n8n Tutorial – Zero to Hero Course

Master the future of process automation.

Nin is an incredibly powerful open-

source platform that enables you to

integrate APIs and orchestrate

intelligent workflows without the usual

coding headaches. This course from

Maronei will guide you from the

foundational concepts of nodes and

architecture to deploying advanced real

world systems. You'll master essential

skills like connecting various services,

configuring API keys, and handling

complex data flows. Also, the course

goes into cutting edge AI integration,

teaching you how to build advanced

retrieval automated generation or rag

agents and coordinate multi-agent

systems. By the end, you'll be able to

automate sophisticated business

processes, giving you a competitive edge

in DevOps, AI, and data engineering.

Automation is no longer just a buzz

word. It's reshaping how teams work,

collaborate, and scale. At the heart of

this revolution is Nen, an open-source

extensible automation platform that lets

you connect APIs, orchestrate workflows,

and even build intelligent AI agents

with ease. From sending simple automated

emails to creating advanced multi-agent

rack system, NAN empowers you to

streamline operations, boost

productivity, and unlock whole new

levels of innovation. Welcome to NATE

zero to hero course by Cookcloud. I'm

Maronei and together we'll go from

beginner to advanc as we explore how to

build powerful automations with NIET.

Whether you're a DevOps engineer, an AI

enthusiast or someone from a

nontechnical background looking to

automate real world processes, this

course is designed for you. We begin

with the introduction section where you

get familiar with Naden itself, the

playground environment and the code

structure. You'll also understand the

objectives so you know exactly how each

step builds towards your automation

skill set. Next in the foundations of

edit end we'll cover the essentials

understanding notes inputs and outputs

and we'll talk about data types in editn

and finally how wful run under the hood

using an endn's default logic. You'll

also configure API keys for services

like OpenAI, Anthropic, and CodeLab

Keyspaces so you're ready to integrate

AI into your workflows right from the

beginning. From there, we jump into

hands-on AI agent workflows. This is

where the fun begins. You'll build your

first email AI agent to drop responses

on autopilot. Create a multi-agent

research workflow that pulls knowledge

from tools like proplexity and open AAI,

turning hours of work into minutes.

Explore the HTTP request node. your key

to working with APIs from web scraping,

data fetching or calling other external

agentic tools. Experiment with creative

workflows from text to image, text to

video, and even image to video

automations using bleeding edge models

like Google's V3 and Cand. And yes,

we'll even build a Slack workflow that

lets AI reply to your co-workers and

your managers on your behalf. So while

you're sipping coffee, Edit end is

already answering questions for you, aka

doing your job. In the optional setup

section, we'll explore how we can

self-host nit with Docker and use Ola as

your local LLM, as well as hosting on

AWS EC2 in our playground environments.

This flexibility means you'll know how

to run NAD in a way that matches your

scale, budget, and preferences. Then

we'll dive into rag agents retrieval

augmented generation. You'll learn how

vector databases like pine cone gives

workflows memories and contexts.

Together, we'll build a customer sport

rag agent, the kind of workflow that

real businesses use to provide

intelligent contextaware support. We'll

also explore MCPs and see how they

compare to traditional workflows,

unlocking new possibilities for

reusability and scalability. In the

multi-wordflow advanced build, we'll

combine multiple agents into enterprise

style system, showcasing how subflows

let you manage complexity without chaos.

Finally, we'll cover retries, error

handling, best practices, and how to

leverage the editor workflow template

marketplace to accelerate your builds.

By the end of this course, you'll have

learned how to build a practical real

world automations like Slack AI agents,

multi-agent research workflows,

multimodal automations, customer support

rag agents, and advanced multiworkflow

orchestration systems. And along the

way, you'll engage with hands-on labs

designed to reinforce learning by doing

so. So, every concept becomes second

nature.

At Cook Cloud, we believe in learning by

building. And you'll be part of a

vibrant community where you can share

insights, ask questions, and grow

together.

So, ready to move from zero to hero with

Nin. Let's dive in and unlock the full

potential of workflow automation.

Hey there and welcome. In this section,

we're going to walk through the

foundations of NIDN, an open source

workflow automation tool that makes

connecting apps and services simple.

Here's what we'll cover. We're going to

do a quick introduction of NIDN and its

building blocks nodes. And then we're

going to explore AI agent architecture

and API flow. And later on, we'll

showcase how it works in an email to

select use case. I'll also show you how

input formats and expressions work in

NIDN. And finally, we're going to talk

about how Nadn handles data types. Let's

dive in. All right. So, what is NN?

Well, NN is short for notation, and it's

a free open-source workflow automation

tool. Think of it like a universal

connector. You can link different apps,

services, and even AI agents together

without writing tons of custom code. The

magic happens through something called

nodes. You set up workflows visually

step by step and edit end handles all

the data passing in between. In add

workflow with a trigger event that you

define like a new email arriving or a

web being called and then there's the

action nodes that do the work like send

an email, interact with an LLM or call

an API. By chaining these together, you

can turn everyday manual tasks into

smooth automated flows. Now, let's look

at AI agents inside Nitn. Just like any

other node, an AI agent within a

workflow will be triggered by an event

or a trigger node. And there are three

important elements that make this node

an AI agent. Namely, the LLM, which is

the brain that powers the agent. And

this could be open AI, plot or llama, or

any LLMs that you hook up to it. And of

course, the way you connect them would

be through an API access token. And then

there's a context window memory which

gives the agent context of the workflow

interaction and the tool that you can

use to complete tasks like Gmail, web

scrapers, or any tools that you connect

to it. And of course, you connect that

node to an output into either the next

node or just as a chat response. Here's

how it works. Your workflow grabs some

user input, say a simple chat trigger.

The agent thinks about that input and

based on what you've asked, whether it's

sending an email or just having a quick

chat, it decides if it needs to call

tools like Gmail or any other tools that

you hook into it through the APIs. Then

it pass the output back to your

workflow. So it's input going into the

agent, deciding whether to use the tools

or not, and then passing out the output.

And that's how you get intelligence

mixed into your automations. And here's

how that looks inside N itself. A chat

trigger kicks things off. An AI engine

agent node hooked up to GPT does the

thinking. A Gmail node can send messages

and an output node fires the result

straight to Slack. It's the same logic

but now mapped visually into the

workflow editor. Now by default addin

run nodes sequentially.

So it's trigger to the next node to the

next note and the next node. If the

nodes are connected in a line,

everything runs in order until the

workflow finishes without skipping any

nodes. Which means if along the line one

of the nodes fail, it will cause the

workflow to error out. In this case,

you'll see that there's a Tavly tool

attached to the AI agent. And because it

is only a tool hooked up to the agent,

it is optional for the workflow to run

this. And it depends on whether the AI

agent decides that it is appropriate to

call on this tool based on the

instructions received. And in the

middle, you'll see that there's a if

conditional loop. And this just ensures

that the workflow doesn't error out when

it doesn't get the result in time from

external APIs. But we'll get to the

details of that in the next few

sections. But workflows don't always go

in a straight line. You can brush them

out. So one input might trigger two

parts. Maybe one branch post to Slack

and the other goes to Gmail. In this

case, each branch runs independently.

It'll run the top one first and then the

second one sequentially, giving you two

actions from the same trigger. AI agents

need memory. And there are two main

flavors. We have the context memory,

which is a short-term memory inside the

AI's prompt. It's great for chat history

and to get the AI to stay in context.

And it is a difference between a

stateless and a stateful response from

the AI agent. And we'll run some

examples to show you the difference in

practice later on. And then there's a

vector database or documents rack

memory, which is long-term searchable

memory. Here your documents get stored

as embeddings and the agent retrieves

only what's relevant when answering.

This is a knowledge source outside of

LLM's training data. So simple memory is

leveraged for context and continuity to

make sure that the AI agent stays

context and rag is used for scale and

accuracy.

Every node in addit has an input and an

output. In the input panel, you see the

payload data that the node is receiving

from the previous node. So in this case,

it is a chat message received from the

user because the previous node was a

chat trigger node and the output is

basically the response of the LLMs or

the AI agent and the data structure

ranges from very simple with one

variable or multivariable with different

ranges of data types and additional

informations. But it is worth noting

that every node has an input and an

output data and the output data of this

node will be the input payload of the

next node and that's how the logic

strings the entire workflow together. So

this makes it super easy to see what's

flowing through your workflow and data

doesn't always look the same. So in

ended end there are three main types of

data that you'll see in the input and

output panel. So the first one is schema

which is a defined data structure and

the second one is a table format with

rows and columns handy for spreadsheets

and then of course there's that

underlying JSON which is the most common

structured and flexible way to show the

data and sometimes you get other types

like binary for files and images. It's

worth noting that these are different

representations of the same data. When

you configure a node, you can type in a

fixed value like a static string or you

can use expression which pulls dynamic

data from earlier nodes. Expressions

allow your workflow to adapt to

real-time inputs instead of being

hardcoded in. And the cool thing about

N&N is the fact that you can drag and

drop the variables or parameters into

the fields within their low code UI. Now

let's talk data types. The basic

building blocks inside NN. So in Naden

there are many five different data types

which is strings, numbers, boolean,

arrays and objects.

So strings are like text for example

hello world. Numbers as it suggests are

values like 42, 3.14 or -100. Booleans

are simply true or false statements and

arrays are essentially lists like a list

of numbers or a list of names. And

objects are structured data like what

you see in the example which contains

the user credentials including name and

email. Objects can even nest and you

often access them with dot paths and

expressions and knowing the differences

and mastering these makes everything

else click. One of the coolest things

about NN is the community nodes. These

are extra connectors built by users

extending NN way beyond the official

nodes. So, if you don't see a built-in

integration for your favorite tool,

check the community first. Chances are

someone's already made it.

