By the end of this video, you'll have

your own inbox agent that is classifying

and labeling all of your incoming

emails. We're going to build all of this

together step by step with no code so

that you guys will understand how you

can actually start to scale up this

system to become a full inbox manager.

So, I don't want to waste any time.

Let's get straight into the video. All

right, so the first thing we're going to

do is actually wireframe this out just

so we're all completely aligned on the

way that this system is going to work

before we actually hop into NIDN and

start building. So, I'm going to take

you from zero to having an inbox agent

in this video and we're going to use

just three tools. Nitn is where we're

going to be building this actual system.

You can get started on a 14-day free

trial using the link in the description

and then after that it's like 25 bucks a

month. We're going to be using Gmail

today specifically, but if you do have

Outlook or a different email provider,

you can follow along. It just might not

be the exact same clicks at the exact

same spots. And then we're going to be

using a tool called Open Router, which

is where we actually connect to

different AI models. So, like I said,

we're going to be wireframing this out,

which basically means a visual diagram

of what the system is. And this is

something that I strongly encourage you

do before you build any sort of system.

You have to have all of the steps of the

process mapped out clearly before you

actually start building. And this is

something that I teach in my full course

as well. If you want to check that out,

the link will be in the description.

Anyways, every single process starts

with some sort of trigger. And our

trigger for this workflow today is going

to be a Gmail trigger. So you can see

that we have Gmail. We're saying it's a

trigger. And the trigger action is

whenever we get a new email in our

inbox. So that's going to kick off the

rest of this workflow. So when that

email comes in, it's going to have

information. It's going to have stuff

like who sent this email, what time was

it, what's the subject, what's the body,

and then some other metadata as well.

And so, we're going to feed this Gmail

information into an AI classifier node.

What we want the AI classifier node to

do is it's going to look at the subject

and the body. So, I'm going to real

quick just say basically the incoming

information will be the subject of the

email and the body of the email. And

then it's going to basically read

through that and understand what type of

email is this. And so for the sake of

today's video, we're going to do four

generic categories which are customer

support, finance and billing, high

priority, and promotion. So we're

basically going to explain to this AI

classifier how to interpret the emails

as one of the four. And then it will

shoot that email string down into one of

these four categories. So then it's up

to us to basically handle what goes on

the rest of that flow. So let's start

with just one example. Let's say we get

an email. the AI decides, okay, this is

a customer support related email, so I'm

going to send it up here down this first

branch. So maybe we know if we get a

customer support email, we want to go

ahead and actually label it in Gmail as

customer support and keep that in this

specific inbox or that specific folder.

Next, we may actually want an AI agent

to go ahead and send off a draft or even

just a reply to this person. So we can

basically teach this AI agent how to

respond to customer support related

inquiries. And then what we can have it

do is shoot off a reply right back in

Gmail itself to whoever sent that

initial email that triggered that

workflow back over here. So for customer

support, let's just say that's what

we're doing. So the idea is based on

what category of email comes in, you're

in full control of what you want to

happen. We probably want all of them to

have some sort of label where it gets

put in that category right away. But

then from there, I don't know if we want

to instant reply. I don't know if we

want to just create a draft. I don't

know if we even maybe just don't want to

respond at all and just notify someone

else about this inquiry. What do you

want to happen? That's what you have to

set up here. So when we get into end to

end, I'm going to show a few different

things that could happen after each

path. But basically, an email would come

in, it would get classified. Let's say

it gets classified as high priority,

then we would take the next steps only

in that high priority branch. But if a

promotional email comes in and gets

classified as promotion, then maybe a

different order of operations would

happen down here. So now that we've seen

a wireframe and we understand the system

that we're building, let's actually hop

over to end, open up a new workflow, and

go ahead and get started here. All

right, so this is what our blank

workflow looks like. And as you saw

based on the wireframe, we know that we

have to start this process with a Gmail

trigger. So I'm going to click on add

first step. I'm going to type in Gmail

and click on this button right here.

Automatically, this pops up with

triggers. There's 26 different actions

we can take, but it knows that there's

nothing else on our workflow yet, and it

has to start with a trigger. So it

prompts us to click on the onssage

received trigger. Now the first thing

that we have to do is actually sign in

to our Gmail account. So you would

basically open this button right here

and it would probably just say create

new credential. You click on create new

credential and just go ahead and sign in

to the email that you want to use for

this inbox manager. And if you're

self-hosting or locally hosting your NIN

and it's not as simple as just signing

in here, then I will also drop this full

complete beginner's guide on NADN and

setting it up with Google in my free

school community. The link for that will

be down in the description. When you get

in there, it will look like this. You'll

click on YouTube resources or search for

the title of this video. And then when

you find the post associated with this

video, there'll be a file right here,

which will be this Google guide. And

then you guys can hopefully use that to

get set up. Anyways, once you get

connected here, it should look like

this. You basically can set up how often

you want this to pull. So, every minute,

every hour, every day, whatever. We're

just going to leave it on every minute.

