Free YouTube Transcribe

Video transcript

Build Your First AI Agent with n8n (step by step)

CXL · 11,731 words · 54 min read

Want to search this transcript, jump the video from any line, or download it as TXT, SRT, or VTT?

Open in the transcript tool

Full transcript

0:00Welcome everybody to the first

0:02installment of the n8n webinar week. And

0:06in this session we're going to cover the

0:08foundations and basics of n8n. And also

0:12by the end of this session you're going

0:13to walk away with a live n8n agent

0:16regardless of whether you had any

0:17experience

0:19with n8n.

0:23So quick intro about me.

0:26My name is Marconi Don Rowan and I'm the

0:28director of revenue operations at Code

0:30Cloud and Code Cloud is a subscription

0:32based platform

0:33for DevOps engineering more of IT

0:36professional enablement. I'm also an n8n

0:39trainer and facilitator at Code Cloud.

0:41So I do have a n8n zero to hero course

0:45at codecloud.com if you guys want to go

0:47and check it out. Apart from that I also

0:50run a YouTube channel called automate

0:52with Mark. Where I teach a step-by-step

0:54guide to non-technical folks most of the

0:57time

0:58you know on how to use powerful AI

1:01automations with none of the technical

1:02jargon, right? So if you want to get in

1:05touch you know I can find me on YouTube

1:07LinkedIn Instagram

1:08scan the QR code and all my resources

1:11and contacts are there.

1:15So jumping to the goal for today is

1:17really to do a basic introduction to n8n

1:20what is n8n why we're using n8n

1:23and of course the first part of it first

1:2510-15 minutes or so we're going to cover

1:27some of the theoretical basics to n8n

1:29and don't worry we're just going to

1:31cover those that you actually really

1:33need to build your first workflow. And

1:35those are important concepts for you to

1:37understand how it all works together.

1:38But right after that it's going to be

1:40very hands-on right? So we're going to

1:41make sure the session hands-on. So I

1:43hope that you guys are in front of your

1:44laptop or your system so you can get

1:46signed up

1:47grab you know all the necessary tools to

1:50build your first live agent

1:52on n8n even if again Uh, if you

1:55completely new and never heard of n8n

1:57before, you're going to walk away with

1:58that today.

2:01All right, just a quick poll before we

2:02start. I wanted to understand the level

2:04of familiarity. Uh, I see a lot of

2:06people from uh, various different places

2:09um,

2:09on how familiar you are with n8n, right?

2:11From a scale of 1 to 5. 1 being, you

2:13know, you've either never heard of it or

2:15heard of it but never signed up, never

2:17really worked on any workflows before.

2:19And 5, you know, you're really advanced

2:20and technically you shouldn't have been

2:22here cuz these are uh, we're going over

2:23things that you already know. So, yeah,

2:25if you can just type into the chat to

2:27see

2:28uh, you know, what that looks like.

2:30It'd be great. Okay. So, 1 2 3.

2:36Someone with a 4. Okay. Okay. A lot of

2:381s.

2:40All right, somebody with a 5. So,

2:43cool.

2:45All right. So, um, a lot of 1s and 2s 1s

2:47and 2s. So, um, I think these sessions

2:48are uh, going to be pretty useful for

2:51for a lot of people.

2:52Um, but of course, we're going to cover

2:54the basics first and then we're going to

2:55show a couple of more advanced flow um,

2:57on what you can build with uh, different

2:59types of tools and stuff like that. But

3:01the ones that we're going to build

3:02together are going to be the ones that,

3:04you know, have all the free tools that

3:06you can grab, you know, stuff that you

3:08don't need to put in your credit card to

3:09pay. I want to make sure that everyone

3:11can just follow along um, you know, to

3:13build. So,

3:14let's jump right in.

3:18All right. So, what is n8n, right? Um,

3:21so, n8n, if you don't already know, is

3:24basically uh, in simple terms is an open

3:26and uh, free workflow platform, right?

3:29So, what it does is just connects uh,

3:30one process to another process, one app

3:32to another app, different services

3:34together in a very visual way. So, it's

3:36very visual. That's why there's there's

3:38this huge uptake about it because you

3:39can actually see it in that work. So,

3:41um, it has a huge selling point just

3:43because of that. People can actually see

3:44what it's doing. Um,

3:46in technical terms it actually runs

3:47under Node.js. Uh, with with Node.js

3:49under hood, you don't really need to

3:50know what that is, Um, but essentially

3:53that's to run the JavaScript and it can

3:55either be cloud or self-hosted. In this

3:57session, we're going to talk about the

3:59cloud-hosted one, specifically N8N Cloud

4:01because that's the easiest one to set up

4:03and also the easiest one to maintain. If

4:05there's any update from N8N, that's

4:06going to get updated first as opposed to

4:09self-hosted. If you self-hosted,

4:11technically you can self-host it on, you

4:13know, your own system or VPS, but you

4:15got to make sure that the maintenance

4:17are being taken care of if you're doing

4:19that, all right?

4:22And you might be asking, you know, with

4:24all the tools that are out there, why do

4:26we choose N8N, right? Of course, as any

4:29automation platform, orchestration is

4:31the one of the most important, if not

4:33the most important thing that an

4:35automation platform does. But of course,

4:36apart from that, N8N comes with its

4:38reliability, right? So, because you're

4:40building the node yourself or, you know,

4:43even with the help of an AI, you are

4:45able to see the processes in each node,

4:47which makes it really reliable and, you

4:49know, the control to that is very high.

4:52It's also a very scalable. If you have

4:54the workflow configured, it's very easy

4:56to scale it, very easy to connect new

4:58apps into it.

5:00Flexibility-wise, of course, you can

5:01build basically anything, all right? So,

5:02I think one of the

5:04tagline of N8N is you can automate and

5:06pretty much anything.

5:08So, if you have third-party apps,

5:10third-party tools, as long as they have

5:11API exposed, apps to be exposed, you

5:13pretty much can connect them to the

5:15workflow and essentially that gives you

5:18all the control and flexibility that you

5:20need.

5:22And with all the tools out there, you

5:23know, Cloud, Zapier,

5:26Open Claw, Manifold, you've probably

5:28heard of that.

5:29You're probably asking, why should we

5:31use N8N? Why should we consider N8N,

5:33right? So,

5:34I don't want to get into the weeds of

5:35comparing N8N, which is an automation

5:37platform, with something like Open Claw,

5:39where it is more of a 24/7 always-on

5:41agent because structurally they're

5:42different. But with all of these

5:44automation tools, you technically can

5:46put them uh position them in this

5:48sliding scale of the spectrum between

5:50control and reliability to independence

5:53and power and flexibility. So, for

5:54example, if you're using Claude uh

5:56sorry, Open Claude and Claude, um it

5:58comes with a lot of power flexibility.

6:00You can tell it something and and they

6:01could, you know, uh connect it with

6:03certain tools, etc. But, the steps that

6:06it's taking is not necessarily visible

6:08to you. And that's why, you know, you

6:09probably hear some stories about how it

6:11could go wrong, etc. and, you know, how

6:14it could actually use certain tools in

6:15the way that you didn't anticipate

6:17before.

6:18And um and N8N and Zapier, uh basically,

6:22being the automation platform, is the

6:23ones that actually you build uh from

6:26ground up. So, you actually know what it

6:28does. The guardrails are there. So, it's

6:29more deterministic than actually um

6:32tools, you know, like Open Claude and

6:33Claude is. So, so yeah, so that's

6:36probably how you want to view, you know,

6:38where these tools fall in terms of uh

6:40where these spectrums is. So, all right.

6:42So, diving into the foundations and uh

6:45the theoretical side, the concepts, the

6:46important concepts that we're going to

6:47cover today, right? So, again, I'm going

6:50to try to keep this as brief and as

6:51concise as possible. Only the

6:53need-to-know basis type of stuff, right?