Once you're in the workflow, the very

first step that you want to do is to

introduce a trigger node into the

workflow. The trigger node, as it name

suggests, is a node that's going to

trigger your entire workflow when the

event that is dependent on it is

triggered. So let's go ahead and hit the

plus button on the top right hand side

here and it will generate a list of

options or possible trigger nodes that

you can start your workflow with. Now

the most simplest one is the trigger

manually node which triggers every time

you click on it. But in this case what

we're going to choose is an on chat

message trigger node.

And once you click on a note there

usually are going to be some fields that

you need to configure. But in this case,

because it's a simple chat message

received node,

what happens is that a terminal will pop

up here where you can actually interact

with the chat node. In this case, I'm

just going to type in hello to populate

it with the relevant data.

Now, a quick way for you to

differentiate between a trigger node and

a regular node is by the lightning icon

here on the left that you see.

Now let's open up a chat message receive

trigger node to see how the data is

populated within the note.

As this is a trigger note, you only have

an output data from the node and you do

not have the input data.

Now if we covered in the previous

sections, there are three types of

representation of the same data that

you'll be able to see within each nodes.

In this case, the underlying data format

is JSON with three different information

that's contained within the payload. Now

the first information is the session ID

which identifies a session of the chat

and then we have the action message

which in this case is send message

because we're sending message through

the chat input and there's a chat input

content which in this case was what I

typed in which is hello.

Now when you toggle over to the table

section what you'll see is the same data

representation in a table format. As you

can see this session ID, the send

message action at the chat input which

is hello.

Now if you toggle over to schema, you

can see the same representation of the

data in a schema format which makes it

easy for you to drag and drop to the

next node when you need to.

Now let's go back to the workflow. Once

that first node has been set, the second

node that you want to hit is the AI

agent node. Now in this case I'm going

to click AI. I'm going to select AI

agent.

Now when you open up a node the data

from the output of the previous node

automatically populates the input of the

AI agent node. And this is true for any

types of nodes when you connect it to

the previous nodes. The output payload

of the previous node is going to be the

input of the current node.

Now in this case because the AI agent is

already connected to the chat trigger we

can leave this as a connected chat

trigger node

in terms of the prompt or the user

message it is correctly identifying as

JSON chat input which is the content

here which is hello and you can see that

hello is represented here as the actual

value of the variables.

Now with every AI agent node, what you

need to pay attention to are three key

elements that make this node an AI

agent. So the first is of course the

LLM. So LLMs are the lash language

models basically is the brain behind the

AI agent note. Now in this case there

are long list of options of LLM that you

can choose from and each of these are

good for specific use cases. In this

case, we're going to choose an OpenAI

chat model as the model that we want to

hit.

Now, I already have an OpenAI account

set up here, but I'll show you how to

connect your OpenAI account or your

OpenAI API keys pretty easily. In this

case, just hit the dropown and click on

create new credential.

And in this field, what you want to do

is to populate it with the API key that

you can either obtain from CodeCloud

Keyspace or the OpenAI developer

platform. We'll cover how you can get

the OpenAI API key from CodeCloud

Keyspace or the OpenAI developer

platform in another section. In this

case, you can also fill in your

organizational ID or it could be

optional.

Now, once you've key in your API key,

you can hit save.

And once you've connected your OpenAI

account, what you can do now is to pick

from a list of GPT models that OpenAI

has to offer. In this case, we're going

to pick GPT for Mini because it's

intelligent enough for most use cases.

Now, moving on, the second element that

we don't want to attach to the AI Asian

node, it's a memory node.

And what does memory mean for the AI

agent? Well, this is basically the

persistent memory that tells the AI

agent or gives the AI agent the context

that it requires from previous

conversation.

And I'm going to show you very quickly

what happens when an AI agent has no

memory versus when it does. Now, in this

case, we're going to type in hello,

my name is Mark.

And when I type that in, the agent now

access this open AI LLM chat model to

infer the information and respond to me

by saying, "Hello, Mark. How can I

assist you today?" Okay. And in the next

thread, what I'm going to ask is, "What

is my name?" So, since I've just told

him that my name is Mark, it should be

able to tell me what my name is.

But as you can see in the output, and

let me just move this up. It says, "I'm

sorry, but I don't have access to

personal information about users unless

you shared it with me in this

conversation. How can I assist you

today?" So, as you can see, it has no

memory context of what I just said,

which is my name is Mark, even though I

just typed it in a minute ago.

So, now let's attach a memory bank. In

this case, we're going to choose a

simple memory. And as you might have

noticed, there are various options of

different types of memories that you can

attach to the AI agent. But the simplest

way uh to get started is to click on

simple memory and choose that as the

option.

And there are a couple fields that come

with the simple memory as well. But in

this case, because our connected trigger

node is a chat trigger node, we can

leave this as that.

And the context window length basically

means how many past interactions to

model that the memory will retain. So in

this case, we're going to keep it as

five and then it will remember five

different iterations or five previous

iterations of the conversation that you

have with the agent. So we're going to

leave it as that.

Now you might notice that now the AI

agent node is showing a yellow highlight

around it with a triangular icon and

that means some changes has been made to

the AI agent and in this case the change

was the memory node being attached to

the AI engine and showing that so that

you can run the workflow again so that

the data can be repopulated.

In this case I will do the same thing. I

will say hello my name is Mark.

As you can see it's doing the same

operation but this time it's actually

dipping into the simple memory and now

are able to store that information.

So I'm going to ask it again what is my

name?

And this time as you can see it says

your name is Mark. How can I help you

today Mark? So now it retains the memory

of the information that I've just given

it a minute ago.

And you might notice the terminal in the

middle here are basically the logs of

what's going on within the node. So in

this case, what happens was a chat

message was received. The data was

passed on to the AI agent and the AI

agent is zipping into the simple memory

to check what my name is or the

information that was given earlier and

then using the open AI chat model to

return the information.

Now that the chat model and the simple

memory has been set, the third thing

that we want to attach the AI agent to

or provide the AI agent with is the

email tool. So now there are various

tools that you can attach. But in this

case, because we want the AI agent to be

an email agent, we want to attach the

Gmail note.

when you're attaching a Gmail note, you

need to set up your Gmail credentials to

be associated to the account. I already

have my Gmail account set up here.

But if you haven't set up your Gmail

account before, it is pretty easy. The

way you would do it is to create a new

credential. And if you're using an

cloud, you would sign in with Google as

an ooth to recommended way.

or if you're using a self-posting

method, you might want to choose a

service account and fill in the

necessary information that's required to

set up the Gmail account properly.

Now, as you can see, there are a couple

fields to configure here.

The first one is tool description. We're

going to leave this as set automatically

because we're going to let the AI model

determine whether they want to use Gmail

or not.

The second field is a resource. In this

case, we want to do message because we

want to be able to send message through

the Gmail tool. And the operation of

course is send. And so you can see there

are other options as well. So when we

attach a Gmail tool oftent times it

doesn't mean that we just want to send

an email. It can also do many other

operations such as reply a specific type

of email or mark an email as read or

unread or delete an email or many other

options. Now going into the information

of the email itself, there are a couple

of information that we need to pass to

the AI agent for it to determine what to

do with the email. So first of course

who is the recipient of the email. Now

because this is going to change

depending on your input. Now remember

because your input is going to be in

natural language whether that's English

or any other languages this is going to

be different every time. So what you

want to do in that case is to let the

model define what the email address is.

That way whatever your input is, the

model is going to infer the information

and try to extract the necessary

information that you think is correct

for the email address

and the subject itself is also going to

be determined by the model. We're going

to let it do that and we can do that by

clicking the same blue button here to

let it determine the subject.

When it comes to email type, in this

case, we want to choose a text email

just for simplicity. HTML is very good

when it comes to imagerich type of

email. So, if you're looking to attach

any image or any thumbnail and you want

it to be aesthetically formatted within

the email, then you would choose HTML.

But in this case, we're going to go

ahead with text.

And then on message, we want to choose

determine by model as well because that

way we're going to let the AI determine

what kind of body of the message that is

going to attach.

Okay. So now the three elements have

been configured.

We want to go back to the AI agent node

to specify certain prompts. But before

we do that, there are a couple of things

that I think it's worth highlighting. As

you probably have noticed, with every

single field that comes within a

particular node, there is a fix and

expression toggle here. Now, I just want

to highlight the difference between fix

and expression. So, because the AI agent

node is connected to the chat trigger

node, we've left this as connected to

chat trigger node. But what you can do

is actually to define what the input is.

So, when we click on define below, we

specify exactly what kind of input that

goes into the prompt or user message

prompt.

So in this case I can drag and drop the

chat input and it will fill it up as an

expression.

Now the difference between fix and

expression a fixed format is basically a

natural language format with which you

can tell the AI agent a specific but

permanent definition of what the user

message is. So for example, if I were to

put the user message as please send an

email, this is going to be the input for

every single run that I execute of this

workflow. And that's not ideal because

we want the AI agent to be able to get

the information dynamically depending on

my input. So in the first run, I might

put what is my name instead of please

send an email. So we want the prompt or

the user message to change together with

my prompt. So how do we do that? That's

when we toggle into expression. And what

happens is we can now drag the chat

input variable in this case and simply

drop them into the prompt user message.

