going to build here is going to build that part

where it's going to take the content that has been generated by Vera

and insert it into the HTML template that we have.

So all the templates that you see here like I said these are basically HTML templates.

So for instance when I look at this example here you can see that some of it have longer word characters than some.

So this one for instance has longer characters.

This has a different layout and stuff like that.

So then what is going to happen is that the back end that we're going to build is going to look at a template that you select and

then adapt the text that Vera generated here to fit that template.

And this is what I call the auto fit feature.

And for this to work,

we have to build a render engine that is going to look at the content that Vera generated

look at a destination template.

So here for instance,

if I say that,

hey,

I like this content.

When I come to approve and pick templates,

then it ask me which of the templates I want to use.

So whenever I select the template that I want to use,

then the back end that we're going to build here is going to look at the template that we selected,

look at the number of words and characters that are in there

and then adapt the text that

Vera generated to fit that template.

Now imagine that if the words that Vera generated are way longer than what the template can fit,

then the carousel just becomes ugly and that is why we build this render take care of that for us

and after that we're going to build a

pipeline to talk to the crew.

So currently whenever I'm talking to the crew right

I'm chatting with Kite able to chat with Vera

I'm not able to chat to our publisher

but we want to be able to chat with them from the dashboard

and by chat what I mean is that when I come

to the idea tab and I click on generate idea

this sends a message to Atlas who is our research

look for ideas of what we can build.

I'm talking to Atlas through a code

where I send a trigger to Atlas that hey generate some ideas

and the same thing applies for all the other agents.

Whenever we want to publish something

we're not going to come to Telegram and tell them to publish.

We want to push a button the dashboard

that is going to trigger the agent to publish the content.

And these are the things that prompt 17 is going to do.

It's going to build an interface

where whenever we click a button on the dashboard,

it sends and triggers an agent to do that particular task for us.

And then we're going to have prompt 18.

Prompt 18 is going to be exclusively for Atlas

who is our ideas guy and our researcher.

And I'm basically just telling Atlas how I want the carousels to be designed.

So here, these are just examples of what I put in.

I'm telling Atlas that hey whenever you're writing the carousel slides

I want to have this written in plain simple language

short words short sentences

no hype no buzz words and stuff like that.

So this prompt is basically giving Atlas an orientation of how we want the content

and also Vera who is our content writer how we want the content of the slides to look like.

And here we're also giving them an orientation on how he's generating an idea.

He should generate four ideas

and basically an orientation on how we want the whole carousel slides to look like.

And then we're going to have this last prompt here

which is going to be basically for our publisher

whenever we want to publish.

So for instance when we come to the dashboard

and I come to my studio tab for instance

if I want to publish something here

I'm just going to click on post

and then this can be published directly to Instagram

to Tik Tok or wherever you

want to apply this to.

But for that as well we need to build a backend

and that is what this prompt is going to do.

This prompt is going to build a backend

so that whenever we trigger an upload

then the upload can be pushed directly to whatever social media platform

that we want to push this in.

So I'm going to render all these prompts.

The first prompt being the one that is going to take the text that has been generated and adapt it to the template.

The next prompt

which is going to be taking care of that

we can use all the buttons on our dashboard

to trigger an action to the agent.

The prompt 18 which is going to take care of how we want our ideas to be generated.

The number of ideas that we want to be generated,

how we want the cast,

how we want the carousel slides,

the content to look like.

And we would also have the last prompt

that is going to take care of publishing to the server

or the social media platform

that we want to publish this to.

So I'm going to run all these prompts.

If I get any backs,

I'm going to pause

and I'm going to explain this to you

so that you guys can also follow

what we would do there.

Right.

So Kite is done with processing all those prompts as well.

Like I said, I didn't have to respond to anything.

I just copy pasted all the prompts to Kite and he processed all the back end.

Now in the next stage, we're going to grab some API keys and I'm going to explain every API key and why we need that.

The first API key

that we're going to search

is for publicity.

So perplexity is one of the AI models that you can use that is very very strong in research.

So I added this feature

so that when you are writing a carousel

and you want the agents to really use recent data

then for instance you can use perplexity as a research model

where Atlas is going to query perplexity to do the research

before the contents of the slides have been written.

Just to give you an example of where we would need this.

For instance, when I come to my dashboard and I click on generate ideas,

you see that I have a couple of ideas that have been generated for me.

So for instance, let's go for the builder AI automation map.