6:55So, we're going to do a quick

6:56introduction to nodes, right? The types

6:57of nodes that there are in N8N. Uh we're

7:00going to also explore some AI agent

7:02architecture. Uh what it all means. How

7:03do you How do you connect it uh to an AI

7:05agent on N8N. Um and then, what are the

7:08input formats? What is the input? What

7:10is the output? What are the expressions?

7:12What is the fixed format? Um and then,

7:14also understand how N8N handle uh the

7:16data types. And these four things are

7:17super important uh because without this,

7:19even if you can automate the build, you

7:21wouldn't understand what's going on. So,

7:23N8N, uh if you don't already know, is a

7:26short form for node automation, right?

7:28So, essentially, what it is is just a

7:29bunch of nodes connected together uh to

7:31create a workflow. So, nodes are really

7:33the building component of N8N. So, in

7:35N8N, there are two types of nodes,

7:36right? You've got the trigger nodes,

7:37which there is, as the name suggests,

7:39the nodes that kickstarts the workflow,

7:41right? So, um it's an event that

7:43kickstarts the entire workflow. And then

7:45you have the action nodes, which is the

7:47node that performs a task, either send

7:49an email, store data, you know, have

7:51certain maybe even wait a certain period

7:54and stuff like that. So, those are the

7:55action nodes. So, 95% of the time you're

7:58working with action nodes, but the

7:59trigger node is usually the the first

8:00one that's going to start the entire

8:02workflow. And this is basically the

8:03typical basic AI agent architecture that

8:06we're going to work with today.

8:08Uh it comes Yeah, everything starts with

8:09a trigger node in N10. So, there's a

8:11trigger event that we define. And the

8:13trigger event can be on schedule, it

8:14could be let's say in every day 9:00

8:16a.m. or every week on a certain day, or

8:18it could be a chat trigger, right? You

8:20chat with N10 through Telegram or the

8:22native chat interface within N10, or

8:25form subscription, for example, right?

8:27So, all that data goes into what we call

8:29the AI agent node.

8:31And in N10, the AI agent node will then

8:34have an output into the next node. And

8:37this next next node could be another

8:39loop back to, you know, how you sent the

8:41message to the AI agent. Let's say if

8:43you sent it with Telegram, it may loop

8:45it back to Telegram as a reply. Or if

8:48you don't want it to do that, if you

8:49want it to do a certain action, that

8:50could be the case as well, right?

8:53But an AI agent node in N10 is actually

8:55you can

8:56you can think about it like an empty

8:58shell, right? If you don't connect it to

9:00three things, it's not going to be an AI

9:01agent. So, these three things are the

9:03LLM, sorry, which is you can view that

9:05as the brain of the AI agent. So, AI

9:07agent needs a brain.

9:09Um and that could be OpenAI, that could

9:11be Claude, that could be Ollama, could

9:14be Google Gemini, right? But all of

9:16these require API access token to

9:18connect it to the AI agent. And the

9:21second thing that you need to connect to

9:23an AI agent is the memory, right? So,

9:24this creates the persistent memory that

9:27an AI agent needs to be able to

9:28understand, contextualize the

9:30conversations or the flow events that's

9:32going into the AI agent.

9:35And third, you're going to connect it

9:36with the tools, right? That's really

9:37what's making the AI and AI agent is

9:40being able to leverage a tool, right?

9:42So, that tool can be anything from Gmail

9:44to Google Docs to, you know, Tavily to

9:48Firecrawl scraper, all sorts of stuff.

9:50So, again, like I said, in N8N, as long

9:53as the tool is exposed with API MCP, you

9:56can actually connect it to the AI agent.

9:57Although, a lot of tools these days

10:00already have their own community notes

10:02created within N8N. And again, to

10:04connect this uh to, you know, all these

10:06tools, you need either an OAuth or some

10:08form of API access to connect them to

10:11the tools, okay? So,

10:12the uh basically the architecture that

10:15we talked about, this is the way that is

10:17represented within N8N, right? Almost

10:19visually similar to what we just talked

10:21about, having a trigger node, which is

10:24right over here.

10:26Let me just turn on the laser pointer.

10:28Having a trigger node,

10:29oops.

10:30Having a trigger node going into the AI

10:32agent, and the AI agent having connected

10:34to a certain type of model, a memory,

10:37a tool, and then having that response

10:39set to another node. Now, quick one on

10:41the default run logic, right? So, no

10:44matter how extensive your workflow is,

10:47the way you can think about it is that

10:49it's going to do operate or it's going

10:50to process node by node, right? So, each

10:52node is going to be processed, it's

10:53going to run before the next node,

10:56you know, fetches the input and then

10:58process it and then

11:00kicks out the output. And we'll take a

11:01look at what that input and output looks

11:03like in a little bit. But I just wanted

11:05to clarify that if you have a branch in

11:08a workflow like this,

11:09the logic to run this is always first

11:11it'll it'll run the one at the top, and

11:13then it'll run the one at the second,

11:15always in that order, and they will

11:17always try to run both of them,

11:20you know, one after the other. Okay, so,

11:22with every node, when you open it up,

11:23and I know this is an AI agent node, but

11:25with every node, the concept is the

11:27same. We have an input

11:29and an output, right? So, the input is

11:31basically the data that's going into the

11:33node.

11:34And that is uh represented in three

11:36different format. And also the output is

11:39also represented in in three different

11:41format in terms of schema, table, and

11:43JSON, right? So,

11:46the schema uh template looks something

11:48like this. So, you have a variable key,

11:50and then you have a value, right? So, uh

11:53it's a key-value pair, and there's a

11:55reason why you want to keep this in mind

11:57because later on I'll show why that's

11:59important when it's uh when you're using

12:01it to drag and drop into the next node.

12:04And then there is a second version of

12:06that, which is a table format. And this

12:08is basically representing the same

12:09information, right? So, the information

12:11that you're seeing here on the left,

12:12schema, is just represented in a table

12:14format here. As you can see the the

12:16information or the data is exactly the

12:18same, uh which is chat input hello over

12:21here.

12:22And then

12:23the raw format is actually JSON, uh

12:25which is what's happening behind all of

12:26this. Uh in fact, the JSON format is

12:29basically the basis for that for any

12:30end-to-end workflows, right? Um so, you

12:32can toggle on any and you can toggle

12:34between these three, but essentially

12:35just to note that uh these are

12:37representing the same thing uh when

12:39going into the input and and output

12:41panel.

12:44So, in the input panel, as you can see

12:45here, you can later on uh we'll we'll do

12:48this live, but you can drag um you know,

12:51a certain variable into

12:53uh a prompt into the AI agent, for

12:55example. And the reason why this is

12:57powerful is because uh you can see see

12:59here there's a fixed and expression

13:01toggle.

13:02If you toggle this to the uh to a fixed,

13:04what happens is that uh and you know,

13:06and you type in hello, what happens is

13:08that this is going to be hardcoded every

13:10time uh when the AI agent runs. So, the

13:12AI agent's going to take hello as a

13:14prompt every time there's a run of the

13:15workflow. And that's why most of the

13:17time you don't want it to be a a fixed

13:19value unless you intentionally want that

13:21to be. Uh what you want to do is drag

13:23the chat input variable, drop it into

13:25here, and uh what that does is basically

13:27makes it a dynamic value. So, every time

13:30it runs, let's say the second

13:32uh thread of the conversation is no

13:34longer hello, it's how are you, this

13:35will change into how are you um

13:37automatically. All right, so I promise

13:39this is the last few slides on the

13:42conceptual side and then and then we'll

13:44jump right into the the hands-on

13:46session. But, data types are super

13:47important in n8n and in n8n there are

13:50five different data types, right? So,

13:51we've got the strings, we've got the

13:52numbers, we've got boolean, arrays, and

13:54object. And what are these things?