And what happens is as you can see it

creates an expression in this case a

JavaScript expression of the variables

of the chat input. Meanwhile the actual

value of the variable is dictated here

which is what is my name question mark.

So, as you can see

on each separate run of the workflow,

it's going to populate the information

based on my input instead of a prefix

input if I were to choose a fixed

format. I hope that explains the

difference between fix and expression.

Well, and in this case, the next thing I

want to do is to go to options and under

add option,

I'll go to the dropdown and click on

system message.

Now for those of you who are not sure

the difference between a system message

and a user message, the user message is

a set of instructions or input that

comes directly from the user.

Meanwhile, the system message describes

the core function of this particular

model. So in this case, what I want to

tell it is that you are a helpful email

assistant which helps craft effective

and succinct email based on users

instruction. You also help with sending

the email by using attach email tool

when asked. Now this is a very simple

system prompt and I would suggest that

for a more effective system promp that

you go to chatgpt or llm of your choice

to prompt engineer an effective system

prompt for a particular AI agent. That's

what I would do in a production

scenario. But in this case, because

we're just running a quick demo, we're

going to stick with a simple one just so

that you understand the principle behind

it. Cool. Now that we have the system

prompt defined as well as the user prom

variable in place, we can now take it

for a spin. Now in this case, what I

want I want to do is to say the same

thing, which is hello, my name is Mark.

And as you can see, it's accessing both

the open AI chat model and a simple

memory to store the information, but

it's not necessarily using the Gmail

tool. And the reason for that is because

I did not specifically ask the agent to

send any email just yet. So as you can

see from the response, hello again Mark,

what would you like to do today? Is the

response which is also the same response

here. Now I would say I would like to

send

an email

to my boss

at

maronei@codecloud.com

which is not my real boss's email

address is actually my email address

because I don't want to necessarily send

my boss a test email about

the upcoming marketing meeting

on

26th of August

2025.

So, I'll hit that.

As you can see, it's not necessarily

using the Gmail tool straight away

because it is intelligent enough to know

that it needs a little bit more

information. So, it says, "Sure, what

would you like the subject and the

message of the email to be?" So in this

case, because I'm too lazy to come up

with the subject and email, please,

please come up with it yourself. Okay.

So I'm going to say that

and it gives you a draft of the email.

So in this case, it says subject

upcoming marketing meeting on August 26,

2025. Dear boss's name, I hope this

message finds you well. I wanted to

remind you about the upcoming marketing

meeting scheduled for the 26th of August

2025. Let me just blow this up. We will

be discussing our strategies and plans

for the upcoming quarter. Please let me

know if there are any specific topics

you would like to address during the

meeting. Looking forward to your input,

Mark. And I would like to make changes

because it says, as you've noticed, my

boss's name. So, let's say

my boss's name is Moonshot.

Okay, so now it registers that

and gives me a new draft. Dear Moonshot,

I hope this message finds you well about

the upcoming marketing meeting. Da da

da.

And then it says, shall I go ahead and

send this email to

moonshotaronecloud.com?

Again, this is not his email. This is my

email because I don't actually want this

to be the email that sent to him. So I

would say yes, please go ahead and send

it.

So once I've confirmed that, what

happens is, as you can see, the AI agent

is now accessing the Gmail tool to send

the email to actually in this case my

own email. So let's take a look what it

looks like in my inbox.

Now this is the email that's been sent

to me. As you can see, because the Gmail

account over here

is connected to my email or my Gmail,

which is marone@codecloud.com,

it is essentially sending from

marone@codpow.com

to me. That is this in the content of

course. Dear Moonshot, I hope this

message finds you well. Wanted to remind

you of the upcoming marketing meeting

schedule for 26th of August 2025.

So, everything looks nice. I can almost

send this directly to Moonshot. But what

I want to do is as you can see at the

end of it, it says this email was sent

automatically by NAND. So that's not

ideal because Moonshot will find out

that I'm actually automating this email.

So what I want to do is go back to the

NAND workflow and go to the Gmail tool

and scroll down to options. Under add

options, I'm going to click on the

dropown

and click on append edit and

attribution.

And what I want to do now is to toggle

this off. In this case, it will actually

stop attaching the message of the end

attribution in the email.

So in this case, what you can do when

you want to execute a step is to hit the

play button on top of the note so that

you don't have to rerun the entire

workflow. So let's just click on that.

Okay. So now that that's done, I want to

take a minute to explain some of the

little functions here that you see and

what they do.

The first one is the play button which

is the execute step button. So this

button is super useful when you want to

run the node without essentially running

the entire workflow again. So if you

change anything within the AI agent node

and you only want to run the AI agent

node again, you would hit this button.

It will only run the node here.

The power button as you can see here is

to deactivate the node because this is

the only node basically that does any

processing in this workflow. But as your

workflow grows more complex, you might

have five or six branches of nodes. And

some of them, if they're not in use, you

might want to deactivate. So if you

click on deactivate, what happens is

it's going to say that the node has been

deactivated. And basically, the workflow

is not going to work for any processes

that go through this particular node.

And we're going to reactivate that

again. And of course, the bin icon is to

delete the entire node together.

Now another very useful function in

within any is a pin. Now if you were to

open up the node and go to the top right

hand corner you can see that there's an

option to pin the data. Now this is

super useful because now you don't have

to key in or populate this particular

node with new data all the time.

And this is extremely valuable when

you're running a workflow that uses API

tokens or might cost you for every run

of the workflow.

So to avoid running the workflow again

and again and using or activating the

API token, what you can do instead is to

pin the data to the node so that every

single demo run or every single trial

run is using the data that's already

been populated.

So once I pin this data, every single

run of the chat input is only going to

say yes until I push it to production.

Now once you've done all this, this

doesn't mean that your workflow has been

pushed to production. So the way you

would push to production is to toggle

the activate workflow option right at

the top here.

And when you do that, it's going to give

you a warning message to say you can now

make calls to your production chat URL.

Okay. So click on got it. And there you

go. Your workflow is now in production.

You do not have to click on execute

workflow to run the workflow. And every

single message that it received is going

to run the way you set up the workflow.

So just to try that out, I'm going to

use hello.

I'm going to have to pick on pin and

send.

And there you go.

It's a new response. Now, of course,

this workflow isn't very useful because

the only way you can interact with the

AI agent is through the terminal here.

Now, how can we provide a publicly

available terminal so that more than one

users or people who don't have access to

your end to end workflow can also use

this AI agent? Well, if you open up your

chat note, what you can see is that

there is a make chat publicly available

option. And what you want to do is just

to toggle that. In this case, it's going

to give you a URL

that you can copy and open up.

And you might have to wait for a couple

seconds for it to stand up, but once you

do, this is a page that is publicly

available as long as you pick the option

to make the chat publicly available

and you can access the AI agent

as any user would. Cool. And as you can

see, there are a couple options here

that you can configure if you want to

include an authentication or if you want

to have an initial message that is

crafted or customized to your needs.

But in this case, we're going to keep it

simple because in this section, we just

want to run through the principles of

how to create your first AI agent

workflow on NN. And I hope that has been

helpful. And just before we go, you

might notice that there are three

options to choose from from here, which

is the editor version where we have been

actually building the NDN workflow on

which obviously as the name suggests

where you edit all the nodes and

configure all the nodes in the workflow,

but also the executions button

where you can check out the logs of the

executions.

As you can see, it gives you the date,

the time, and the type of execution. The

one with the beaker icons are test

executions or basically the workflow

runs that we did when it was in demo

mode. And once I push it to production,

it is an actual execution to the

workflow in production.

And one of the things you can do once

your workflow is in production is also

to copy the data that's been populated

during the execution for troubleshooting

purposes. So if you click on copy to

editor

what happens is your editor will now

show the information that is derived

during the execution of that production

workflow.

So remember in the chat we said hello

and here it is the chat input content is

hello

and similarly the AI agent response is

hello how can I assist you today which

is the response that we got from the

live chat.

Now, this is a very useful option

because it just makes your

troubleshooting as well as your

iteration a lot easier

as you build or modify workflow in

production.

So, I hope you were able to follow

through on the demo of how to build your

first AI agent workflow on Edn. In the

next section, we're going to give you a

lab where you build the exact same AI

agent within that particular lab and see

for yourself if you're able to create an

AI agent that crafts email and sends it

on your behalf. This course comes with

free labs that could be accessed right

in your browser. So, you don't need to

set anything up yourselves or use your

credit cards. The labs are challenge

based, meaning we give you a challenge

and ask you to solve it. Use the link in

the description below to sign up for

free labs on codecloud. I'll now walk

you through your first lab. In the

upcoming labs, you're going to

experience hands-on building AI agents

on NAD environments. And in doing so,

you're going to need an API access to

the LLMs of your choice. And in order to

do that, you need the API keys. And

there are a couple ways to get your API

keys. Naturally, you can go to OpenAI to

get your OpenAI API keys or platforms

like Open Routers on which you can

obtain API keys of several different

models. But those platforms require

payment information before you're able

to get the API keys even if you're just

trying out. So, one of the things that

we've done here at CodeCloud is we've

created the CodeCloud code key. And what

it is, it is the all-in-one AI

playground where you can get the API

keys to the following LLMs of your

choice. And at the time of this

recording, models like GPT40, GPT4.1,

cross on 4, Gemini 2.5, and Grog 3 are

available through the code key

playgrounds. And the code key playground

is available on various different

subscription plans, but you get the most

access with AI plans and the business

plan. Now, if you were to go to this

page, which is

codecloud.com/ai-playground/code

key with the right access, you can hit

launch now and click on start playground

and you'll be led to a dashboard where

you can essentially create and grab your

API keys by selecting the models as well

as the different types of ways or

methods with which you can access the

LLM model. Now, in this case, I'm going

to do an example of how to call a GPT

4.1 model through a code key on your NDN

environment. So, there are two ways to

do that. The most straightforward way is

to go to your end workflow. I'm going to

start the trigger node with a chat

trigger because I want to be able to

communicate with the agent and I'm going

to just populate the node with something

like hello. And we're going to introduce

a second node which is the OpenAI node.