Now when I want to promote this article or when I want to promote this idea into slides,

I do have this option where I take research with publicity.

If you take research with publicity,

then the agent will query publicity as a research tool

to gather this information on this new idea.

If you do not take this article,

then the agent will just do a standard web search

to get ideas for this content

or probably what the agent is also already trained on.

But if you want really research based content,

then you can toggle this research perplexity

and then the agents will basically do a researchbased content

for the article or for the idea that you want to create.

So that's going to be the first key that we're going to get.

Now to get this,

you can come to Now to get this,

you can come to console.plexity.ai.

I'm also going to provide a link to this in the description section.

Then when you're here, just create an account or login if you already have one.

And then when you are logged in into the console, you do not have to do much here.

You can just come to the left section here and click on API keys.

And then under API keys, just come to the top right section here, click on generate API key.

And you can put in web key you want to go with.

For this tutorial, I'm going to call this comp mechanic tutorial test

and then just click on create a key.

And after that, you see that an API key would be generated for you.

So just copy this API key, save it in the safe space,

and we're going to use this API key.

The next stage we're going to get an API key from imageB.

But let me explain to you why we need this.

So what we want to do with our setup is that we have our cadence

and we want to push images from here to Instagram for an example.

Right?

The issue is that if we want to push our slides directly from cadence to we need to get Instagram API key.

If you want to do this for Tik Tok, you need to build your own Tik Tok flow.

If you want to do this for Facebook, you have to create your own keys,

your own flow and stuff like that.

And this is very very complex, right?

We're going to spend more time building those connections

than basically focusing on the dashboard

that we want to create.

So what we're going to do is that

there are some platforms like buffer or metrical.

And what these platforms do is

they have access to all the social media platforms

and they have worked on all those backends for you.

So all we would have to do is to push our slides through a platform like metricle or buffer

and they would push the slides to the social media platforms that we want to go to.

There is one c however that at least from my testing buffer doesn't allow us to push our images directly to that platform.

So what we're going to need is an intermediate storage where the images that we generate will be pushed to

and then from there the images would go to buffer

and for that we're going to need a platform called image BB.

Now this BB is free.

So what is going to happen is that whenever you push an image

or whenever you push a story or a carousel from our cadence

this would be first be pushed to BB

and then from image they would go to Instagram.

Now, it looks complicated from the setup that I'm explaining to you,

but we're going to wire this in.

And when you click on post,

like for instance, when I come to my dashboard here

and I click on post

and I choose, for instance, I want to post to Instagram.

This is going to happen.

All this transfer is going to happen automatically.

So, you do not have to go here, push from here to here, push from here to here.

We're going to wire in all those connections.

And to do that, we're going to come to image.b.

It's completely free.

So, you can just create an free account.

And then once you create the account,

come to the left section here and click on API.

And this brings you to the API page.

When you are here, just click on add API key.

Create an API key, store the API key,

and then we can proceed to the next API key that we will need.

Now, the next API key that we will need

is the platform that is going to push our slides

to the social media platform that we want to go to.

Now, here there are several platforms that are out there.

As an example, we have Buffer who can do this for us.

And we also have Metro who can also do this for us.

Both of them work.

I know that a lot of people use metrical because they have a very very powerful dashboard that they use.

But from my testing, I realized that for you to be able to push our slides from Cadence OS to Instagram or Tik Tok,

if you want to use metrical, then we have to go for a higher payment plan.

It's only the advanced option where you're going to pay 43 in my case a month that would come with the metrical API.

So this was quite expensive for me

because in my case

if you really need metric code to work with kers

then you would have to go with advance

where you pay dollars a month a month

to be able to get access to their API.

Remember an AP is an interface that would allow us to be able to communicate with the platform.

And in my case I had just two social media accounts that I was using to test.

So I didn't want to pay €43 a month just for two social media platforms.

But advantage with metricle is that you can go for instance with their starter plan

where you would pay $16 a month and then you can add so many social media platforms

without having to increase your pricing.

So for instance if you have four social media platforms Tik Tok, Facebook, LinkedIn etc.

you would still be paying $16 a month which I find pretty cool

and I'm going to show you when I would go with metric.

We have another platform that is called buffer

and that is what I used in what we are building with hidden.

Advantage with buffer is that they would allow us to be able to use their API assets

even with a standard plan.

And I think in my case I went with the five plan which is essential right?

So even with this you have access to three API keys

and you would be able to use this

but with metrical you can only get access to the API key if you have the dollar plan.