And then the event here is going to be a

message received. And then what we're

going to do is real quick, I'm going to

shoot myself a sample email from a

different one. and we'll pull that in so

we can just take a look at how that

comes into nit. Okay, so I've got this

Google sheet here with sample emails and

we've also got definitions that we'll

use later. This will also be available.

So if you want to grab that so you can

follow along with the video or you could

also go to chatbt and just have it make

up some. This will also be in my free

school community along with this guide.

So what I'm going to do is grab this

customer support type of example. I'm

going to grab the subject which is issue

logging into dashboard and throw that in

as a subject. I'm going to grab this

body and throw that in the email and

then I'm going to send it off. And now

back in edit end, I'm going to hit fetch

test event which is going to check for

the most recent message in this Gmail

inbox. And you can see right here that

we have the email that I just sent

myself and the subject was issue logging

into dashboard. Now, one thing you're

going to notice real quick is we just

have a snippet. The issue is this is

just going to give you a short summary

of the email body and a lot of times

it's going to be too long to actually

fit here and that's because we have this

little toggle on which says simplify. So

I'm going to toggle that off. Hit fetch

test event once again. And now you can

see we get way more information back

about this email. And if this was a long

email, we would have gotten the full

text right here. So I'm going to leave

this as simplify turned off. And now we

have this data to play with. So, what

I'm going to do is hit this button in

the top right called pin data. And this

is basically just going to save our data

so that if we refresh or something like

that, we don't lose that and have to go

pull it in again. We now just have this

data sitting here that we can use. And

like I said, we can test the rest of our

workflow with it. Okay, cool. So now,

according to our wireframe, as you can

see, the next step is we need to set up

our AI classifier node to read the

subject and body of that email and

classify it in one of these four

categories. So to do that, I'm going to

click on the plus right after the Gmail

trigger and I'm going to type text

classifier as you can see right here. So

this is how nodes work. On the left hand

side over here, we have our input. So

this is everything about this Gmail

trigger. We can pull any of this

information in to the middle and have it

be classified or whatever we want to do

with it. The middle is the

configuration. So we'll set up the node.

And then the right will be the actual

output data. So in the middle, the first

thing that I'm going to do is you see

there's a big box that says text to

classifier, which basically means what

am I looking at? And so what I'm going

to do is real quick change this to an

expression, which means that we can use

variables. And variables are basically

just these things over here that could

basically change. And you you will see

what I mean by that in a sec. I'm going

to click on this button to open it up

full screen. And now I'm going to give

this node what to look at. So we know we

want it to look at the subject and we

want it to look at the body. So I'm

going to just basically indicate to it

that that's what it's about to receive.

And then on this left hand side, I just

need to find the subject in the body. So

if I keep scrolling down, I should

eventually find out of all of this messy

metadata right here is the subject. So

this is what I mean by variables. The

Gmail node every time will output a

variable called subject. And that

subject is a variable because it

represents whatever actually is the real

subject. So no matter what the subject

is, as long as we have this variable

here, the agent or the text classifier

AI node will actually be able to read

that. And then same thing for the body.

This comes in a variable over here

called text. And I'll just drag that in

as well. And now every time we get an

email, this AI node will be looking at

the subject and the body of the email.

Okay, so we have that set up. But now we

need to actually tell it the different

categories to

place the emails in. And so you can see

there's a button here called add

category. And I'm going to just go ahead

real quick and add four because we know

we're going to set up four different

categories. But before this, as a

prerec, what you want to do is go into

your inbox in Gmail. You can see on this

lefth hand side, I already have four

labels right there. And what you would

do is just create a new label, name it

whatever you want. You know, you can

choose the color. And then just make

four of those if you want to follow

along with this demo. So you can see

I've got customer support, finance and

billing, high priority, and promotion.

And so now what we need to do is give

this AI node those exact same

categories. So I'll start by writing

customer support. Down here I'll do

finance/ billing. The third one I will

do high priority. And then for the

fourth one I'll go ahead and type in

promotion.