13:56Right, so string is basically a string

13:58of characters or alphabets, right? Like

14:00hello there. And then numbers are

14:02basically numbers, right? They could be

14:04They could be whole numbers like these

14:06or decimals or negative. And boolean is

14:08just either true or false, it's kind of

14:10just an expression. And arrays is

14:13basically a series of uh data values,

14:16right? This could be uh in strings or in

14:18numbers.

14:19And objects are basically next nested

14:21data and nested values, um which is a

14:24combination of all of that, all right?

14:26So, um you don't need to know about that

14:29too deeply, but just try to understand

14:31it uh because there are different data

14:33formats, a lot of the times the errors

14:34that you encounter while building n8n

14:36workflows because uh if a certain node

14:38is expecting a string input and you're

14:41inserting an object, that's when

14:43typically the error happens. So, it's

14:45important to note um you know, the

14:47different data types as we go along uh

14:49building these workflows.

14:52Okay, uh last but not least, uh

14:54community nodes. These are nodes that

14:56are uploaded and maintained by the tool

14:58providers. So, for example, this is, you

15:00know, a Firecrawl node, which is uh not

15:04uh which is um you know, uh created and

15:06maintained by Firecrawl, uh which is a

15:09uh

15:10uh tool that crawls the, you know,

15:11internet internet and scrape a certain

15:13page. Um you know, actually very useful,

15:16so

15:16um just to note that with n8n cloud, the

15:19community nodes is automatically

15:20uploaded, but if you are self-hosting,

15:22then there are a couple of steps to do

15:23uh for you to set it up. Cool. All

15:25right, so those are the concepts, the

15:28important concepts that you need to know

15:30to build your first end-to-end workflow

15:32and agent, right? So, there are two

15:34things that you need to have in order to

15:36build your first end-to-end

15:38workflow. And what you need is you need

15:40to sign up on n8n.

15:42And the second thing that you need to do

15:43is you need a Google AI Studio. And the

15:46reason for that is we're going to grab

15:47the LLM API keys from Google AI Studio.

15:50And it's one of the

15:52you know, it's one of the rare ones that

15:53gives you

15:54free credits without signing in putting

15:56into putting in your credit card into

15:58the system. So,

16:01I'm just going to give you guys maybe 2

16:04minutes to sign up and

16:06I think to recommend I'm not sure if the

16:08link has been shared with the audience.

16:13For n8n, if not, you can you guys can go

16:16to n8n.io and

16:19um

16:19you basically go to that page. All

16:21right, okay.

16:23So,

16:24I am going to walk through step-by-step

16:27on how to get signed up. I know most of

16:29you already probably signed up before or

16:31maybe you're already in the process of

16:32signing up, but if you haven't, go to

16:35n8n.io or the link that's been shared.

16:38And hit get started. And Tarek, I just

16:40want to make sure that I'm sharing the

16:41n8n page. Can or somebody can somebody

16:44from the CXL side confirm that

16:47I'm sharing that. Okay.

16:49Yeah, I can see it now. All good.

16:52All good. Okay. So, here it says, you

16:54know, company email, but technically you

16:56don't need to. It can be

16:58your Gmail.

16:59Doesn't matter, right?

17:03Submit that and then it's going to send

17:04you an email

17:06to so to your inbox. So, what you want

17:08to do is just uh

17:11make sure that

17:13you grab that. You can either can the

17:15link or just

17:17copy it

17:19and go back to

17:21the page, paste that here.

17:25on

17:26and

17:29Let's do random one there and then here.

17:35Okay, so the full name, password, you

17:38know, you guys I can fill that in

17:39accordingly, but account name, uh again,

17:41you can just name it based on whatever

17:43you prefer. Um click on verify that

17:47you're human and start the free 14-day

17:49trial. Again,

17:50um no credit card will be necessary.

17:53And just give it

17:55a couple seconds and it'll ask you some

17:58onboarding question there, right, which

18:01you know, you can answer accordingly.

18:03Team you're on, let's say marketing.

18:07Configuring API authentication.

18:11How did you hear about

18:12YouTube.

18:15Okay, so it's going to ask you to invite

18:17team members to your workspace and you

18:18can skip that for now.

18:23And just give it maybe 30 seconds or so

18:26um to spin up the workspace for you.

18:31All right, there we go. That's ready.

18:32So, I'm going to hit

18:34start automating. There we go. So, it's

18:37automatically spin spinning up a work

18:39flow for me immediately. But, if you are

18:43stuck at the overview page, what you can

18:45do is you can just hit the plus sign

18:47here

18:48and you're going to go to workflow and

18:50we're going to hit personal.

18:53And here, you're going to spin up your

18:55first workflow, right? So, we're going

18:56to add the first step here.

18:59All right, and I'm just going to pause

19:00here for a moment, just make sure that

19:02everyone that's signing up has

19:04successfully signed up.

19:05And if you have can you just

19:09type in yes in the chat and make sure

19:11that you know

19:13we're all on the same page.

19:16Great. All right.

19:18Cool. I see a lot of yeses.

19:20All right. So, going to go ahead and

19:22here you're going to hit the plus sign

19:23in the middle. So, this is where you're

19:25going to add your first trigger note,

19:26right? So, again, there are a bunch of

19:28trigger notes here they can pick from.

19:29You can trigger manually which means you

19:31it's just only good for you know test

19:33execution when you have to click on it.

19:35Again, you can hit on schedule which is

19:37you know scheduling every day, every

19:39hour. But, what we want to do here,

19:41we're going to connect it with a chat

19:42trigger, right? So, here if you type in

19:44chat

19:45and click on chat there's under trigger,

19:48there's an on new chat event, right? So,

19:50we're going to choose that

19:52and this is going to pop up, right? So,

19:55we can hit test trigger here.

19:58And this panel should pop up right

20:00below. This is a panel that is available

20:02if you are doing the N and N native

20:05chat. So, here we just going to type

20:07hello, all right?

20:08The reason why we not do that is we want

20:10to populate this node, right? So, what

20:11does populate mean? It basically

20:13fills in the input of the node and also

20:16in this case because it's the first

20:18node, it only shows the output which is

20:20the chat input hello.

20:22All right. Okay. So, now this is the

20:24first node, the very first node in your

20:26N and N workflow. So, what you want to

20:27do next is to build the next node. So,

20:29in the next node, what we want to do is

20:31hit the plus sign here and we're going

20:33to choose the AI agent node here. Sorry.

20:36We're going to hit AI agent.

20:40And this should pop up

20:42on your monitor.

20:44And here it knows already that you're

20:46connected to a chat trigger node. In

20:48this case, it is a chat trigger. So,

20:49we're going to leave it as that. And

20:51under prompt, it also knows that it also

20:54knows to take the chat input. So, you

20:55don't even need to drag and drop

20:56anymore, right? So, just to show you

20:59guys, you know, if let's say you want to

21:00define it yourself, you can hit that and

21:03now you can actually hardcode, you know,

21:05hello or you can toggle it to

21:07expression,

21:08and then you can drag the chat input and

21:10drop it over here. And you can see that

21:12below it says hello, which is reflective

21:14of what the value is here, right? So,

21:17this is how then you can dynamically

21:20put in any value that's coming in, and

21:22you can pick and choose which input data

21:24that you want to put in here, all right?

21:26So, I'm going to toggle this back to

21:27connected to chat trigger node, because

21:29it's the same, which is the chat input

21:30here,

21:31right? And then you see this three plus

21:34signs down below, the chat model, the

21:36memory, and tools. So, those are the

21:37three things that we need to make this

21:39an AI agent, right? So, we're going to

21:41hit chat model.

21:43And here, we're going to pick Gemini,

21:45right? So, literally type in Google

21:47Gemini,

21:49and we're going to click on that.

21:52And here,

21:53we're going to fill in the credential

21:54and the model, right? So, under

21:56credential, as you can see, there's no

21:57credential set up yet because it's a

21:59fresh account. So, I'm going to hit set

22:01up credential.

22:02Okay?

22:03So, there are a couple things that I

22:04need to fill in. We're going to leave

22:06the host URLs the same, right? So,

22:08everything we're going to leave it as as

22:10that.