In this case, we're going to choose a

message and model node. And what we want

to do here is to create a new

credential. And we call this code key

demo for example.

And what you want to do under the base

URL, replace this base URL with the base

URL that you can obtain from the code

key page. Copy that. Go back to the

workflow and paste the base URL in. And

of course for the API key, we're going

to take the API key here and we're just

going to paste it in and we're going to

hit save. And you'll see that sometimes

depending on the version of the NA cloud

that you're in, this red bar might pop

up. Don't worry too much about it

because I'm going to show you how you

can keep using it even if the red bar

shows up. And this is just simply

because the UI itself is not designed

for users to patch in their base URL

themselves like that. And sometimes that

causes some form of incompatibility

which causes the red bar to show up. In

this case, under model, what we want to

do is we're just going to define it by

ID because we're going to type in the

name of the model ourselves because this

is not going to be the models that are

in the list from OpenAI. And if you were

to try to grab it from the list, it's

not going to load at all. So, we're

going to hit buy ID and we're going to

manually type in open AI/GPT-4.1.

And that's the model that we want to

call here. As you can see, is GBT-4.1.

And with that model, we're going to

transfer or rather drop in the chat

input variable. So, that essentially

everything is connected. And we're not

going to add any tool at this point

because we're just going to hit execute

step. And as you can see it is coming up

with an output or response to the chat

input which is hello. It says hello how

can I help you today. So with that

simple setup you can now chat with the

model here. For example I can ask why is

the color of the sea blue and as you can

see under the chat the output is here in

the response the sea appears blue

primarily because of the way water

absorbs and scatters sunlight. Um, and

that's the response for that. Now,

that's one way to gain access to a LLM

model through code key. The other way to

do that is to call a HTTP node. So, I'm

just going to add another trigger node

here. And this time, I'm going to select

trigger manually. I'm going to move this

down right here. I'm going to add the

HTTP request node. And if you were to go

back to the code key documentation, as

you can see, we can actually toggle this

to the curl command and I'm going to

copy the entire chunk here of the curl

command and go back to my workflow. I'm

going to hit import curl. I'm going to

paste the entire thing and it's a post

API call to this particular endpoint

URL. And of course there's some

authentication here which is already

prefilled actually because I pasted the

entire curl command. Ideally we want to

set up the authentication on the headers

but we'll go over that part later on in

the next section. But in this case

without doing too much what I want to do

is I want to just hit execute step and

there you go. So under JSON body you can

see that there is a preconfigured

content which is in a single sentence.

Why should someone choose codecloud over

other platforms to learn DevOps? That is

the user prompt that has been

prepopulated there. So you can

essentially replace this with some other

variables that you receive from the

input. Like for example, if it's a chat

trigger that's connected, you can drag

those chat trigger as the user input and

drop it in here to replace this whole

line. But here I just want to show you

that it is responding to the user prom

as someone should choose code cloud over

other platforms to learn devops because

it offers highly interactive hands-on

labs and real world scenario based

exercises that accelerate practical

skills development far beyond

traditional video based courses.

So cool. These are the two ways that you

can gain access to the LLM API keys via

KOI. And as I've shown earlier, you can

choose from these various models right

now. For the next few sections, this

might come in handy for you. Otherwise,

you can always go to the actual provider

like OpenAI or Anthropic to get those

API keys. I'll see you in the next one.

So I want to run through the differences

between running your instance on end

cloud versus the lab playground that we

have on the codecloud course. Now the

first thing you'll notice as you go into

the lab is that you still have to

provide your email and your first name

last name credentials as well as a

password in the instance. And don't

worry this is not saved anywhere. So you

don't actually have to save the

credential and the password somewhere.

You can use different emails and

password for each of these instances. So

once that's filled in you can hit next.

And there's going to be a series of

onboarding questions here which you

don't need to fill. So you can hit get

started and in the same way for the paid

features information you can just skip

and there you go. Now you're in the

admin dashboard that is very similar to

the end cloud environment. However,

there's still some nuances and some

differences as we go into the workflow.

So here what we're going to do is we're

going to click start from scratch. And

the very first workflow that you're

going to build is the email AI agent.

And the first trigger to that is a chat

trigger. And I'm going to speed through

this cuz this is very similar to the

workflow that we've done. But I want to

just show you quickly the differences of

using the KCloud keyspace as well as the

OpenAI API key. So if you're using the

KCloud keyspace, what you want to do

when you select your model, for example,

in this case, I'm just going to select

the OpenAI chat model. And what I want

to do is to follow the instruction in

the left hand bar right here. And you

see that there's a link to Code Key. I'm

going to just hit that URL. I'm going to

hit launch now. and it's going to lead

me to this dashboard right here. I'm

going to click start playground and

there we go. So, this is the dashboard

on CodeCloud Keyspace. And what I want

to pick is the OpenAI GPT 4.1. And in

this case, I'm going to just copy the

API key here. I'm going to go back to my

workflow. And here, I'm going to hit

create credential. And I'm going to

paste the same exact API key. I'm going

to skip the organization ID. And for the

base URL, I want to make sure that I'm

replacing this with the base URL that I

obtained from the KCL keyspace. Go back

to my lab and paste the base URL right

here. Okay, so I'm going to hit save

right now. As you can see, it says

connection tested successfully. However,

there's a couple of things I want to

point out here cuz if you pick from

list, as you can see, the list doesn't

really match the known models of GPT.

This is because it's not really working

based on the UI that's been built. So if

you try to run it based on the chat ID

or chat model that we've selected, I'm

just going to run a hello message here.

It's going to go to the AI agent, but

it's going to error out. So what you

need to do is go into the chat model and

instead of picking from list, you want

to go by ID and as suggested from the

instruction on the left hand bar here,

you want to copy the OpenAI/GPT-4.1.

Just copy that and paste it all word for

word. And let's do a test run again. And

this time it should actually be able to

access the appropriate model. So that's

if you're using the CCloud keyspace API

keys. Now what I want to point out is

the difference between this and using

the OpenAI API key is that it's much

more straightforward when you use OpenAI

API keys. So to show you the difference,

what I'm going to do here is I'm going

to create another new credential and I'm

going to call it OpenAI account 2. This

time I'm going to head over to

platform.opai.com.

And what I want to do is head over to

API key section. I'm going to create a

new secret key in and then email

integration. I'm going to hit create

secret key and I'm going to copy the API

key and head back to my lab. I'm going

to paste the API key and leave the base

URL as is and I'm going to hit save. So

there you go. So connection tested

successfully. And the difference is

instead of by ID, I can now just pick

from the list and it should load up the

correct list of models that I might

possibly want to use. So for example, if

I choose GPT 4.1, it's just going to be

that. And we're going to run this node

again. And as you can see, it's calling

the correct model right here. Okay. So

the next thing I want to point out is

when you add your Gmail note on the next

workflow or Google tool for that matter

the difference between doing that in our

labs versus edit cloud is that on edit

and cloud you often see when you create

a new credential with Gmail that you can

actually have a button which you can

sign in directly using your Google

account if your browser happens to be a

Google Chrome browser. However, with the

labs what you need to do when you create

a new credential is you actually need to

connect it with the o method. So what

you want to do is you want to head over

to console.cloud cloud.google.com.

And the first thing you want to do is

create a new project. And for the new

project, I'm going to name it Enit in

email app. All right. So, I'm going to

leave this as no organization. I'm going

to hit create. And it's going to take a

couple seconds to create a project. And

once that's done, I'm going to select

the project. And as you can see, it says

end email app project. And the very

first thing I want to activate is I want

to go to Gmail API.

So what I'm doing right now is I'm

creating a project because that's how

Google recognizes each of these ooth

access that we're giving it. But the way

the security works is you need to enable

the particular tool that you want to use

within the project. So in this case I

want to use Gmail. So I want to make

sure I enable the Gmail API. All right.

Once that's enabled, I want to head over

to OOTH consent screen. And right now

there's no OOTH consent that's set up.

So I'm going to just hit get started.

And in this case, I'm going to have to

call it an app name as well. So I'm just

going to say it is an email

app. Okay. User support email. Going to

put this.

And we're going to select external. And

by the way, each of these steps is

documented on the left hand side panel

of the lab. So you don't have to

memorize any of these. But we're going

to go to next. And under contact

information, just going to put my email

here. And once you're done, just hit

continue. and create. So, just before we

go, I just want to go to audience and I

want to add a test user here, which is

an email that you're going to use to

send out basically the emails that you

want the agent to send out. So, in this

case, it's maroneia atcookcloud.com.

We're going to save that. And lastly,

we're almost there. We're going to go to

API and services again, and this time

we're going to go to credentials. And

what we want to do is to hit create

credentials with OOTH client ID and

application type we want to choose web

application and under name you want to

name it Canadian email oath client and

then we're going to add the authorized

redirect URLs which we can obtain from

our lab. So, going back into the lab

here, you see that this is the oath

redirect URL and we're going to copy

this and we're going to head back and

just fill this in and we're going to hit

create. As you can see, we now have the

client ID and client secret to the app

that we just created. So, we're going to

copy this and head back to client ID.