So now it knows okay these are the four

categories. And what we need to do next

is give each of these categories a

description or a prompt. So the AI

basically can read this and say, "Okay,

which one does it fit in based on the

description?" And so I made this real

simple for you guys. You can go to that

Google sheet that I gave you earlier,

click on the definitions tab down here,

and just copy and paste in these

definitions. So here's our customer

support definition. I'll paste that in

right there. I'm going to go get the

finance and billing description, and I'm

going to paste that in right there. And

I'm just going to do the same for the

bottom two. You guys don't have to see

me do that. Now, what I will hit on is

the fact that descriptions and prompts

are very important and being as specific

as you can while still being concise is

really, really important. So, right

here, let's just read one of them. For

customer support, I said emails

requesting help, troubleshooting, or

guidance for any product or service.

Usually describes issues like login

errors, bugs, or malfunctioning

features. And then I also gave it some

common key words like error, issue,

problem help access support fix

reset, troubleshoot. And so what you

would do is over time if you realize,

huh, my AI node is classifying things as

customer support way too often, then

you'd come in here and you'd make a

change to this prompt. Or, huh, my AI

node is missing a lot of customer

support emails. I wonder why. Let me

come into this prompt and make this a

little bit more detailed and specific.

And that's what you would want to

monitor with all of these other

categories as well. So anyways, this was

a customer support email example. I'm

going to go ahead and hit execute step.

And what happens is we get an error. And

the reason why is because we need to

connect AI to this node. And we

currently haven't connected AI. So

you're going to go to open router.ai.

You're going to create an account. And

then once you create an account, you

need to do two things. The first thing

is go up here to your credits and you

need to add billing information. You do

have to pay for AI, but if you add five

bucks just to play around, that will

last you a while and then you just need

to add more later. And then once you add

credits, you can go over here on the

lefth hand side or back up here in the

top right and click on keys. And this is

an API key that you will use to access

that billing information. So it's kind

of like a password, which is why you

shouldn't give other people your API

key. So what I'm going to do is click on

create API key. I'm going to name this

one inbox demo. And I'm going to create.

And now it gives me this long string of

letters and numbers and a few dashes

that I'm going to go ahead and copy. And

now when I go back and edit in, I need

to add a chat model. So I'm going to

click on this plus button. I'm going to

type in open router, which is the

platform we were just on. And then it

will prompt you right here to create a

new credential. And that's where you'll

paste in that API key right here. So you

can see there you go. I just pasted in

my API key. And this is the exact same

long string of letters and numbers that

I went ahead and grabbed over here in

Open Router. And don't even think about

trying to use this one. I'm going to

delete it as soon as I finish recording

this video. So anyways, I'm going to go

ahead and hit save. And now you can see

we got the green connection tested

successfully. I would also get in the

habit of naming these credentials so you

know what they are. And now you are

connected to Open Router and you have

access to tons and tons of different

chat models, which is why I like to use

Open Router and keep all my billing

information in one spot. So for the sake

of today's demo, I'm just going to go

with a 4.1 mini. It's got a good balance

of cost and speed and power and

everything. So, not saying that this is

the most optimized model for this use

case or anything, but for the demo,

let's just stick with that. Keep it

simple. All right. Now, if I go back

into the text classifier and I hit

execute step, it's actually going to

work. And AI just read this email. And

look what it did. It output it in the

customer support branch. You can see we

have one item. And then finance,

billing, high priority, and promotion.

We have no items. So you can see now it

sent it down the right path just like we

talked about in our wireframe of having

the AI send it down a particular path.

All right. So now it's up to us to

figure out what do we want to do now

that it has been labeled as customer

support. And real quick I'm just going

to click on this node and hit P which is

going to pin that data. So if we

accidentally lose it we don't have to

rerun AI and spend more money. So let's

just say what we want to do first is

label this email in Gmail as customer

support. So, I'm going to click on the

plus. I'm going to go ahead and type

Gmail. And now we have a list of actions

that we can take. I'm going to go ahead

and look for label actions. Create a

label, delete a label, get many. No,

we're going to look for message actions.

And it right here we have add label to

message. So, I'm going to click on that.

Then, you have to make sure you have the

correct credential connected. So, not my

True Horizon email. We're doing this

demo email. You can see the resource is

a message. The operation is add label.

And then it's asking for a message ID.

So all we need to do is grab the message

ID of the email that triggered this

workflow. So we can actually grab it

right here. It's coming from the text

classifier. But just to show you guys

where that's actually coming from, I'm

going to close out a text classifier.