22:11But under API key, this is where we want

22:12to fetch our API key, right? So, API key

22:15is just like a key for you to be able to

22:17be able to access the LLM. So, if you're

22:19not familiar with the term API,

22:23think of it as a way of communicating

22:26with your LLMs, right? So, how does

22:28Google know that you're authorized to

22:30use the LLMs? Because every use of LLMs,

22:33they you use tokens, right? So, tokens

22:35are our money, tokens are compute. So,

22:39you know, they need an API. So, the API

22:41key is basically the key to reflect that

22:42you are actually the authorized user

22:45for that particular,

22:46you know, to use that particular

22:47account. So, where are we going to get

22:49this API key? This where you want to go

22:51to

22:52Google Studio. So,

22:56go AI Studio.

23:02I'm sorry. So, it's aistudio.google.com.

23:07Right here, we're going to hit

23:08get started.

23:11Okay, I'm already signed in here. But,

23:13if you haven't, what you can do, and if

23:15you have a Google account, it should be

23:16a simple

23:18um

23:19it's a a simple sign-in with your Google

23:20account. And uh where you can go, once

23:23you're logged in to Google AI Studio, is

23:26to the left-hand panel here.

23:28And you can click get API

23:30key. All right. Again, I just want to

23:32pause here for a second. Just want to

23:34make sure that everybody's following.

23:36So, um again, I saw a great explanation

23:39about too fast. Okay. So, I'm going to

23:40slow down. So, just want to make sure

23:42that everyone

23:44is on the same page with the

23:45aistudio.google.com.

23:50So, if you are, uh please type in yes.

23:52If you're not, um yeah, type in no.

23:54Okay. Okay. Okay. A lot of yeses. Couple

23:57of nos. So,

23:59just want to Okay. So, a couple of uh

24:01requests to go back to the workflow

24:04and to repeat. So, I will do that. One

24:06second.

24:07So, in the workflow,

24:09the first thing you want to do, and I'm

24:10going to I'm going to just start on top

24:12here, right? Just to make sure that

24:13everyone everybody's following. If you

24:15are, I'm assuming that the chat message

24:17uh sorry, the the chat note

24:20has been, you know, you you managed to

24:22set that up.

24:23So, the AI agent note, once you have

24:25this on, you want to choose connected

24:27chat trigger note.

24:28And then, uh in the prompt, you're going

24:30to leave everything as is, right? The

24:32moment you create the AI agent note, um

24:34you're going to leave everything as is.

24:36And if you're

24:37wondering where you can get the chat the

24:39AI agent note, you can hit the plus sign

24:41here.

24:42And under AI,

24:44pick AI agent, and that note's going to

24:46pop up, right? So, this is a new one.

24:47I'm going to close this.

24:50And yeah, I'm going to delete that. All

24:52right. Um

24:54So, how I created the uh uh called the,

24:57you know, Gemini chat model is under

24:59chat model there's there was a plus sign

25:00if you click that it's going to open up

25:03the Google Gemini chat model.

25:06And right now there's no credential set

25:07up so you want to hit set up credential.

25:09This is where it's going to ask for an

25:10API key. All right, so that's the recap

25:13there.

25:15Okay, so under API keys what you want to

25:17do is you want to hit create API key.

25:20Right, so you can name your API key but

25:23I'm going to keep it as Gemini API key.

25:25Typically you can maybe name it okay,

25:27maybe name it edited and memo, right?

25:31And you're going to need to select a

25:33project. If you don't have a project

25:35you can create a project

25:37and hit create project.

25:40And some of the interfaces I've noticed

25:42that are different. Sometimes you

25:44already have a default if that was your

25:45first time creating it you may have a

25:47default API key over there.

25:49So if you already if you already have a

25:52default API key you can actually copy it

25:53and use it. Right, but uh otherwise you

25:56can create it the way I did and just hit

25:58copy API key

26:00over here with this button.

26:02And then go back to your workflow

26:04and paste the API key.

26:06And what you want to do now

26:08is hit is hit save. Sorry, I'm going to

26:10toggle this to expression so everyone

26:11can see the API key. I'm going to revoke

26:12it afterwards of course. So you want to

26:14be you want to keep this safe, right? So

26:16API key is always keep it safe because

26:17it um once someone has your API key

26:20basically can't access

26:22uh your Nello Labs all this stuff,

26:24right? So uh you want to keep that safe

26:26and then you want to hit save.

26:30Okay, so connection tested successfully

26:33means connection is live and we're good

26:35to go there.

26:38Cool.

26:39Okay, so now

26:41what we want to do the second thing let

26:43me just check to why this is showing

26:45that.

26:45Okay, so once you connect uh once you've

26:48set up the API account

26:51what you want to do is you can choose

26:53the model that you want to use, right?

26:54So, under Gemini, there's a couple of

26:56models, but in this case, I'm just going

26:58to keep it as Gemini 2.5 Flash. Uh

27:01that's good enough for the stuff that we

27:02want to do. So, we're going to leave it

27:04as that. We're not going to change

27:05anything, so we're going to toggle out,

27:06right?

27:07Now, the second thing you want to do is

27:08to connect the persistent memory, right?

27:10Because an AI agent with a brain but

27:13with no memory is not going to be able

27:15to remember what you just told it,

27:16right? So, let me just give it for

27:18example, if you type in hello now,

27:23All right. So, you can see the input is

27:24hello. The output is, "How can I help

27:26you today?" Which means it's like, all

27:28right, I can tell it my name

27:31is Marconi, and it's going to say,

27:33"Hello Marconi, it's good to meet you.

27:34How can I help you today?" I'm going to

27:35ask it again,

27:37"What is my name?"

27:45Right? So, it says, "I don't know your

27:46name. As an AI, I don't have access to

27:48personal information about you unless

27:49you choose to tell me." Right? So, it

27:51just completely forgot what I just told

27:53it. So, that's why uh the memory is a

27:55very core important part of building a

27:58uh you know,

27:59uh an effective AI agent. So, sorry, let

28:01me just redo that because I was too

28:03quick. So, what you want to do is to hit

28:05the plus sign,

28:07and there are a couple of memories that

28:09you can choose from, right? Uh typically

28:11Postgres is is a good one, but here,

28:13because we want to keep it simple, uh

28:14this is the first workflow you're

28:16building. So, we're going to hit simple

28:17memory because this is the native uh

28:20simple memory that N den Cloud has, um

28:23you know, and it's simple for you to use

28:25and it's quite reliable as well. So,

28:27it's actually pretty good and you know,

28:29to

28:30uh for you to be able to even build

28:32something that's pretty complex. So, um

28:34we're going to choose simple memory,

28:36and context window length is basically

28:38how many past interactions that you had

28:40with the particular AI agent, right? So,

28:41in this case, we're going to leave it as

28:43five. We're not going to change anything

28:44here,

28:45um and then we're going to toggle out.

28:49Okay? So, now you've got the LLM and the

28:52simple memory connected. The last thing

28:53we want to connect is the tool. So,

28:55we're going to hit the plus sign here

28:57under tool.

28:58And we're going to search for Gmail, all

29:01right?

29:02And here the two Gmails, we want to make

29:04sure we are choosing the right one.

29:06Um, so human in the loop is if we want

29:08to ask approval through Gmail, but what

29:11we want to choose now is not human in

29:12the loop, it is the Gmail tool. So, this

29:15is the second uh one if you type in

29:17Gmail. So, just make sure you choose the

29:18second one here.

29:20And once you choose that, this is going

29:22to pop up.

29:23And uh with Gmail now, it's super easy

29:26if you are on a Google Chrome, you can

29:27hit sign in with Google.

29:34Okay? And then it's going to ask you

29:37for, you know, a credential pop-up is

29:39going to show.

29:42And then you're going to select all the

29:45um, permissions that it's asking for.

29:47And you're going to hit continue.