Paste it in. Client secret. Copy that.

Paste in the client secret. And you'll

see now that there is a sign in with

Google button that pops up. So, what you

want to do is just hit that and a Google

login popup will show and then you just

want to correctly select the email

address and you'll say that Google

hasn't verified this app. But because

you're the one who created it, you know

it's safe. So, we're just going to hit

continue and we're going to select all

because we want to have the agent be

able to do all these actions with our

Gmail. So, we're going to hit continue

now. And as you can see, it says

connection successful and we're good to

go. So, just wait a couple seconds here

within the labs and it's just going to

load up. And there you go. I already

have my credentials set up here and we

just want to run an execute step to show

you that everything is working. So in

the workflow you'll see that we've

chosen to define all of this by the

model. So we're going to let the AI

agent define this and we're going to

start chatting and say hi can you send

an email to.com

to just say hello. All right so we're

going to hit this. So as you can see now

the workflow has run and it's actually

sent an email to my Gmail. So let me

take a look and there you go. It says

hello Mercury just want to say hello

best regards. Okay. So obviously it's

not very sophisticated because actually

in the AI agent we didn't even specify

any system prom. So the whole point is

just to show you the main difference

between running the environment on N

cloud and within our playgrounds

specifically covering the part where

code key is being used as well as when

you're going to access any Google tools,

Gmail, Google sheets and stuff like

that. You do need to go to your Google

Cloud Console to set up the project, the

app, and the OOTH clients in order to

access Google services with the

workflow. And as you explore Nit in the

course, you're going to realize that

there's going to be some differences

between running edit cloud and edit

within your own self-hosted environment.

For example, the availability of

community notes, supported versions, and

a few other features that might only be

available on edit cloud. So if you're

facing any issues any part of that

build, just keep in mind that it might

be because you're using a self-hosted

version or a lab hosted version from

CodeCloud. And it's not necessarily a

limiting issue. There's always a workar

around for that. It's just something to

keep in mind about. I want to take a

minute to go through the different types

of authentication that we can make in

our approaches when we make HTTP request

nodes and the corresponding API calls.

Now, as you've seen in the previous

section, the HTTP request node is a very

convenient way for us to replicate an

API call without configuring everything.

And it's especially useful when we can

just go to an API documentation and hit

the import curl, essentially import the

entire curl command. Now, as you've seen

in the previous example, there are a few

ways that we can configure the

authentication method in these API

calls. Now, most of the API calls are

going to be header authentication. As

you can see in this section after I've

imported the curl, it automatically

configures the header section and

populate it with the name and the value

format of the API keys for

authorization. Now you can do it this

way or you can actually set up a

credential type here. Now I want to

explain the difference between the two.

So there are two main reasons why you

would want to do it this way by setting

up a credential type. The number one

reason is it's just super convenient

because the moment you have set this up

with the header off credential, you can

just pick from the list of credential

you've created in the next upcoming

nodes or even in another workflow. And

this is especially useful when you have

an API call that has a post and a get so

you don't have to set it up again in

next node. Now the second reason why you

want to do it this way is for security

reasons because setting up credential

type allows you to approach the

authentication without hard coding the

API key into the parameters and this is

just a better practice to ensure

security especially when you're sharing

the workflow on the project with

multiple team members and individuals

and all your credentials is going to be

stored under credential tab you can have

an overview of the list of credentials

that you've set up and of course you can

delete it or you can recon configure it

from this page. Now, that's it about

authentication. I'll see you in the next

section. So, in this section, we're

going to run through how to set up your

Google Cloud Console and connect it to

your NDN. That way, you're going to be

able to start using the Google notes

such as Google Drive, Gmail, Google

Sheet, Google Docs, and a variety of

others without setting it up over and

over again. So, the note that we're

going to start with here is the Google

Drive node. And the reason why we're

choosing this is because the way to

connect it is going to be similar across

all nodes. In fact, once you've

connected it once, everything else would

just be a simple matter of enabling APIs

through your Google Cloud console. So,

in this case, under credential, what I

want to do is to create a new

credential. And in order to get your

client ID and client secret, where you

want to go is to go to

console.cloud.google.com.

And depending on whether you've worked

with Google Cloud before, and by the

way, if you have and you already know

how to get connected with Nitnen, then

feel free to skip over to the next

section. But your page might look

slightly different from what I'm seeing

right now. But in any case, where you

want to go is on the top leftand corner.

You want to pick the project here. And

again, as you can see, I already have

one set up, but I'm going to set up

another one. Click on new project. And

I'm going to call it end demo KKK. And

under location, I'm going to pick new

organization here. And I'm going to hit

create. And there you go. It's creating

a Niden demo KKK project. I'm going to

click on select project and I'm going to

pull up the left hand sidebar here.

Where I want to go is under APIs and

services. I want to go to the OOTH

content screen and nothing set up yet.

So I'm going to hit get started. So app

information

and then demo KKK. I'm going to call it

the same name for the app and click that

as the email. And for the audience, you

have the choice of internal and external

depending on whether you have your

Google workspace set up. Internal is

only available if you have a Google

workspace. But in this case, I'm going

to hit external. And contact

information. I'm going to use the same

here. Hit next. Check on that and hit

continue. And we can go on create. All

right. So the next step, I want to hit

the create oath client. And what

essentially we're doing here is that

we're creating a project even though

it's not a real project. Google just

likes to use this as a structure to

identify and keep track of the oath

clients that people create. So

essentially where what we want to do

here is we want to pick application type

web application and in the name we're

going to choose nan demo kk again and

we're going to leave this as is and

we're going to add an authorize to

redirect URLs here and where we're going

to get this URL is from our workflow

or run our credential here and click on

copy and go back to our

Google console. Paste that and hit

create. So now we have the client ID.

We're going to copy that. We're going to

go back to our workflow here. So client

ID, we're going to paste that in. Go

back to our console here. And what we

want to do is to add a client secret.

And we're just going to copy that. Go

back to our work and paste the client

secret in. So just before we hit sign in

with Google here, which will activate

the connection, we're going to head back

to our Google Cloud Console and go to

branding and scroll down to authorize

domains. So in this case, what we want

to do, we want to add an authorized

domain, which is editn.cloud.

We're going to hit save. And the other

thing we're going to do is to type in

drive in the search bar and hit Google

Drive API because we want to enable this

API here. And of course, we're going to

have to enable Gmail API as well as

sheet and docs as well. But in this

case, we're going to just enable the

drive and we're going to try it out. And

once you've enabled the drive API, the

last thing you want to do is head over

to audience

and essentially publish the app. All

right, we're going to confirm that and

head back to our workflow here and hit

sign in with Google. and you're going to

be redirected to a Google signin page

and you're going to have to click your

email and it's going to say Google

hasn't verified this app, but it's okay

because it's you who's launching this.

And once you've done that, there's going

to be a couple permissions that you need

to run on the credentials. So, we're

going to select all of these and we're

going to hit continue. And there you go.

Connection successful. So, we're going

to head back to our workflow here. As

you can see, account is connected. And

once you've connected your account,

you'll be able to see the list of

folders that you have on your drive. So

in this case there's only one folder in

my drive which is KK test folder. So I'm

going to hit that and essentially we can

watch for file created for example right

so anytime a file is uploaded it's going

to trigger this. So we're going to hit

fetch test event. I didn't upload any

file so it says no data with the current

filter. So this is the folder right

here. I'm going to open that up. I'm

going to upload a dummy file, which is

an S SOP file. And there we go. So,

we're going to go back to workflow here

and rerun the test again. And as you can

see, now it's fetching the data of the

file that has been uploaded. So, cool.

For the other nodes such as Google

Sheet, Google Docs, what you need to do

is just to go back to your Google Cloud

Console and essentially enable your APIs

by typing in the Google notes or the

Google tool that you want to use. For

example, in this case, Google Sheets

API. You want to enable that and you can

go back and connect that on your edit

end and you're good to go. All right,

I'll see you in the next section. In

this section, we're going to build a

simple workflow of an AI research agent

that goes out to the internet and search

for the latest AI news and send it

across to you via email and at the same

time log those particular headlines so

that the next day the workflow is going

to check against the logs so that it

doesn't repeat the same headlines.

And this is a workflow that's going to

be super useful in this day and age when

AI news pops up every other day.

So this is what the entire workflow will

look like. We start off with a scheduled

trigger node which is preset to a

certain time of each day. Let's say 9:00

a.m. And then this goes to a Perplexity

search. So as you know, Perplexity is

one of the leading search AI agent out

there. And we're going to use that to

search the latest news or headlines that

is pertaining to AI development. And

then we're going to kick the output to a

checking agent. And what this does is it

dips into the Google sheet which locks

all the previous headlines of the news

to make sure that it doesn't repeat the

same headlines.

And after that, it's going to send a

summary of this AI news and send it out

to the intended recipient via Gmail. And

the final step is going to log those

headlines to make sure that the AI agent

stays up to date with those logs.