I'm going to open up the Gmail trigger.

And right here now you can see I have

ID. This is the actual message ID of the

one that triggered this workflow. And

once again, this is a variable. So this

will change every time. And now that we

have that, I'm going to look at this

label names or ID and I need to choose

which one I want, which we know is

customer support. So, I chose that. And

now what I'm going to do is execute the

step. But before I do that, I just

wanted to show you guys the email in my

inbox is right here. Nate Hurkeleman

issue logging into dashboard back in

Naden. We're going to hit execute step.

We're going to get a success message

over here. And then if I go back into my

inbox right here, we just saw that got

labeled as customer support. and it is

now down here in the customer support

label. Cool. Once again, just going to

pin that so we can keep that operation

saved. And now let's continue moving on

down this customer support branch. Okay.

So, according to the wireframe, what we

would do after we label is we'd have an

AI agent that creates a response and

then we'll have it send that response

back. So, what we're going to do is

we're going to click on the plus. I'm

going to type in AI agent. And now we

have an AI agent that we have to

configure. So, what we have to do first,

just like we did in the text classifier

node, is tell it what it's looking at.

We're not looking at a connected chat

trigger node because that doesn't exist.

So, what we're going to do is change

this to define below. And then I'm going

to do basically the exact same thing as

we did earlier. I'm going to open this

up. I'm going to say here's the subject,

here's the body, and we're going to go

back to the Gmail trigger, and we're

going to drag in those exact same two

variables so that the agent that's going

to respond to this email actually has

context of what the email is saying. So,

there we go. We got our subject and our

email body in this agent. Now, next

thing to do is we have to prompt this AI

agent what to do. So, it's looking at

the subject in the body, but it doesn't

know does it have categories to push it

into like earlier, what does it have to

do? And so, that type of information,

stuff that is like instructions or ro or

things like that, we put that down here

in the system prompt. So, I'd click on

add option and I'd click on system

message. You can see by default it says

you are helpful assistant. We're going

to go ahead and get rid of that and I'm

going to change this to an expression

just so we can make this full screen.

And so this is where you would put in

any information like your customer

support policies or how you would want

an agent to respond to people if it was

dealing with customer support related

issues. But what I'm going to do real

quick is I'm going to go to chatbt and

I'm going to whisper flow and say, can

you please create me a concise system

prompt in a markdown code window. This

is for an expert customer support

assistant for AI automation society.

This agent's name is Nate's AI

assistant. So it will always sign off

emails with Nate's AI assistant. It's a

friendly assistant that responds to

users queries with the correct

information and if it can't find the

answer to something, it will tell the

customer to reach out to nate

herk88@gmail.com.

Okay, so I got that brief system prompt.

Like I said, I'm just doing this quick

for the sake of the example, but we have

a bit of a role. You are Nate's AI

assistant. Um, you are a professional

customer support representative for AI

Automation Society. You always sign off

every message as Nate's AI assistant.

And then you've got some instructions

right here. So, like I said, you would

customize this for your specific use

case, for whatever you want to happen

down that branch. And now, once again,

before we test it, we have to give this

agent a brain. So, what I'm actually

going to do is we know that we have Open

Router down here connected to GPT4.1

Mini. And I'll actually just name this

GPT4.1 Mini just for organization. And

if I want to use the same chat model, I

could just take this and drag it into

there. But if I wanted to change it, I

could then of course connect my own

separate open router and I could change

the model. So let's actually just say

for this case we want to use claude 3.7

sonnet. Sure. And so now we have claude

3.7 sonnet in the customer support

agent, but we're using 4.1 mini to

classify our emails. So what I'm going

to do is hit this little play button

above the agent to actually run it. You

can see right now it's using its brain.

It's reading that incoming email and

it's going to create an output for us.

And if I go ahead and click in to read

it, it basically says, "I'm sorry to

hear that you're experiencing login

issues with your dashboard. This is

definitely frustrating, but we can help

you sort this out." And then it lists

off some options that could be causing

the error. And then you also see at the

bottom, it says, "If these steps don't

resolve the issue, please email Nate

directly at this email." And then it

signs off Nate's AI assistant. So that's

perfect for now. And now I'm just going

to real quick pin this so we don't lose

that data. And the last thing we have to

do on this customer support branch is

have it send back that email in the same

thread to the email that kicked off this

whole process. So I'm going to click on

the plus right after the agent going to

type in Gmail. We're going to do a

message action which is right here.

Reply to a message, not send. We're

replying to one. So I'll click on reply.