29:52So, that's pretty much it, right? So, it

29:54it says here credential created

29:55successfully, your account has been

29:57connected successfully.

29:58So, now we're not going to do anything

30:00because we know that we are connecting

30:03this Gmail because we want the agent to

30:05be able to use the Gmail to send emails

30:07out. So, the operation send is correct.

30:09We want everything else is set up

30:11correctly, we're going to leave it as

30:12that. Now, to is basically who we're

30:15going to send this email to. This we're

30:17going to let the AI

30:19be able to decide this by just clicking

30:21on this blue button here, this blue

30:23sparkly button. Once you click that, it

30:25will say define automatically by model.

30:27So, the model is going to decide who to

30:29send it to. Now, that may sound scary,

30:31but it's going to follow your strict

30:33instruction. If you tell it,

30:34um, you know, send it to uh

30:36automatewithmark@gmail.com,

30:37it's only going to send it to

30:38automatewithmark@gmail.com. It's not

30:40going to send it to uh anybody else,

30:42right? Same with subject, we're going to

30:44let it be able to decide what the

30:45subject line is. And then message, we're

30:48also going to let it decide what the

30:50message is. All right, so we're going to

30:52you know, click on the blue button for

30:55these three here. Email type we're going

30:56to leave it as that.

30:58And that's it. We're going to toggle

30:59out, right?

31:01Okay.

31:02So, now that we have all these tools

31:05connected, I'm going to take a pause

31:07here again. Just want to make sure that

31:08everyone is following.

31:10Um I see somebody saying that they don't

31:13see the pop-up. So, it might be because

31:15I'm the screen share thing, but it's

31:17basically just a simple Google

31:18credential authentication pop-up.

31:21So, yeah. So, if you guys are on the

31:23same page, please type in yes.

31:26Um

31:26and then I'll move forward. Cool.

31:28Awesome. All right, seems like everyone

31:29is on the same page or most most

31:31everyone. So, here once we have all

31:34these three connected, now we can

31:35actually ask it to do something, right?

31:37So, we can ask it, "Hey, can you

31:41come up with

31:44a 300-word

31:46blog

31:48post

31:49or blog for let's say CXL,

31:53a

31:55premium

31:56B2B

31:58marketing

32:00um and education

32:02company

32:03and send it

32:06to my email at automate with

32:11Mark

32:12at gmail

32:15.com. All right, we're going to add a

32:16plus sign here just so that it does a

32:18differentiation, but

32:20All right, so we can tell it that. So,

32:21essentially I'm just telling it

32:24Uh

32:25but I cannot Okay, it's telling me I

32:26cannot write the blog 300 posts for you.

32:28Okay. So, maybe I need to connect some

32:30tools, right, to be able to

32:32come up with a good blog. Sometimes it

32:35will go ahead and write it for you. In

32:37this case, you know, it's it's refusing

32:39to. Maybe the model that I chose is not

32:41up for it. But what I can tell it, for

32:43example,

32:44can you

32:46instead, can you send an email

32:49about my appointment

32:52with Mark at 2:30 p.m.

32:56to

32:58automate

32:59with Mark.

33:01Hello.

33:03And Miller, oh.

33:05All right. So here you can see that the

33:07green sign is actually going and using

33:11the LLM twice. So it's referring to

33:13that. It's actually,

33:14you know, referring to the memory as

33:16well as it's also utilizing the

33:20the Gmail tool here to send the email.

33:22All right. So here, let's go to our

33:23email and see if we actually receive

33:25anything.

33:26So here, I actually received, "Hi Mark,

33:28this is a reminder about our appointment

33:30today at 2:30 p.m." All right. So

33:34So yeah, so right now we have

33:35successfully created an AI agent that's

33:37capable of sending email on your behalf

33:40to anybody

33:41as long as you tell it to send it to

33:43them, right?

33:44Of course, you know,

33:46when you look at this, this is a very

33:48simple workflow and you say,

33:50you know, some of my cloud agent or chat

33:52chat GPT agent might be able to do that

33:54as it is. So why are we doing that?

33:56Well, the reason is we're going to build

33:58the The reason is today we're we're

34:01building the basic foundation of a

34:02workflow of an AI agent.

34:05And later on I'll show you, you know,

34:06two or three other use cases where it

34:09gets a little bit more advanced, a

34:10little bit more complex,

34:12where it then it kind of makes sense how

34:14you know, all these things can be

34:16automated when put together. But it's

34:18just to give you an introduction, give

34:20me a feel of

34:21as to how these AI agents actually

34:23leverage tools, right? Like for example,

34:25earlier I said, you know, create

34:28a 300-word blog post and it's not

34:30confident in being able to use that

34:32or to do that. So, you can connect tool

34:34like for example, you know, Google Doc

34:35is another tool that you can connect,

34:37right? And the same thing with Google

34:39Doc, you can sign in with Google and be

34:41able to connect it, and then it will

34:43it'll be able to, you know, draft those

34:44blog posts into the Google Docs and be

34:47able to share that Google link to

34:50you know, via email, via Telegram and

34:52stuff like that, right?

34:53Um

34:54So, before I move on to, you know,

34:57demoing the other workflow, I just want

34:58to point out the difference between

35:00adding a tool here and adding a Gmail

35:04tool or Gmail node over here, right? So,

35:07you may be wondering, "Hey, you know,

35:09what's the difference between adding it

35:10as a tool to the agent

35:12uh versus adding it over here?" So, for

35:13example, I can add this node, right? I

35:15can add a send message node

35:17over here, and then it will

35:19it will it will still run that, right?

35:21So, the difference is that if I attach

35:23it as a tool, the agent has a choice on

35:26whether or not to to use the tool. So,

35:27it's going to infer from my request

35:29whether or not I'm asking it to send an

35:31email, right? So, if I say send an

35:32email, then it will infer that, "Okay, I

35:34actually need to use a tool to fulfill

35:36that request." But if not,

35:38uh you know, if I wire it up this way,

35:41what happens is that whether it doesn't

35:43have a choice, right? The workflow

35:44doesn't have a choice uh whether or not

35:45to send the email. It will always run

35:47the agent node and then subsequently run

35:49the send message. So, if you want it to

35:51always send a message via email, this is

35:53what you're going to do. You're going to

35:55connect it to the Gmail node this way.

35:57But if you want it to be

36:00uh multi-agentic, multi-tools, so, you

36:02know, not always all the requests are

36:04going to be involving sending email, you

36:06would connect it this way as a tool so

36:07that it has a choice

36:09uh to use it or not.

36:12Okay?

36:15All right.

36:16So, that is the uh I hope everyone is

36:19was able to get that up and running.

36:22Um

36:23Yeah, if you were able to get it up and

36:25running on your end, please type in yes.

36:27Just want to make sure that um

36:29Yeah, somebody also Yeah, somebody

36:31received a 300 word word blog post, so

36:33that's great. Awesome. A lot of yeses

36:35there. All right.

36:37Great. Yep, so um that was an easy one,

36:41uh you know, to get started um so that,

36:43you know, you guys have a feel on how it

36:45all works.

36:46Uh of course, you know, there are more

36:47advanced build that we can go ahead and

36:49build, but you know, this being a

36:5045-minute session, I can't squeeze all

36:52of that uh into one session, but don't

36:55worry, the next few sessions it's going

36:57to get progressively more advanced and

36:58and it's going to uh get you there uh by

37:01the end of it, right? Uh but here I just

37:03want to demo some of the few

37:04possibilities that we can we can do and

37:06we can come up with, right? Like for

37:07example, this particular uh

37:10uh workflow, and I'm going to explain

37:11each and every node uh to let you know

37:13what's happening here, right? Um this is

37:15a inbound sales, so this is a sales use,

37:18and later on I'll I'll show you another

37:19marketing usage as well. So, this is a

37:21sales usage where it will uh respond or

37:24the trigger event is actually an on form

37:26submission, right? So, this on form

37:27submission

37:28uh it can uh it's actually a form that

37:31you can embed uh to your website or to

37:33your landing page. So, I'm going to

37:35execute workflow here just to show you

37:37guys what it looks like, right? So, this

37:39is an inbound form. Again, this can be

37:40embedded into your web page uh into your

37:44landing page. So, for example, I'm just

37:45going to fill in, let's say, John Doe

37:48um

37:48email address. Again, I want to put my

37:50own email address here.