Now, let's build it together. The first

thing you want to do is introduce the

trigger note, which in this case is a

scheduled trigger note to kickstart the

entire workflow. With the schedule

trigger node, you can choose what the

trigger intervals that you want. In this

case, it could be days, hours, minutes,

or seconds or even longer than that. As

we want this workflow to kickstart every

single day, we want to choose days. And

the days in between triggers, we want to

keep that as one because we want it to

send out at every single day. The

trigger hour is something that we can

choose based on preference. And in this

case, we're going to choose 9:00 a.m.

because we like the news to be delivered

to us in the morning. And we're going to

keep the trigger at minute zero. And

we're going to hit execute step just to

populate the data. And there you go. We

have set our first trigger node. Now the

second note that we want to introduce

here is a perplexity note. And usually

when it comes to using thirdparty tools,

what you typically do is go to the API

documentation of the third party tools

and make a HTTP request towards those

API endpoints. However, in this case,

Perplexity has a native node that lives

within NN which you can choose from and

is easy to use. In this case, I'm going

to choose message model note from

Propexity.

And I already have my Propexity account

credential set up. But I'm just going to

walk you through very quickly how you

can easily do that. So in this case,

you're going to pick the dropown and

click on create new credential. And as

you can see is as simple as keying in

the API key from Plexity.

What you could do is go to Plexity.ai

and in the main dashboard, go to your

account and click on settings. And under

settings, if you look at the lefth hand

side tab, you want to hit API keys. And

your API keys page may not look like

this because you do need to set up your

API billing before being able to access

these keys. You can easily do that by

heading to API billing and putting in

your credit card information and topping

out the credits at a minimum of $5.

But in this case, I've already gone

ahead and do so. What we can do here is

to hit the create key button and it's

going to create an API key ready for us

to copy. Now head back to the workflow

and copy the API keys into the field.

Hit save and you have your Plexity

account registered to NAD. And there you

go. This says connection tested

successfully and your Plexity account is

connected to your NADN. And you only

need to set up the credentials one time.

Afterwards, you can simply pick from the

account credentials that you've set up

after you've connected your account or

your API keys in this case. What you can

do is to choose the operation that you

want perplexity to carry out. In this

case, it's a messenger model. And the

model that we want to work with is the

Sona Pro in this case. And there are

other models that you can choose from

such as the Sona Deep Research or Sona

Reasoning Pro. And each of these are

dependent on the use case that you want

to achieve. The deep research is going

to give you a large size of output rich

with data information and is perfect if

you're looking to do a very detailed

research about a topic. But in this

case, because we're only looking for the

headlines of the AI development news, we

want to choose Zona Probe. And as with

the other AI agent nodes, the typical

components that you have is the user

prompt as well as the system prompt. And

you can do that by adding a message and

changing the user to a system. And there

is a slight difference between the

workings of a perplexity node and those

of the other AI agent nodes. And

specifically, one of those is that

perplexity will require the system

message to come first before the user

message. So, what you want to do is to

swap this around and choose the system

prompt as the first prompt and the user

prompt as a second. Now, within the

system prompt, here's a prompt that I've

come up with. So, it says, "You are an

expert AI research analyst specialized

in tracking and summarizing the latest

developments in artificial intelligence.

Your task is to deliver concise and

up-to-date news summaries, focus on new

AI model releases, research

breakthroughs, and strategic moves by

key AI players. So that's the system

promp. As for the user prom, what I want

to do here is to prompt it effectively

so that it summarizes and extracts the

most relevant information or news for

me. And I'm going to blow this up so

that's easier for you guys to read. And

here I go. It says find and summarize

the most recent AI news within the last

24 hours. For reference, today is, as

you can see, this is intentionally left

blank because I want to drag the

variables of today. And in order to find

that, you would go to variables and

context. And this is applicable for

every single workflow. You're going to

have a default variable of the current

time as well as the current date. And in

this case, I want to drag the value or

the variables of the current date and

put it here. And this will dynamically

change the value depending on what the

date is. And so you can see on the right

hand side it says today is date time

today's date. And the reason why I want

to do that is because a lot of the times

the AI is prone to hallucinate the

timing of the day. So when you tell it

to get the most recent news, it doesn't

have context on what today is. So when

you tell it to get the most recent news

in the past 24 hours, often times it

might hallucinate the day to be a day

within last year or the year before. So

this way we ensure that the AI has

context of what today is and be able to

focus on news that are coming in from

the past 24 hours. And also I've added

in to make sure that it prioritizes

research breakthroughs and key

announcements by organizations like

OpenAI, Anthropic, Google, Meta,

Mistral, XAI and Hunking Face. And this

is just some guardrails to make sure

that it focuses on the news that are

most interesting to me which is the

development and the release of new

models of AI. Okay. So now that the

system and user message has been set,

the last thing I want to do is to go to

add option and click on the dropown

button and perplexity has a very neat

feature which is the recency filter. So

we can set the search recency filter to

just within the day. So in this case

it's going to restrict the search to

within the most recent news that have

been put out there. Okay, now we're

ready to execute the step. And as you

can see, the output is various types of

data that has to do with the research

output as well as some of the relevant

information regarding the search. I'm

not going to go through each of these

variables. But what is interesting to us

would be the number of citations here,

which means are the number of news

outlets and links that it has found to

be relevant to your search and also the

search results.

And finally, the output content which is

in natural language where perplexity

will respond to you and say there have

been several notable AI developments and

announcement for the last 24 hours

particularly concerning major strategic

moves and those reinvestments.

And what we want to do here is to pin

the data so that we don't have to rerun

that every single run is going to use up

the perplexity API token. We want to

make sure that we're being cost

effective and prudent about it. The same

thing with the schedule trigger. I'm

going to pin this data so that every

time we run through the workflow, it's

not going to exhaust a new API token and

provide new data. After the flexity

node, what I want to introduce is the

formatter agent or in this case the

checking agent. We're going to use

OpenAI in this case and we're going to

pick a simple message and model action

as part of the workflow. And as you can

see, the output of the perplexity node,

which was the previous node, becomes the

input of the OpenAI node. And again, I

already have my account all set up as

we've covered in the previous sections.

So, you should have the same account set

up here as well. And we're going to

leave the resources as text operation as

message and model. And we're going to

pick a model here. In this case, we're

going to pick a GPT

4. That should be enough to check

against the log. And just a reminder of

what we're doing here with this node is

we want this node to check against a

specific Google sheet log of past

headlines that has been recorded. And

the reason we do that is because some

headlines are going to be reported

twice, three times, four times, or

repeatedly depending on how big the

headlines is. If we don't add this

particular note within the workflow,

what happens is it's going to report

some of the headlines again and again

thinking that it's a new headlines in

the past 24 hours. We're going to rename

this news

checking agent. Now for the user prompt,

what I'm going to do is I'm going to

give it a set of prompt here and I'm

going to blow it up so that you can see

it says here's today's AI news list from

Perplexity.

So I'm going to add the summarized

content from the previous perplexity

node which is under choices under

message and the variable is called

content and we're going to just drag and

drop as you can see in the right hand

side this is the whole chunk of the

summarized news headlines that came from

perplexity

and then later on I'm going to ask

please cross check the news against the

past news lock sheet and remove

duplicates

So this is the name of the news lock

sheet that I'm going to create later on

so that they can compare and check

against such that there won't be any

duplicate news on the output. I've also

added here that we should ensure each

item has a clear impactful one to two

senses summary. So there's a bit of

summarization task here. Also after each

summary include the full unfold source

URL. This is to make sure that you're

interested in that particular news. You

can click on that URL and get access to

where it's been cited from. Use bold to

highlight company names or major

updates. This is just for readability

and add spacing between news items for

readability as well. And at the end,

I've added tip perplexities output news

are all duplicates from the past news

lock sheet. Respond appropriately with

no notable AI development news in the

past 24 hours. is just to make sure that

it's only generating content when there

are new headlines and not recycling

headlines again and again. So that is

the user prom. Meanwhile, we need to add

a system prompt. In this case, we choose

an add message and we're going to choose

a system message or a system prompt. And

in the system prompt, we're going to do

a simple prompt. And I'm going to blow

it up to say you are an AI news format

and your role is to process, check and

clean AI news result fetch from

perplexity to ensure readability and

that past headlines are not repeated.

I'm going to add when checked

against

Google sheet tool named past

and use lock. So this is just to ensure

that it knows that the tool exists and

that you're supposed to use that to

counteract the news headlines and make

sure they're not repeated. Okay, cool.

So now we are going to add a tool here

which is essentially the Google sheet

tool. All right, click that. And I've

already have my Google Sheets account

connected, but we've covered how to

connect your Google Sheets or Google

Docs or your entire Google account

through a Google Cloud console in the

previous section. So, you should have

the same setup. Under the tool

description, we're going to set

automatically because we're going to let

the AI decide when to use this tool. At

the same time, we're going to have to

determine some of the configuration

here. And under resource, we're going to

want to to refer to a sheet within the

document. And the operation is to get

rows because we're not trying to create

new sheets. We're not trying to update

the sheets. We're simply trying to check

against the sheets if the headlines are

repeated. So, we're going to use get

rows. Now, we need to fetch a document.

But just before we do that, we need to

create the actual sheet where all the

information of the past headlines are

going to be logged. So I've created a

sheet here called the passi news log and

it's a simple sheet with a date column

and a headlines column. So the date is

going to correspond to the date of the

headlines. And the headlines basically

is going to say the content of the

headlines in one or two sentences. Now

let's go back to the workflow. Once you

have the sheet set up, you can then find

the sheet just by selecting from list

and clicking on the right one which is

pass AI news lock. And then later on you

can choose from the sheet the correct

sheet in this case which is sheet one. I

only have one sheet in the document. So

that's the correct one. You should be

able to see now

and that's it. And just before we go, we

want to rename this to pass AI new law

because we want to make sure that the AI

knows that this is the particular Google

sheet that we're referring to when we

told it that there's an attached tool to

it. Now, going back to the workflow,

we're going to run this particular node

here based on the configuration that we

filled it. You can see based on the data

that came from perplexity

is checking it against the news log

which in this case is an empty sheet. So

there's no headlines to be checked

against but still doing that. And

finally the output is here's the

formatted AI news which is Julius AI has

successfully raised 10 million funding

round amongst other things. In this

case, the next node that we want to

introduce is the email node because

right now all the output lives within

the workflow. Just like in the previous

section, we're going to attach a Gmail

node. But in this case, we're going to

still choose the send a message action.