And then we have a few things to do.

First of all, make sure that you're

connected to the right account. Second

of all, we need the message ID. So, I'm

actually just going to go all the way

back to the Gmail trigger and grab the

ID right there. Once again, email type

can be HTML or text. I'm going to go

ahead and switch this to text. And then

we have the actual message itself, which

we know is the output of the customer

support agent that we just built

together. So, I'm going to drag this

output into the message field. And then

the last thing, which is a little extra

tip, is I'm going to add an option that

says append edit an attribution and turn

that off because by default, this would

send an email. And then at the bottom

what it would say, this was

automatically sent by Naden. So I just

want to turn that off. And now I'm going

to go ahead and hit execute step. This

is going to give us another success

message over here. You can see it says

sent. And I'm going to go into my email

right here. This is the email that we

had that was labeled as customer

support. So I'm going to click on it.

And now you can see we have a reply

which says, "I'm sorry to hear you're

experiencing login issues with your

dashboard. This is definitely

frustrating, but we can help you sort

this out. Here is the actual email that

it generated for us." And then we also

see it signs off with Nate's AI

assistant. Now, I don't like that

there's bold formatting right there, but

we could once again just go ahead and

prompt it to change that. The reason why

it did that is because in the prompt,

you can see it explicitly does have bold

like this. So, I would literally just

get rid of those bold things. And now,

it wouldn't sign off like that anymore.

But anyways, what I'm going to do real

quick is unpin all of this data. I'm

going to go back into our sample emails

right here and I'm going to send one

more customer support example and we're

going to run the whole flow in basically

one fell swoop and make sure

everything's working right. All right,

so here's the moment of truth. I'm going

to hit execute workflow. This pulls in

an email. It's being classified. It got

sent down the right branch, customer

support. It got labeled. The customer

support agent is now making that message

and it got sent off like that. So now

that took maybe 15 seconds and what

normally would have taken you at least a

few minutes to read the email, respond

to it, and classify it just basically

happened instantly. So let's go back

into the email. Let's see. This is the

new one that we just got. Let's go ahead

and refresh. We can see that it is now

labeled as customer support and there's

a reply. So I'll click on it and you can

see we got, "Hello, I understand you're

experiencing some issues on with your

CRM integration, not syncing leads, blah

blah blah." Um, here are some things you

can try and you can also reach out to

Nate directly. And it also signed off.

There we go. With no more bold because

we fixed that system prompt. So, as you

can see, we have now basically set up

the customer support inquiries that may

be coming in. And now, all you would

have to do is continue to make this

prompt better and better so that the

agent answers better. And if you needed

to, you could even connect it to some

sort of database that has more external

knowledge. We're not going to dive into

that in today's tutorial, but I've done

plenty of stuff with Rag on my channel

if you want to check out some other

videos about Rag. I'll actually include

my full beginner's guide course right up

there. I'll tag it right there. So

later, if you want to come back and make

these agents even smarter, you can go

ahead and watch those RAG videos. So the

process is basically the exact same for

all of these other branches now. So

we're not going to build out all of them

fully, but I will show you maybe a few

different things that you could do for

different branches. Okay, so I just sent

myself a new email. This one is the

subject incorrect charge on invoice. And

of course, this came from the sample

email database right here. So, what I'm

going to do is in this workflow, I'm

going to hit execute. We're pulling that

email. And hopefully, this one gets

classified as finance and billing, which

it did. As you can see, it says one

item. And so, now we need to set up

what's comes next for the finance and

billing flow.

So, the first thing I'm going to do

right away is just have it get labeled.

Real quick, just going to pin this data

so we don't have to come back if we end

up losing it. And what we can actually

do is just copy this node from up above

because we know this node is adding an

email and we know that the variable

should already be set up correctly as

well as you can see. Now all we'd have

to do is just change the label. So

instead of customer support, we want

this one to go down. I spelled finance

wrong. Down the finance and billing um

label instead. So we can see in our

inbox there it is. I'm going to go ahead

and execute this step. And now in our

inbox, this should change to finance and

billing any second now. Maybe I just

need to hit refresh. There we go.

Finance and billing. And we've got one

new email down there. So maybe for

finance and billing emails, we just want

to notify the finance and billing team.

And maybe we just want to give them a

quick summary of what just came in and

who sent it. So when I click on this

next button to send them a notification,

this could be in Slack, this could be in

ClickUp, this could be even on WhatsApp,

this could be on Telegram, whatever we

want. And I don't want to just add more

tools into today's tutorial just to keep

it simple. So, we're just going to do a

Gmail notification, but the point is

there's so many different integrations.