37:52The reason for that is I don't want it

37:54to randomly send it to other people,

37:56right?

37:57And then phone number, I'm going to

37:58leave it blank.

38:00So, this is a similar to

38:03let's say any inbound lead form that you

38:05have uh you know, as a business. You

38:08know, you want to know, you know, what

38:10their businesses are and stuff like

38:11that. In this case, I am uh acting as

38:14Kode Kloud, right? So, this is, let's

38:15say, how Kode Kloud inbound form is. You

38:18know, we sell to engineering teams,

38:20DevOps, um IT departments um who wants

38:23to upskill. So, let's say how can

38:25Co-cloud help? Let's say upskilling in

38:27AI skills.

38:28And let's say the person submits this

38:29form on your webpage, right? So, once

38:32they submit the form, it goes into an AI

38:34agent similar to what we've just built.

38:36But in this case, you see I didn't

38:38attach any memory, I didn't attach any

38:39tool. Why? Because this agent, I just

38:41wanted to come up with a prompt, right?

38:43So, I've given it some system prompts

38:46as a as a video prompt agent.

38:48And the reason the reason why

38:51uh why I want to I want it to create a

38:52prompt is because that prompt is going

38:54to get passed on to HeyGen, right? So,

38:57what HeyGen is

38:59I'm going to

39:00toggle it to here.

39:01Uh if you guys are not familiar with

39:03HeyGen, it is one of the tools that

39:04generate uh lifelike videos, uh

39:07AI-generated videos based on your

39:10similarities, based on uh based on your

39:12your videos, right? So, it can actually

39:14clone you or impersonate you. Uh so, in

39:16this case, I have a few avatars that uh

39:18I've created of myself. Um so, in in

39:21basically, once you've done that, you do

39:23need to be on a pay plan. Uh but once

39:25you've done that, you can grab the API

39:26keys uh in similar fashion that we did.

39:29Uh slightly more complex, which we won't

39:31get into the weeds of today. Uh but once

39:33you've done that, you basically call the

39:35HTTP request to post a request to make

39:37to make the video, right? So, HeyGen's

39:39actually processing the video.

39:41And then we're going to wait, you know,

39:43a certain period of time because it

39:44needs time to process. So, you know,

39:46it's processing the video based on the

39:47prompt that this agent comes up with.

39:49And then it goes to get the video,

39:51right? Why do I want to get the videos?

39:53Because hey, the video's been generated,

39:55right? Um so, we want to get the video.

39:57And this loop is just to make sure that

39:58if the video is not ready, it's going to

40:00go into a loop and it's going to try to

40:02get the video again and again until

40:04um you know, the video is ready. And

40:06then the if gate is if the video ready

40:08is ready, it goes into uh um another

40:12open AI model. And the reason why we

40:15want to feed it to that is because this

40:17is a caption model, right? So, if we

40:19look at if we look at at the content

40:21here, so it's coming up with the caption

40:23or the content based on what we've

40:25filled up on the form earlier here,

40:27right? So, it's taking inference from

40:29the form submission. So, what the uh

40:32prompt agent and the or the content

40:35caption agent is doing

40:37is uh sorry, not the caption agent, the

40:39email writing agent is doing is taking

40:41reference uh this guy called John Doe um

40:44based you know, from CXL and coming up

40:47with the email copies and and video

40:49contents and scripts uh in order uh to

40:53send them a a an email with a short

40:56video inside uh

40:58uh which is an intro video to get them

41:00to join or come into a demo call uh with

41:03your sales person, right? So, in this

41:04case, it's already sent an email to

41:07automate with mark@gmail. So, we're

41:08going to toggle over to

41:11automate with mark and

41:15and let's look at the

41:17and see what the email looks like,

41:19right? All right, cool. So, I should be

41:21seeing the email now. So, this is what

41:23the email looks like, right? So, um I

41:25received it 2 minutes ago, right? So,

41:28based on what I filled in, right? I

41:29filled in John Doe. Go, "Hey John, I

41:31took a quick look at CXL because I put

41:33in that my business URL is CXL and

41:36thought you might appreciate a brief

41:37intro on how Quick Cloud helps teams

41:39ramp up DevOps skills with hands-on

41:41training." Exactly how we would talk to

41:43engineering team, right? So, hands-on

41:44labs, uh real cloud sandboxes with

41:46Kubernetes, Docker, Terraform, AWS, etc.

41:49So, these are all these technical terms

41:51that the engineering team would like to

41:53hear and want to upskill in, and then it

41:55created a quick 10-second intro video

41:58here uh with the thumbnail so that, you

42:01know, people can click on it and it

42:02actually downloads the video. Okay, so

42:05that is a video that's generated based

42:07on I don't know if the audio is coming

42:08through, Tarek. It's

42:10I know you have to share screen with the

42:12audio. It happens a lot with Zoom. It's

42:14annoying.

42:17Did the Did the audio come through or

42:18no? No, they didn't come through. Like

42:20you do when you share your screen, you

42:21have to share and you have to enable a

42:23share sound as well.

42:27Okay, I don't know if

42:29But I believe you. But I trust you. It

42:31has a I know like I I faced the problem

42:33this problem so many times. Yeah, yeah.

42:35Yeah, yeah, yeah. So anyway, um

42:38assume that there was an audio. Uh you

42:40guys can see the movement and then the

42:42video, right? So I did not say anything

42:44or move like that or or or or said any

42:46any of those things in terms of the

42:48script, right? So this is all AI

42:49generated of myself. All I did was take

42:52took a 1 and 1/2 minute video on HeyGen

42:54and that generated my avatar. So now

42:57I have a workflow Okay.

42:59So basically with this workflow, every

43:00single lead that comes through the

43:02inbound form, it's going to immediately

43:04get an email that is personalized to

43:06them with a video that is talking to

43:08directly to them to their pain points,

43:10knowing what the business

43:12you know, pain points are. And you can

43:14add even layers to this, right? So right

43:15now it's very generic, you know, I built

43:17this in maybe 15 20 minutes, right?

43:20It is currently very generic. What you

43:21can do, you can add an Apollo layer, you

43:23know, data enrichment layer to find out

43:25who their managers are, you know, who

43:26their uh co-workers are and stuff like

43:29that, decision-makers, etc. and have a

43:30multi-thread channel. So there there are

43:33lots of potential that you can build in

43:34terms of workflow here

43:36uh in terms of sale, right?

43:39Cool. All right. I am mindful of the

43:40time.

43:41Uh so I'm just going to show

43:43my last uh workflow here with the

43:46marketing one. This is uh

43:48a bit more complex, but um basically

43:50what this workflow is trying to achieve,

43:52it is trying to create uh reels for

43:54Instagram, right? So um you have a

43:56folder and an edited folder with

43:59currently

44:00you know, some random pictures of uh

44:02gym,

44:04some coffee, and some croissant. So I'm

44:06assuming, you know, this is a coffee

44:08shop or a gym that is trying to create

44:10social media post, social media reel.

44:12Uh so, it goes through this node, which

44:14is a randomizer. So, it randomly picks

44:16an an image and then uploads it to a

44:18database, uh gets the real idea and the

44:21prompt, and then it posts it to Wave

44:23Speed. Again, Wave Speed generates

44:25videos from the particular image. Same

44:27thing here with what you saw earlier,

44:29wait for the image Oh, sorry, for the

44:31video to be generated. Um and then try

44:33to get the video. That's a loop there,

44:36and then goes to this thing called a Sub

44:37Magic. So, Sub Magic is, you know, uh

44:40it's an auto caption tool. Again, there

44:42are many tools that I'm covering here,

44:44but not all of them have alternatives,

44:46too, right? This is by no means like

44:47they're the best tools out there. Um so,

44:50same thing with calling any tools that

44:52that needs time to generate, you need to

44:53wait, and it goes into a loop uh before

44:55sending a message for approval.