But the difference, as you can see, is

that we're not attaching the Gmail as a

tool for the checking agent to leverage

on. Now, why is that? And what's the

difference between attaching Gmail as as

a tool versus Gmail as a subsequent node

to the AI agent node? And the difference

is that when we attach the Gmail tool to

an AI agent note, the AI agent has the

liberty to choose whether or not to

leverage the tool or use the tool. In

scenarios where AI agent hallucinates,

they might actually use the tool when

they don't need to. They might not use

the tool when they actually need to.

Having Gmail note as a note that is

chained up to the workflow like this

ensures that any then runs through the

note regardless of what the AI agent

thinks. In this case, we make sure that

the note is run no matter what. And this

is extremely useful when we're sure that

each time of the day we want to receive

an AI news output into our Gmail because

a note has to be run every time a

workflow is run. It is just best

practice to chain it up as part of the

workflow instead of connecting it to the

AI agent as part of the tools that it

may or may not use. Now, let's go back

to the Gmail node. And you can see that

in terms of configuration, there's some

slight differences between a Gmail node

that is chained up to the workflow

versus a Gmail tool that is attached to

the AI agent. You now don't have the

option to let the AI choose to define

the recipient, the subject, and the

message. you actually have to hardcode

it into the node. And I use the word

hardcode, but actually you're simply

dragging the variable and dropping it

into the field. The fields appear to be

the same as when you have a Gmail node

attached to your AI agent. So, we're

going to leave it as that. But here's

where it gets a little different. So the

recipient that you want the output of

the Gmail to be sent to should be

consistent because you want this to be a

daily operation sent to the same set of

email addresses. So in this case I'm

going to put my email and under the

subject I'm going to put daily AI news

digest.

And this doesn't need to change because

I know what it is. And email type

because it's all going to be text. I'm

going to choose text and the message is

actually from the content here. And as

you can see the content itself, it says

here is formatted AI news. So this is

not ideal in the way that I want to

receive the summary. So we can actually

prompt this a little bit better. So once

we're done with this, let's go back to

the news checking agent and under the

user prompt and we blow it up. Let's add

an output format rule output in text

only and start the

news

daily news summary with here is today's

AI development news

today is and I'm going to just track the

today variable and drop it here. So, I

wanted to start this way so that it

makes sense. Every time I open up the

email, it's going to say, "Here's

today's AI development news. Today is

that particular day." So, I know that

I'm getting the latest news from the

headlines. Okay, now that's done. As you

can see, when I make any changes, it's

going to turn the node yellow. So, we're

going to run it one time and let's see

what the output now is. Here's today AI

development news. Today is 7th, sorry,

today is 29th of July, 2025. And then

the news. So, okay, that's the format

that I want it to be. I'm going to open

that up again and pin it so that I don't

have to rerun it every time. And within

the Gmail, this is the content that I

want to drag into the message field. So,

that's good. And remember in the

previous sections, we know that it's

going to automatically come with an

added attribution within the email. So

what we want to do is just click add

option and it toggle off the append

attribution and then we're going to try

to execute step. As you can see there

has been some actions and then email has

been sent. So let's go to our email

inbox and see what we've received. So

this is the email that I've received and

it says here's today's AI development

news. Today is this date. Of course this

formatting is not ideal. You can go into

the prom engineering again and try to

format it in a more readable way, but

essentially in terms of formatting, it's

much easier to read now. And of course,

these are the news, the headlines,

and if you push it to production, you're

going to be receiving these emails day

in day out and keeping yourself updated

with the latest AI development news.

Going back to the workflow because we're

not quite done yet. Even though we're

now receiving the daily news update and

I'm going to pin this as well is that

remember the AI news log that we've

introduced here. We want to make sure

that whatever news that have been

researched in the past 24 hours has been

logged into the sheet. That way the very

next day the agent or the news checking

agent can then check the log and make

sure that the headlines are not repeated

again. and the agent will always be

checking against past news to make sure

that it's not repeated. So in this case,

what we want to do is we want to

introduce a Google sheet node. Click on

Google sheet. As you can see, there are

a couple actions that we can do with

Google sheet. And what we want to do

here is to choose an append row in

sheet. And the reason why we choose

append row in sheet is because we want

to add new information on an existing

sheet. So we've got the Google Sheets

account connected already as usual as

what we've covered in the past section

and under resource we want to access the

sheet within the document and the

operation we want to do is to append

rows or append new information. Now

under document what we want to choose is

the past AI news. So just type in the

title and you can choose that. The

alternative is actually you can link it

by typing in or copy and pasting in the

URL or the ID of the document. And once

you've typed that in, you can then

choose the list of sheet. And again, we

have only one sheet. So the sheet sheet

one and once you've done that, you'll be

able to see the columns that now exist

within that particular sheet. And within

the sheet that we've created right now,

it's completely blank because there are

no headlines. They've been logged in. We

have two columns, which is date and

headlines. And if you go back to the

workflow,

there's the date column and there's a

headlines column. So now what we do is

we want to populate this particular rows

in under these particular columns with a

specific data. So in this case, we want

to choose under variables and context

the today variable and drag that and

drop it under date. So that way it's

going to always populate the date column

with the value of today's date. And

under headlines, what we want to do is

we want to choose under the news

checking agent the output of the

content. And now as you can see when you

drag variables from other nodes, you

don't necessarily always have to drag it

from the previous or the immediate

previous nodes. You can drag it from

nodes that were from previous instances

in the chain. So in this case, I'm going

to drag the content variable from a news

checking agent and just drop it into the

headlines.

Now with that done, I'm going to try the

execute step to see what it does. And as

you can see, it's telling you that it's

added the date component or rather

today's date into the date column and

headlines into the headlines column. And

of course, the way the logic works is

it's going to pick the topmost row that

is unpopulated within the sheet. So it's

always going to pick the next available

rows as you go. So now that that's done,

we've ensured that there is some form of

loop in terms of the way the workflow

checks for new headlines. And now with

the workflow, we've created a form of a

loop where the perplexity agent every

day would go out and search for the

latest information on AI development and

a news checking agent checks against

past logs of the headlines to make sure

that the information is not repeated

before sending it into your Gmail. All

of that gets stored within a Google

sheet every day. And then the next day,

the same operation is going to happen

again. The research agent is going to go

out into the internet and check the most

recent and relevant news for you. And

then we also have the agent checking

that against the Google sheet together

with the news that were locked in the

previous days before sending it out into

your Gmail. So that way you can be sure

that the headlines that you receive

every day are really the most relevant

and the most recent news that you can

get. There are many other options of

tools that you can use along this

workflow. Perplexity is just one of the

options even though in my opinion it's

one of the most advanced out there when

it comes to search. OpenAI can be

replaced with any other LLMs as well.

And you can replace a Gmail with a

Telegram node or Slack node or any other

platforms that you spend your most time

on. In any case, I hope this section

provides a good understanding of how a

general logging mechanism of the

workflow can possibly work. So that's

the end of the section and I'll see you

in the labs. In this section, we're

going to have a little bit of fun. So

since we now know how to create an AI

agent on end and how to connected to our

Slack accounts, we're going to try to

see if we can have the AI agent fully

take over our profile and respond on our

behalf, aka do our job for us. So in

this case, this is the same workflow

that we've built in the past section.

The first thing we need to do to have

the AI agent respond on our behalf is to

go to our Slack API and reconfigure some

of the OOTH permissions.

And what we want to do here is simply go

down to the scope section. And in the

past section, we filled up all the bot

token scopes as required. But in this

case, we're going to scroll down to the

user token scopes and fill it up with

the required scopes. And these are the

list of required scopes that are

typically recommended in order for it to

function well. So as you can see we have

channels history to read the channels

just in case we want to respond and

interact uh within the channel and we

obviously we have the channels read chat

right groups history as well as some

read and history permissions and this

all depends if you want the bot to

respond on your behalf in a channel in a

direct message or in a group message.

So once you're done adding all the new O

scope under the the user token scopes,

what you need to do would be to

reinstall the app. There will be a

notification popup that comes up here.

I'm just going to show you right here.

So if I were to add

let's say an admin apps read. What will

happen is it's going to ask me to

restore my app over here.

So just go ahead and do that and you're

good to go.

And just before we go under the bot

token scopes, you do need to add one

more permission which is essential which

is the chat write.customize.

And this just allows the bot to send

messages on your behalf.

All right. So once you're done, you want

to copy the user o token and go back to

your n workflow.

And here you're going to create a new

credential which I've done here.

But essentially what you need to do just

do the same thing. pick access token and

paste the user O token here.

I'm not going to click on save because

I've already created one, but it's the

same exact process as we've done in the

previous section. And once you're done

with that, you want to toggle over to

the event subscriptions. And what you

want to do is just unfollow the

subscribe to events on behalf of users.