So, however you want to send or receive

notifications, you can set up here. So,

what we're going to do in this one is

send a message. And you may be

wondering, we don't have any message to

send because there's no agent making

one. We're actually just going to save a

few bucks here. Well, not really a few

bucks. The AI models are a lot cheaper

than you think. But, we're going to save

some AI processing power and basically

custom make this message with variables

instead of AI. So, I'll show you guys

what I mean by that. I'm going to open

up this configuration. You can see we

have a two, a subject, email type, and a

message. So for the two, what I would do

is I would notify whoever is in charge

of billing. So for the sake of the

example, I'm just going to put billing

atample.com.

For the subject, I'm going to change

this to an expression. And we can

basically have a mix of variables and

also things that we hardcode in. So just

like you saw earlier, for example, in

here, we always hardcoded in subject,

but then the variable changes every

time. So that's basically what we're

doing in this node. So I'm going to say

new billing email. And then I could

basically just go ahead and put the

person's name of who sent this email. So

I'm going to go back to the Gmail

trigger. I'm going to scroll down until

we find the actual name of the person.

So we have the address right here. We

have the name. Okay, here's the name of

the person that sent the email. So I

could go ahead and you can see it says

from and then there's an address and a

name. So I'm going to drag this in. So,

it basically says new billing email from

Nate Herklman. And then the email type,

I'm going to once again change this to

text. And then for the message, we're

going to make this an expression and

open this up full screen. And I'll show

you guys what I want to put here as like

a sample message. So, I could start this

off by just saying like, hey, billing

team, an exclamation mark. And I could

say we received a new billing inquiry at

and I just want to put the time real

quick. So, the way that I could get to

the current time is I could do two curly

braces. You can see that some stuff pops

up, but I'm just interested in this one

that says dollar sign. Now, I'll click

on that. You can see we have a really

ugly format right here. And this is the

current date and time, though. So, what

I'm going to do is I'm going to type dot

format. Now, what this does is it puts

the date right there. So, 2025, October,

I almost said August, October 23rd. But,

I could format this even farther. So, I

want the actual current time. So, I'm

just going to put a space. I'm going to

do h, which gives us the twodigit hour.

I'm going to put a colon. I'm going to

do m, which gives us the twodigit

minutes. And then, if I go ahead and put

space a, it basically says am. PM. So,

now it's going to always come in with we

received a new billing inquiry at

today's date. And here's the time. I'm

going to put a period and then two

enters to go down again. I'm going to

say it is from and then we're going to

once again go all the way back down here

to grab the name of the person that sent

it. So there we go. Just drag the name

and I'm going to say you can email them

at and then I'm going to also just drag

in the email address. And then just to

give a little more context, we will just

say this is the subject of their email.

And I'll put a colon. And then we just

have to go back up to find the subject

which is right here. And so now

whoever's in charge of billing or maybe

you send this like whatever notification

you send they're getting what time it

came in, who it's from, who to email,

and what the subject is. So that the way

they can see right away if it's urgent

that they can just hop on it. And then

maybe I'll just end with a friendly

cheers. And then we're good to go. So if

I go ahead and exit this and I execute

this step, this is going to send that

email. So we'll hop back into email and

we'll refresh this real quick. And we'll

now see that we sent this. Well, it's a

fake email, so we see address not found,

but we now get this email that has all

of those variables filled in. And it's

always going to have those filled in

every time. And we didn't even have to

use AI for that. Now, notice this is

what it's going to do if you don't turn

off that append, end, and attribution.

So, we would basically just go back into

here, add option, append, end, and

attribution, and then turn that off. And

now, we would be good to go, and we

wouldn't get this little teeny message

down there. Okay. So, hopefully you guys

are starting to understand the pattern

here. So, I'm going to speed it up a

little bit. Let's do high priority. I'm

going to copy the label thing once again

because we already know this is set up

the same. And I'm just going to go into

the label and instead of finance and

billing, this one is going to be a high

priority label. And while we're here,

I'm also going to do that real quick one

more time down here for promotion.

Everything should be set up already. We

just need to change the label from high

priority to promotion right there. And

so now, if I save this, we have all of

our labeling done. And we just have to

configure what comes after these labels.

And so I'm real quick going to go in

here and I'm going to shoot myself an

example high priority message and then

I'll be right back. All right, I shot

that message off. We're going to retest

this workflow. It should classify it as

high priority. And there we go. Boom.