But the caveat with Buffer is that every social media platform that you add you pay an extra $5.

So if you have Tik Tok, Facebook, Instagram it becomes $15.

If you add LinkedIn it becomes $20

and so it can get a bit expensive.

So in that case, if you don't really care about pushing directly from cadence to the social media platform,

I also did add an option on the dashboard where you can basically download all the slides.

Let me do an example here for you.

When I click on slides,

you can see that the slides are downloaded here.

When I click on this,

you see that I have all the slides downloaded here.

Let me go to thumbnails

and you can see that I have all the images.

And what I also

added for you is that you can click on caption where you're going to see the caption that is going to be required for you to post.

and then you can post this yourself and in that case then

probably metrical would be a best option because

they do have for instance for the $16 plan

you would be able to add more social media

you can see here up to 10 brands

which makes it very cool

but if you really want to leverage the push from cadence directly to your social

media platform then buffer might be a cheaper option

I'm just trying to explain how you can decide on which are the platforms that you want to go for

so for this example we're going to be going with Buffer

because like I said I had only two social media accounts

and I didn't want to pay €43 here for my testing

so you can just create an account with whatever platform that you want to go for

and then just loging into your account.

Once you're logged in into your buffer dashboard,

just come to the bottom section here where you have my organization

and click on API

and then here click on new key

and then you can create an API key.

Now with these three API keys from perplexity, imageb and buffer,

we would be able to wire this into cadence

and continue with the build.

Now with these API keys that we have gathered,

we have to store these API keys to our server.

And to do that, we're going to come back to PowerShell.

I'm going to log in back into the terminal.

come back into the

and I'm going to paste in this command.

Now this command is going to ask us for all the API keys

and these are going to be secretly stored into the

you can in principle just give kite directly into the API keys

but this is not a recommended approach

because if gets compromised then someone has access to your API keys

and that is why whenever I'm adding API keys or secrets

I prefer to do it from the rather than directly giving it to kite or telegram.

So just press enter

and then it's going to ask you for your buffer access token.

So here I'm going to paste in my buffer key.

Paste this in.

And after that just press enter.

And then it's going to ask you if your image BB API key.

I'm going to copy this as well.

I'm going to paste this in.

Press enter.

And then the last one you're going to be asked for your perplexity.

Oops.

I realized that I made a mistake.

So the first one that I put in wasn't for buffer.

It was for perplexity.

So I'm going to cancel this and I'm going to run this again.

I hope it doesn't get messed up.

I'm going to tell Kite in the process that I made a mistake.

I'm going to recall the same command that I used.

I'm going to press enter and then

I'm going to copy the right buffer key.

Then I'm going to copy the image PV.

Then I'm going to copy the Plexity API key.

I'm going to paste this in.

And then you can see that all the characters are saved

and we get a notification that all done.

So what I'm going to do now is I'm going to come to Kite

and I'm going to tell Kite that hey,

we've stored these three API keys.

He should do a smoke test

to make sure that all the keys are working.

So I'm going to copy this

and I'm going to give to Kite.

But because I made a mistake and did it twice,

I'm going to send him a

I'm going to write

I did make a mistake and added it twice

so you can ignore this entry.

Then I'm going to tell him to do a smoke test.

I hope that the first mistake that I did,

he would be able to just rectify it

and see that all the keys are working.

Hey, if you're finding value in this content that I'm creating,

hit the subscribe button,

subscribe to my channel,

and hit the like button for YouTube algorithm to keep pushing this video to other people that will be out there.

It doesn't cost you anything,

but it really does support my channel's growth.

A lot of work goes into crafting all these tutorials,

creating that,

creating the prompt,

and that is why I'm not even able to upload a video every week

because I test everything about 20 times before I push it out there.

And that is why I have to upload videos on a bi-weekly schedule

because a lot of work goes into crafting all these things.

And kite is done as well.

So he did a smoke test.

He verified that buffer works,

imageb works,

and perplexity works.

All right.

In the next stage,

we're going to put in prompt 20.

Prompt 20 is basically going to give us a trigger

where we can publish an idea,

create an idea,

render a content,

render a carousel and stuff like that.

So just copy this prompt here

and we're going to give this as well to Kite.

All right.

And Kite is done with that as well.

Now the next three prompts are basically going to be the design prompts.

This is going to wire the ideas tab,

the promotes tab,

and also the studio tab.

And after that,

we're going to basically test the whole thing.

So now we have this idea tab that has nothing there.