44:57And uh once you do that, I'm going to

44:59run the workflow as I'm talking here. Um

45:01it's going to upload the reels into uh

45:05what we call Botato. Botato is a tool uh

45:08that automates your social media post.

45:10And then that creates an Instagram post

45:13at the end of it, right? So,

45:15um

45:16I'm going to fast forward on this. Um

45:18you know, if we go to our Instagram now,

45:21I don't know if it's posted yet.

45:22Typically, you need to wait,

45:24you know, uh like 30 seconds or so, and

45:27uh it will show up. Basically, again,

45:29just to recap, it's taking

45:32randomly one of these images and trying

45:34to create a social media reel

45:37uh to be posted to the to my Instagram,

45:40right? Uh again, because I don't run

45:43cafes and and gyms, I will probably

45:45delete it immediately.

45:47Uh let's see. Okay.

45:49So, yeah, so it's got a So, it picked

45:51the croissant. So, let's open that up.

45:54And

46:01Okay, it's a short one. I don't know if

46:02you guys can hear the audio, but there

46:04was an audio

46:05and the caption is important because

46:07some people on Instagram they're they're

46:09not turning on their audio, right? So

46:12So yeah, so it comes with the captions

46:13as well.

46:15Holiday vibes with a taste of festive

46:16magic etc. So yeah, okay, I'm going to

46:19delete this.

46:20Again, because this

46:22not typically what I would post on my

46:24Instagram. If you do the same demo in

46:26all of your presentation, like your

46:27followers will be like, why are you

46:28obsessed with with croissant? with

46:30croissant. So I'm hoping you're deleting

46:33Yeah, yeah, yeah, so it's

46:35I I should probably create a different

46:38Instagram account to run my demos.

46:40>> like maybe a burger like differentiate

46:42that croissant every time.

46:46That's true, that's true. Yeah, so yeah,

46:48that's a quick rundown on how it could

46:49possibly be used for you know, your

46:51Instagram Reels marketing etc. And

46:54that's all to say

46:56you know, because you are the one

46:57building this, you know what's happening

46:59on each node. You know, it's very

47:00deterministic, right? So they got rails

47:02around, it's not going to go completely

47:04off rails, completely wrong. So that is

47:06to me the power of n8n, right? All

47:09right. Yeah,

47:11conclusion.

47:12You know, what's next, right? So today

47:14you learned the very basic of n8n, how

47:17to sign up, how to set things up, how to

47:19run your first workflow. But to me I

47:21think that's where the magic is, right?

47:22Running your first AI agency, how it all

47:24works, seeing how it all connects

47:25together, seeing how it fetches, goes

47:27into your Gmail, helps you send message.

47:29I think that's the first spark that's

47:31going to help you you know, continuously

47:33learn on n8n and and by no means is

47:35super easy to learn n8n. I'm not going

47:37to sit here and sugarcoat everything,

47:38right? So but with the knowledge you

47:41actually learn also how the fundamentals

47:44of this is actually very much applicable

47:46to other types of tools, right? So the

47:48next thing you can do is n8n has an AI

47:51builder feature, right? So if you were

47:54to go to your workflow, I'm not going to

47:56toggle back, that, if you want to go to

47:57your workflow app on the right-hand

47:59corner, you'll see a little sparkly

48:01thing, and if you click on that, that is

48:02actually

48:03uh the AI help. You can chat with it.

48:05You can ask it, but there's also a

48:06builder version

48:07uh where you can ask it to build uh you

48:09a workflow for certain use cases and

48:11stuff like that, right? However, if you

48:13use that, you'd still need to understand

48:15how it all comes together. You still

48:16need to understand um

48:18you know, where to put the API keys, how

48:20to connect it to certain tools because

48:22they're not going to be able to do that

48:23for you, right? It can build you a

48:24template. It can build you a structure

48:26based on what you need, but you still

48:28need to understand, right? And the

48:29second thing is uh there's a huge

48:31ecosystem of n8n templates and creators

48:33in the marketplace. So, you can go to uh

48:35n8n templates. Some of it is paid, some

48:37of it is free.

48:38Um and uh I I also happen to be one of

48:40the creators there. Uh so, you can go

48:42check out um you know, some of the

48:44workflows that I build. They're all

48:45free, uh and some of them actually come

48:47with, you know, um a step-by-step guide

48:49on on how to put them together. Um so,

48:52yeah, go and check them out if you want.

48:55Um

48:56and then uh you know, some of the

48:58questions that you guys might have is

49:00in today's age with OpenClaw, with

49:02Claude Code, with uh Claude Co-Work, you

49:05know, do we really need n8n anymore,

49:07right? Uh the answer is yes, right? Uh

49:10the reason for that is there are some

49:11workflows that you probably want some

49:14pretty strict guardrails or guardrails

49:15or uh right? So, that's where you want

49:17to use n8n. So, for example, some of the

49:19workflows that I've used uh or that I

49:21showcase in terms of social media

49:22posting and stuff like that, I can

49:24connect that with OpenClaw. I can

49:25connect that to Claude Co-Work or Claude

49:27Code uh to be able to call that

49:29particular workflow. And that that way,

49:32I know for a fact that, you know, Claude

49:34can go ahead to my Instagram channel and

49:36and post something that I didn't approve

49:38because the only thing it can call is

49:40the n8n workflow, which requires my uh

49:43my approve my approval, right? Uh

49:45because I built it that way. So, that's

49:46where n8n could come hand in hand with

49:49all these new tools, all all new agents,

49:51um and uh you know, it's just uh

49:53extremely uh

49:55versatile tool, especially when, you

49:57know, connecting it to third-party apps

49:59that may or may not have built it off

50:01the shelf.

50:03And, um yeah, that's the new future of

50:05automation,

50:06uh you know, connecting it to somebody

50:08gigantic tools that people are rolling

50:10out. I know,

50:12you know, we are all overwhelmed. Every

50:14week there seems to be a new tool that's

50:15coming out. Um

50:17but, yeah, any then I think it's here to

50:19stay just because of how clear it is,

50:22how visible it is in terms of uh the

50:24step-by-step process that it's going

50:25through.

50:27All right, so that's the end of my time.

50:30Um and thank you all for sitting through

50:32that. I hope that was valuable, but

50:34yeah, just opening up the floor for any

50:36questions.

50:37Um you know, please uh go ahead.

50:40Awesome, man. That was like really nice,

50:42and I saw in the chat how everyone was

50:44following along. I think most of them

50:46were able to follow along and, you know,

50:48build their their first agent. In case

50:50you weren't able to catch up, uh we're

50:52going to share the recording with

50:54everyone. If you are signed up to the

50:56CXL platform, it will be available there

50:59within 24 hours. And then we're going to

51:01share the recording via email.

51:03Uh if you registered to this webinar,

51:04you'll receive an email, I think, early

51:06next week with all of the recordings

51:08from all the sessions in the upcoming

51:09few days.

51:11We are obviously out of time, but we're

51:13going to add just few minutes so maybe

51:15you can answer the question or two. Uh

51:17for the rest, you can connect with

51:18Marcony on LinkedIn, and in case we

51:20cannot answer your questions now, he's

51:23more than happy I'm uh speaking on your

51:25behalf. Are you happy

51:26>> No, feel free. Feel free to send him a

51:29LinkedIn message with your question if

51:30we're if we're not able to do so.

51:32Just few things from my my end before we

51:35answer two questions. Sorry about that,

51:37guys, but

51:38uh few things.

51:40Uh speaking of automations, speaking of

51:42AI,

51:43uh at CXL, we have lots of upcoming live

51:46sessions if this is something you want

51:48to upskill in. So from June until

51:50September we have 12 sessions around AI.