High priority. So the email was right

there. If I hit refresh, we now have it

marked as high priority. And what we're

going to do this time is we're going to

do something similar as up here, but

instead of replying, we're going to have

it send a draft instead.

So, I'm going to go into this system

prompt that we made and just say, "Can

you make me another one just like this,

but this time instead of Nate's AI

assistant, this will be called Mr.

Doomsday." And it is not friendly. It's

actually very rude. Okay, so we have you

are Mr. Doomsday, an expert, but

brutally honest and rude customer

support rep. And one more change, Mr.

Doomsday won't be handling customer

support. He will actually be handling

high priority emails. Okay, you know

what? I don't even like the system

prompt as much. Let's just copy this one

because I think it might result in some

funnier outputs. We will see. Anyways,

I'm going to click on the plus after the

label. We're going to add another AI

agent. And we have to once again say,

what are you looking at? So, we're going

to click on define below. I'm going to

change this to an expression. And we

know that we want it to once again look

at the subject and look at the body.

Where do we get these variables from? If

you said Gmail trigger, you would be

correct. You could also get it from text

classifier, but sometimes it's easier

just to look at the source. And so I'm

going to go in here. We're going to

scroll all the way down and find the

subject, which is right here. And we're

also going to find the text, which is

right here. And now Mr. Doomsday will be

looking at the right information. I have

to open up the add option and click on

system message. And I'm going to come in

here and I'm going to paste in the

system prompt. Make it a full screen.

And we're just going to remove the bold

as we remember from last time so it

doesn't sign off with that bold stuff.

And now I have to give it a brain. So

I'm actually just going to bring this

into the claude 3.7 that we set up

earlier so that both the customer

support agent and Mr. Doomsday are going

to be using the same chat model which is

completely fine. So I'm going to go

ahead and hit play to run Mr. Doomsday.

You can see it's using its brain. And

we'll see the type of email that we get

back from him. Okay, so we got Oh,

fantastic. Another urgent system outage

that somehow warrants bothering me. Let

me break this down for you, genius. If

the main production environment is down,

affecting all users, I obviously already

know about it. Okay, so sarcastic. Yep.

Signs off Mr. Doomsday. Let's go ahead

and send this back as a draft because we

probably don't want that to go out to

someone and we could go ahead and make

some tweaks to it. So, I'm going to

click on the plus after Mr. Doomsday and

we're going to go to Gmail. And when I

open this up, you can see that we have a

few things. We have reply to a message.

We also down here later have draft

actions where it says create a draft.

I'm actually going to go ahead and click

on create a draft because in here what

we're able to do is add a thread ID and

then that basically lets us link it to a

specific thread. So one thing you may

notice is that right here it says to

reply to an existing thread specify the

exact subject title of that thread. And

so we're not going to be doing a

specific one. We're just going to reply

to the one that came in. But it will be

interesting because usually drafts don't

have subjects. They just kind of, you

know, keep going on. So, we'll see what

happens here. But anyways, I'm going to

drag the output in as the message that

it's sending. It's already set up as

text. And then for the thread ID, I'm

going to go back to the Gmail trigger.

And we're going to grab right here the

thread ID of that message. And then for

the subject, for now, I'm just going to

put test to see if that even shows up or

not. And I'm going to go ahead and shoot

off this draft. We get this success

message that says that it's a draft. And

then if I go into the email, there is

our high priority message. I hit

refresh. We can now see we have a draft

associated with it. And as you can see,

we don't actually need the subject

because if we were to send this off, it

would reply right there. But it's not

automatically filling in the recipient.

So, what we need to do is come back into

the create a draft node, add another

option, and then right here, it has a

two email. So, we'd click on the two

email, go back into our Gmail trigger,

we would come all the way back down to

the actual person that sent this email,

which is right here, and drag that in as

well. And we actually don't even have

the option here to get rid of the append

edit an attribution which means that it

just naturally won't come in. So I'm

going to go ahead and delete this draft.

We're going to go back out. I'm going to

now refire this node. It just told us

that it created a new draft. And if I go

into here, hit refresh, we see another

draft. And now it's actually being sent

to this person. It has our body. And I

can go ahead and send it off. And that

is why, as you can see, because it's a

reply, it doesn't matter that we have

test in the subject. The subject won't

actually show up unless you want it to

reply to a specific existing thread.