We have the studio tab that has nothing there.

We have the publish tab that also has nothing there.

And also the templates tab.

So what we're going to do is

we're going to give Kite these prompts to wire those up.

And after that,

we can test some functions

so that you guys can see what we are building.

So I'm going to give all these prompts to Kite sequentially

and then we can go to see the test.

All right.

So Kite is done with processing all those prompts.

Now,

I know we've been building for some time now.

Now, let's see a real demo of the end to end to make sure that everything is working.

So, I'm going to come back to the dashboard and I'm going to reload the dashboard and then let's go to the ideas tab.

Now, I see that Kite did a smoke test where he tested one agent tonight before you trusted.

So, he just did a smoke test to make sure that the ideas are working.

Now, let's click on generate ideas to see if the end to end flow with Atlas doing the research,

giving us some ideas is going to work.

So, I see a notification that he's doing the research.

Let's wait till the results come out.

And this worked very well.

So you can see that we have four ideas that was generated by Atlas.

Browser agents should verify, not just click.

Your agent needs a sandbox before it needs more tokens.

Now let's say that we want to promote one of these ideas.

So let's go with stop giving agents your whole app.

And let's click on promote.

And then here,

let's say we want to generate 10 carousel slides.

And I'm going to say research perlexity.

And then here perplexity is enabled

because we want to research perplexity is enabled

just to make sure that we test that the API key is working.

And I also added this option here steroid mode.

Now this is going to create more of like a power sort of words for the carousel slides.

So let's just enable those two as well.

And then let's go to write 10 slides.

And then this is working.

So Vera is triggered.

Vera is the agent that does the writing of the content.

So he's basically reading the idea and doing the research to make sure that everything is working.

Doing the research to make sure that she got us all the information that we need to write that content.

All right.

And Vera created the content.

So you can see here that we have all the 10 slides generated.

We have the cover text which says your agent is not dump

your packet is bloated

and then it basically writes all the content that would be on every slide that we have.

Now let's say we want to approve this

and we want to promote this into a template.

So let's click on approve and then let's pick a template in the next stage.

Now here we're going to be generating an Instagram carousel.

So we need a 4 is to5 aspect ratio.

Let's say just for this example

I'm going to go with this purple one that we have here.

I'm going to select that

and I'm going to click fit copy to this.

Now what is going to happen is that Vera is going to check

if the content the text that she generated

fits the number of characters that are in that template

because if the words are way more than what the template can fit

then the whole design gets ugly.

So what Vera is going to do is

Vera is going to read all the hardcoded text that are in the template

and try to adapt and fit the content that she generated to that particular template

and Vera basically readapted the content

that she generated to fit the template that we have.

So because the template that we chose has less words in the cover,

she just rewrote the package into agent.

Okay, package fat and then basically wrote all the content that would be on the slide.

Now I'm going to click on render now

to push this into the studio tab

so that we would see how this would look

on the template that we selected.

Now you see a notification open in studio.

Now we can click open in studio here

and this would open the slide

that we just generated in studio.

Now let's click on this

and you can see that just as Barard wrote

agent okay packet fat packet fat bad agent

run starts with too much context

whole ripple noise guesses

and then when you click on this

you can see all the text

that was generated

so assuming maybe we don't want to go with this text here for the cover

we want to just fine-tune

you can click on fine-tune the slide here

and this would basically load the slide here

and when we click on this here just

for an example I can write agents

not the problem packet fat

and then you can also move the slides up and down however you want to have them right

this is all possible in the fine tune tab

so I'm going to close this

I'm going to discard

the changes and then let's say we want to push this to Instagram

and this is the part that makes me nervous

and for that I created a dummy Instagram account called content vibe

so that when we push directly from our this would come

hopefully it would go through buffer and then this would come to this Instagram account

open the package again let me click on and then here by the way you can come here

and you can always change the template whenever you want to pick a

different template you can pick whatever template that you want

and then cadence would basically rewrite the content to fit that template.

Now, let me come here.

Let me click on post now.

And then here, because I've configured only Instagram, it's only going to show Instagram.

I'm going to click on Instagram only.

And it's going to warn me and I'm going to get notification.

Hey, are you sure you want to post this?

I'm going to say yes, post now.

And then this is going to be pushed to buffer.

And hopefully, if the end to end stream works, we should see the post coming directly to Instagram.

And we get notification that this has been pushed to buffer.

Let's click on okay.

And this gets put from the studio tab into the publish tab.