In this case, also add these four event

subscriptions, which is message

channels, message.groups, message, and

message.mppim. This just covers most of

the use cases that you might need. And

if you want to know a little bit more

about the permissions and how it all

works, I encourage you to go into the

Slack documentation and to check it out

yourself.

But now I'm going to go back to our end

workflow and reconfigure some of these

nodes. So in the first trigger node,

previously we have the bot/ app mention.

Now I want to add on just any event as

one of the event triggers. The reason

for that is I want to make sure that I'm

capturing direct messages when my team

members are chatting with me. And I also

toggled on the watch o workspace. The

reason for that is again I don't want to

be just watching a single channel or a

single chat. So when I toggle this on

any message received will constitute a

trigger event which would then run the

workflow. And scrolling down under

options what I want to do is also add

the username or ids to ignore.

In this case, I'm going to fetch my own

ID to ignore my own messages because

right now how it works is it's going to

trigger on any event, which means it

will also include any messages that have

sent to any channels as a trigger event.

So, how do I get the user ID?

Well, you can get your user ID from your

Slack channel by just simply clicking on

profile

and clicking on the three dots here. And

you're going to be able to copy the

member ID. So that's the ID and you can

go back to your NDN workflow to the

trigger node

and essentially paste the ID under the

expression for the username or ID string

ignore.

Cool. For this demo, I've created a

fictitious workspace. I've added one of

my colleague Michael Forester into the

channel. And obviously this is not the

real Michael. This is just a profile

that I've created. And again, as you can

see, this is just a demo workspace. And

I'm even on a free account on the left

hand side. You can see that it's 23 days

left on the trial.

And what I've done was I created another

account as a fake Michael Forester. And

it says Michael F, which is me.

It says Michael F, which is this

account. And the left hand side, you see

this is my account. And the reason why

I'm setting up these two accounts is so

that I can show you how it all works

when my team member chats with me on a

direct message. So now I'm going to try

to complete the reconfiguration of the

entire uh workflow. And what I'm going

to do here is go into the agent node and

everything stays the same with the

exception that I've added a system

message here. So essentially what I'm

telling the agent is that you are

maronei a team member at corkcloud and

your job is to impose a maronei as well

as you can and respond to his team

members message on the behalf sound

friendly and natural in a typical tech

working environment. So as you can see

is a pretty broad system message. Um,

you can obviously prompt this much much

better, but I'm going to just start with

that.

But one more thing that I'm going to do

is I'm going to go to a chat model and

I'm going to just toggle this to GPT5

because it just came out.

And GP5 in my experience has been great

at responding to complex queries um and

using tools when required.

So I'm going to leave it as that. The

simple memory is still attached.

And on the last note, which is a send

message note, I made two changes here.

And the first one is, as you can see,

the channel ID, I made it into dynamic

ID. It's showing an error here right now

because it doesn't have any data

populated just yet. But essentially, I

want to make sure that this is going to

be dynamic because the workflow now is

triggered by any events watching the

entire workspace. So, it's not

necessarily going to be restricted to

one single channel. So it could be a

direct message, it can be a group

message as well as a channel message. So

I just want to be able to dynamically

reply to that specific channel where I'm

interacting with people on.

And the second thing that I'm going to

change is under option, I'm going to say

send as user. I'm going to put in my

username here which is Mark. And that's

it.

Okay. So I'm going to hit execute

workflow here.

As you can see, it's waiting for a

trigger event. So, I'm just going to go

back to our Slack app.

In this case, I'm going to use the

Michael Porser account here to chat with

me. So, I'm going to say, "Hey, Marone."

All right. So, let's go back to the

workflow. As you can see, it's getting a

trigger. It's hidden the AI agent and

it's responding to my message. So, let's

see what kind of message or response

that I'm receiving back. It says, "Hey,

what's up? Need an update on anything

AWS/resour

or is there something that can help

unblock?" All right, sounds pretty

helpful. And I think it threw this in

because I said, you know, pretend to be

somebody in CodeCloud. So, CodeCloud's

associated with cloud technologies, etc.

So, it's probably just throwing that in.

Okay. So, I'm going to go back to the

workflow. I'm going to hit execute

workflow again. So that it's now waiting

for my trigger.

And using Michael's profile, I'm going

to type.

How was your weekend? And I'm just going

to stay social. Not going to talk about

anything about work. So there you go.

And it's responding pretty good. Kept it

low-key. Got some rest. Did a bit of

planning over the week. I also sketched

out a couple of tweaks for the AWS and

the zero pipeline that I'll share after

the stand up. Cool. All right. Um, it's

adding it own flare. So, not sure about

that. Uh, probably need to work on the

prompt a little bit, but cool. Um, so,

so it's working quite well now.

Responding as myself, even though I

don't normally sound like that. Um, I

don't necessarily work intimately with

all the cloud stuff, but I'm just going

to go back to the workflow here. And I

want to try to make the agent a bit more

useful and give it more context uh than

it currently is. And this is where I'm

going to try to introduce a pretty

simple rack system. And in this case,

I'm just going to add a tool and this is

just going to be a Google doc tool

because the reason why I'm doing that is

I realized that people are going to

start asking me about projects etc.

So what I did was I created a document

which is a Google doc document and

essentially this is completely

fictitious but it outlines kind of the

update of the project progress in terms

of some of the stuff that we're trying

to do and this is again completely fake

but it's just a document so that we can

see how it all works. All right cool I

have this Google doc drafted out. This

is where I put all the fictitious

updates. So I'm going to connect that

here and under the Google doc node what

I want to do is is a get operation

and I'm going to go back and just copy

link which is a URL link of the cloud

project status

and I'm going to paste that URL here.

All right. So I'm not going to do

anything with this node. This is just an

attach tool. I'm going to go back to the

AI agent and under system prompt I'm

just going to add a tool description.

Use the Google doc tool when asked about

project updates.

Cool. All right. So, I'm going to just

make that change. I'm going to save that

and I'm just going to click on execute

step here and go back. Ask Michael.

I'm gonna ask

how

is the is your project going.

As you can see, it is kicking it to the

air agent. It is thinking it is going

into the document checking the latest

update of the project before kicking it

back into Slack. So, let's see what it

responds to me.

So, there it goes. says quick as your

project I made objective interactive lab

design done 65% complete ETA 15th

October 2025 now let's see if that's

correct

so this says that the lab is 65%

complete and expected completion 15th

October 205 so it's successfully

fetching the relevant information into

the message however as you can

So, it's giving me the right

information, but as you can see, it's

missing a little bit of a human touch.

So, that's probably something I can work

into the assistant prompt for it to

respond more like me and more humanlike.

But at rarely, this gives you a good

understanding on how you could leverage

AI agent. Maybe not to impersonate

yourself just yet, but in a lot of other

use cases. As you can see, once you've

created a rag agent like this, this

could function as your technical support

or just a bot FAQ.

And this could work for HR, legal,

marketing, whatever the case may be, or

even customer support. But there you go.

This section is to really just

familiarize yourself with the select

permission logic and how you can

potentially leverage that to its full

extent.

Again, this is just for fun, and we're

not recommending you to replace your day

job with an actual AI agent just yet.

See you in the next one. In this

section, we're going to walk you through

how to set up the Slack API and connect

it to your NAN. And then we're also

going to go through the various

different permission levels as well as

the logic in terms of integrating Slack

into your NAND workflow. If you're

already familiar with how Slack API

works, feel free to skip this section

and go to the next one. So, first where

you want to head to is the Slack API

page, which is api.slack.com.

And once you've signed in, you want to

go to your app section and hit create

new app. And what you're going to do is

come up with an app name. In this case,

I'm going to call it the end devilbot.

And pick your workspace and hit create

app. Now, what you want to do in your

bot configuration page is first head

over to permission. So, there are a

couple things to keep in mind here in

terms of the architecture of the bot.

So, first we need to configure the

permission level that we want to grant

the bot on Slack. In this case, we want

to scroll down to the permission scope

and we're going to hit add an O scope.

In this case, a couple of basic

permissions that I want to start off

with. So the first being the obvious

which is the app mentions read. So this

is going to let the bot read any

messages that you've sent to the bot

directly by pinging it. And the second

one that we want to add will be the chat

write which allows it to send messages

as any demo bar. And just to make sure

that it could read our messages on

channel we want to add channels read and

group read as well. Lastly we want to

make sure that they could actually see

the name of the people that are sending

the messages. So we're going to add the

user profile read as well as users read.

So that should be enough for now. And so

we're going to go ahead and install the

bot to our Slack workspace, which is

called the Slack automation lab. And

we're going to click allow. And again,

you need to be an administrator of your

Slack account to be able to do this. As

you can see now, we've got a bot user O

token generated. So we're going to hit

copy on this because we're going to use

this to create a Slack credential back

in our NN workflow. So going back to our

N& sheet, the first trigger node that we

want to introduce here is the select

trigger node. And I want to make sure

that we pick from the selection of the

trigger nodes and not the action node.

In this case, we're going to hit the

onbot app mention. And I already have a

Slack vential preset up here, but I'm

just going to recreate another one just

so that I can show you how it's done.

So, we're going to name this Slack and

then demo code cloud. And I'm going to

paste the key here and hit on save. As

you can see, the green bar shows up.

Connection tested successfully. So, I'm

going to go back to the API page here.

And right now I'm going to go to the

event subscription because I want to

make sure that's listening to the events

that I've specified. I'm going to toggle

this to on. And for the request URL is a

URL that I'm going to grab from my endn

workflow. So I'm going to hit the web

hook URLs here. As you can see, there