Then you would specify the subject right

there. So we've done these three. We

have a automatic message going out. We

have a message notifying a different

team. And then we have a draft going

out. So for this last promotional one,

all I want to do is I want to label it

as promotion and then just mark it as

unread cuz I really don't care that

much. So, that's what we're going to do

here. Okay. So, I just sent myself this

promotional email, as you can see right

here. New features, 25% off. We're going

to execute the workflow. The text

classifier should put this down the

promotion path. There we go. And go back

to the inbox, hit refresh. We got this

labeled as promotion. So, perfect. And

now, all I'm going to do is add a new

Gmail node right after it. I'm going to

go ahead to the message actions. And

right here, it says mark a message as

red. So, I'm going to click on that. And

all we have to do after we choose the

right credential, of course, is give it

the message ID, which is really, really

simple. I'm just going to go back to the

Gmail trigger to keep things consistent.

Drag in the message ID. And when I hit

execute step, we get a success message.

And when I come into my inbox and we go

ahead and refresh this. Now, keep in

mind, I haven't clicked on this message

yet. Refresh. It just instantly goes as

unread. So, now we don't even have

promotional stuff taking up any of our

headsp space. So, if you guys just

followed along with everything that I

did, you now have your own inbox manager

up and running and you're in full

control and should understand how you

can customize all of these different

paths based on what you want to happen.

Of course, you could have agents

responding or drafting all of these,

which is kind of what I have set up for

my own inbox system. And it would just

be a matter of coming into the prompts

and making these a little bit better, or

giving them access to information, or

honestly kind of a combination of the

both. One more thing, as I'm editing

this video, I realized that I forgot to

mention is you need to come in here and

make this workflow active. Once you

activate this workflow, it will say your

workflow will now regularly check Gmail

for events and trigger executions for

them. So, what that means is if you were

sitting here now and this workflow is

active and you got an email, you

wouldn't visually see the green lines

and the red circles and everything

spinning in this interface, but it is

actually going to be triggering in the

background and all of those executions

will show up when you click on

executions right here. If you leave this

as test mode, this Gmail trigger will

not actually be pulling every minute the

way that we have it set up. So you have

to make sure every single trigger in

edit end, not just a Gmail one, it's

only going to actually be constantly

running while you sleep if your workflow

is active. So make sure you turn it on

if you're ready to start running the

system. Now, there's lots of ways that

you can scale this up, like I said, and

one of those ways that I would recommend

doing would be having some sort of

logger. So, what I would do at the end

of all of these flows is I would just

have this data writing back to Google

Sheets that just says what time an email

came in, which path it went down, and

like maybe what the reply was or what

what the end result was. And the reason

for that is then you can kind of just

monitor like a Google sheet or an Air

Table or wherever you want to log your

actions. You can see everything that

your agent's doing. And of course, you

could see that by going into the

executions here and just looking at

every time it's ran. But it'd be so much

better to have a clean front-end

interface where you can identify

patterns. You can see how much stuff is

happening. And then you know if you need

to come back into the system and make

tweaks. You can figure out if you need

to make tweaks here, if you need to make

tweaks here, if you need to make tweaks

here, or if you need to add certain

elements in areas that you don't even

have them yet. So, like I said, I wanted

this one to be super simple to follow

along with, and we kept this one simple,

but there's always ways to scale up. And

I've got tons and tons of videos on my

channel if you're interested. I'll also

be attaching this exact template in the

Free School community with this Google

guide with all these sample emails if

you want to just look at my prompts or

look at some stuff of how I set this

thing up. And once again, if you're

really still interested in making the

system better and you want to master

NAND automations, then definitely check

out my plus community. The link for that

is down in the description. We've got a

great community of over 200 members who

are building with Naden, learning with

NAND every single day and a lot of these

people are actually starting to make a

lot of money from their NAND skills. So,

we've got a full course in here. We have

Agent Zero, which is the foundations for

beginners for learning AI automation.

We've got 10 hours to 10 seconds where

you learn how to identify, design, and

build time-saving automations. This is

where we have the process mapping and

the wireframing like I kind of alluded

to at the beginning of the video. And

then for our annual members or if you've

been in the community for six months,

we've got oneperson AI automation agency

where we basically dive into how do you

actually start to make money from this

kind of stuff and lay the groundwork for

building a scalable business with AI

automation. On top of that, we have one

live call per week where I answer all

your guys' questions. We get into some

really fun discussions. So, I'd love to

see you guys on those calls. I'd love to

see you in these communities. But that

is going to do it for today. If you

enjoyed the video or you learned

something new, please give it a like. It

definitely helps me out a ton. And as

always, I appreciate you guys making it

to the end of the video. I will see you

all in the next one.