Once this is pushed to buffer.

Now let's go to Instagram and see if we'll see this post.

I'm going to come to my Instagram and hey you see this it posted directly to Instagram.

So you can see this exact post that we had and generated in our cadence

which is the first post that I've tested with this account.

And then you can see that all the text came just as we rendered them on the cadence website.

And then here as well we have the text that was generated also by the AI for the description section.

So this is working very well.

Now when I come back to the studio tab

and I go to the published

I should see

oh so this is not working.

So it renders that we have one all

but I do not see the slide that was generated.

So I'm telling kite

hey the publish tab is not displaying any published content.

I can see a count of one

which indicates that the item exists

but the page itself is completely empty.

So that kite should fix this for us

and hopefully after that we should see the content popping up here.

Now two more things that we're going to do.

I'm going to be show you how that tab would work

like on the configuration tab.

You would have the opportunity to be able to always edit your brand name,

edit your handle,

write a bit about what content you are creating

so that the agents would always read this to find ideas

and to generate your content for you.

And then you can also select the number of generation defaults that you want to have.

And for that, we're going to give Cadence this prompt here to wire the configuration tab.

But I'm going to wait till he's done with the back that we have on the publish tab.

And after that, we can go into the configuration tab and wire this in as well.

And then also whilst we are here we also do have the calendar tab

where whenever you schedule a post maybe you don't want to publish immediately

you want to schedule this as a particular date

when you come into your calendar you would also see the number

of posts that you have scheduled.

So here for instance you can see that the post that we posted today 24th of July

this is already showing up in the calendar tab.

All right so kite is done with that.

Let's reload the dashboard.

Then let's come to the publish tab.

Perfect.

And now you see that the post that was published is already here.

Now the next two things that we're going to work on like I said

is a configuration tab where you can basically alter the configurations here

and the agent would pick the new updates that you have

and for that I'm going to send kite to this

prompt to wire in the configuration tab.

All right.

So that means that when we come to the configuration put up,

let me reload this.

You can basically change your brand name,

the handle,

the number of slides that you want to default with,

if you want to have perlexity by default.

For instance, when I toggle this off and I save the configuration and I'm generating an idea,

it wouldn't use publicity toggle on as a default.

And after that,

we're basically done with the main aspect of the build.

I do have some other prompts that I'm giving to you as optional things that you can add to your setup to make it even more powerful.

Now what we're going to build now

is going to be the telegram remote control.

So if you remember we do have this studio tab that we created

which ideally we want to also use to be able to generate ideas

without us having to come to the dashboard.

Now the idea here is that whenever you want to create an idea

you can just come to the studio tab here in telegram

generate an idea promoted and stuff like that.

So for that we're going to give kite as prompt here

to set up the telegram remote control.

All right so kite says he's done

he built and verified the telegram remote so far as he could.

He also said that he did a smoke test.

So let's come to the studio tab

and then we can see that he did a test

where we can see the things that he triggered on the studio tab.

Now he did mention something that he couldn't run.

Let me scroll down to the bottom.

He said that he's he has a blocker.

So he wasn't able to restart the hermits gateway himself.

So he's telling us from a separate shell on the machine

we should run this

and then we should just tell him that it's restarted.

So let me copy this.

Let me come to PowerShell.

Let me clear the screen here.

Then let me paste in this.

Let me connect to the server again.

Then let me run the command he gave to us.

All right.

So this gateway has restarted.

So, I'm going to wait a bit for the gateway to completely boot.

And after that, I'm going to send a message to Kite restarted.

So, I'm going to now send Kite a message restarted.

And then hopefully he should be able to continue with what he was doing.

All right.

And he says that he's done.

So, basically, everything is wired in.

Now, let's go to the studio tab

and let's write something like let's forward slash let's run idea.

Now, let me check what the slash command was.

I don't remember if it was ideas or if it was idea.

So, it's ideas.

So, let's send forward slash ideas.

And then hopefully this should trigger the ideas tab on the studio right

in this work in this workown.

So we have fresh ideas stop asking one agent to do the whole job.

So when I come to the studio tab I also see the same thing.

So basically this is mirroring whatever that we have on our dashboard.

Now the reason why we are doing this is that

I want you to be able to build ideas and stuff like that

without you having to always come to the dashboard.

Now let's say we want to promote this idea/promote

and then let's put in the number of that idea 11.

Oops, I made a mistake.

I intended to type in 11.

And this was exactly correct.

He says that idea one was not found.