51:54So n8n, Claude Code, even vibe coding

51:57tools like Replit and lovable. So each

52:00week, each Tuesday we have a session

52:02talking about the digital marketing use

52:03cases. So you can see in June we talk

52:06about how you use n8n and Claude Code

52:08together as Marconi mentioned that one

52:10does not replace the other. You can use

52:12them hand in hand for content creation

52:15uh for SEO automation. In July it's more

52:17about paid advertising. In August it's

52:20all about uh

52:21vibe coding, satellite apps to get more

52:24leads. And in September we have a course

52:26for agencies on how you can adapt to

52:28this AI age. So you can get all of them

52:31for $499

52:33or if not all of them are relevant, you

52:36want one of them that will be $150 per

52:38course. But obviously being biased, I

52:40think if the full program would benefit

52:42you the most. Uh but if you want to get

52:45either one of them, just make sure to do

52:47it fast because after May 31st the

52:50prices will increase. So we going to put

52:53the link of this uh landing page in the

52:55chat. So feel free after this webinar to

52:57check it out. Look at the programs, see

52:59if it's a good fit for you and hopefully

53:01we can see you there.

53:03But yeah, now back to our uh normal

53:06scheduling. I'm just going to go through

53:08we have 18, 16 questions in the Q&A. I

53:11doubt we're going to be able to answer

53:13half of them even. So I'm going to pick

53:15a question or two and for the rest, like

53:17I said, feel free to reach out to

53:18Marconi.

53:20I think the most

53:22uh frequent one here is like what's the

53:23difference between n8n and Make and

53:25other automation tools like Zapier.

53:29That's a great question actually. And um

53:31not too long ago I made a video about

53:33that. Um so so with Make and uh Zapier

53:37uh well

53:38if simply put, I think uh with Zapier a

53:41lot of things and with Make to a certain

53:43extent, right? A lot of connections with

53:45third-party apps need to be off the

53:47shelf. So, if they have built a

53:49connection

53:50within Zapier, that's when you could

53:52choose

53:53that particular tool to be connected to.

53:55Meanwhile, n8n gives you absolute

53:57flexibility. Like I said, anything with

53:59API MCP exposed, you technically

54:02can connect that to n8n. For example, in

54:05one of the workflows that I've In both

54:07of the examples that I've shown, it was

54:09actually through a HTTP request node,

54:12right? So, that is a very That's a

54:13generic node to connect it to anything

54:15that has API or MCP exposed.

54:18Meanwhile, with Zapier and Make, I want

54:20to say that you know, it's easier to

54:23learn in a way. The learning curve is a

54:26lot less.

54:27So, if you're not thinking of something

54:29too complex,

54:30you know, Zapier and Make is a

54:32a good option to go with. However, you

54:35do trade that simplicity with some of

54:37the control. Like I said, with some of

54:39the

54:40So, for more sophisticated workflows

54:42that you might have in mind.

54:44For example, like, you know, having

54:45human in the loop and stuff like that,

54:47which

54:48to be fair, Zapier and Make can do, but

54:51it's not as

54:53as advanced as how n8n built it for. Cuz

54:56you can have human in the loop, you

54:57know, in Slack on

54:59Telegram on, you know, any other

55:01platforms that you want, and also be

55:03able to have another agent maybe check

55:05the previous agent's work and be able to

55:08call, you know, sub workflows, for

55:09example, right? Don't want to get too

55:11ahead of myself here, but maybe you'll

55:12hear this term, you know, in the next

55:14few webinars where you can have a

55:17workflow calling another workflow. So,

55:18having almost like a multi-agent

55:21orchestration layer, right? So, those

55:24are the type of things that may be a

55:25little bit difficult to build in Zapier

55:27and and Make.

55:29But yeah, so to me, that's the biggest

55:31difference between n8n and Zapier and

55:34Make.

55:36All right. So, I think there's also lots

55:37of questions around the pricing

55:39regarding any n there's no free plan as

55:42far as I know they give you a free trial

55:45and then you're hooked and you they take

55:49your money

55:50at least like for me I'm using the basic

55:52one which is $20 per per month. I think

55:55it with this one at least we're running

55:57lots of workflows. It depends if you're

55:59doing lots of workflows maybe you want a

56:01more advanced plan but I think as a

56:03start the basic one is more than enough.

56:05I don't know Marconi if you have other

56:08suggestions. No absolutely. I think the

56:10basic plan will get you a long way. It

56:13is 20 bucks a month but it comes with

56:15again the cloud infrastructure. Again

56:17like I said it's totally open open

56:19source right? So technically you could

56:21have GitHub and

56:22you know if you want to go through the

56:23technical setup you can set it up on

56:25your even your own system even on your

56:27own laptop right? But again as soon as

56:30you close your laptop

56:32you turn off your laptop it's not going

56:33to work because it's depending on your

56:34system to run. Right? So you probably

56:36want to run it in a VPS for example like

56:38hosting a right? So you probably see

56:40hosting a trying to you know

56:42peach their their any n instances and

56:45stuff like that is also a possibility.

56:47But again you need to maintain it. You

56:49need to spend time looking into the

56:51infrastructure. So really depending on

56:53your comfort level and

56:55what you're looking for there.

56:57Yeah and also there are some questions

56:58regarding the API costs. Also it depends

57:01on I would say which LLM you're using

57:03which model. So like if it's for example

57:06GPT-5 mini like it costs maybe few cents

57:09to run it. If you use maybe a different

57:11model it's more expensive. So definitely

57:13keep an eye on the API costs when you're

57:16running these workflows. Once you create

57:18the API key whether it's AI Google

57:20Studio or the platform.openai you can

57:23always see how much you're spending on a

57:25daily basis. But if you're using the

57:27basic modules it should be fine like few

57:29cents there and there.

57:32Yeah. Yeah. I would say, you know,

57:35just from experience that

57:37kind of again, it depends on

57:39what you're running, right? Like if

57:41you're generating video for every

57:42inbound lead that's coming in. It also

57:43depends on the traffic. I mean, inbound

57:45leads that you're getting, right? So, it

57:46can go up pretty quickly also.

57:49But there are ways to kind of fine-tune

57:50that. There are ways to kind of be

57:52selective with the model that you're

57:54using, which is what Anyscale is great

57:56for. As you've seen, you can actually

57:58choose the model the exact model that

57:59you want to choose for certain

58:00operations, right? So, if you think that

58:02for example, the operations of coming up

58:04with the email copies don't really need

58:06such a complex model,

58:08well, just put GPT-3 or whatever in it,

58:10right? And it's going to cost you like

58:120.001 cent for for every run, right? So,

58:15so yeah.

58:17Yeah, exactly. And even for me now, like

58:19we're going to see throughout the this

58:20week, but in some task we use Gemini and

58:23other task we use GPT. Like that's the

58:25cool thing about Anyscale is having that

58:27flexibility to like select which model

58:29to use for which task. Uh there's lots

58:32of more questions that unfortunately we

58:34cannot answer, but feel free to connect

58:36with Marconi after this session.

58:38Guys, we still have Anyscale webinars

58:40tomorrow, Wednesday, Thursday, and

58:42Friday. For tomorrow specifically, it's

58:44going to be more focused on creating

58:46workflows

58:47to for content creation. So, it's going

58:51to be a fun one. We're going to share

58:52lots of demos, live demos so you can see

58:54how you can use them yourself. Uh so,

58:57it's the same time as today. It's going

58:59to be around 11:00 a.m. CT time or 6:00

59:02p.m. CET time. Either way, you'll

59:04receive an email 1 hour before with the

59:08with the joining link. Uh Marconi, man,

59:11it was awesome having you. It was I

59:12think everyone in the chat

59:14enjoyed it. Super nice introduction

59:17session for Anyscale. I think people now

59:19have a good foundation of what they can

59:21do with it. And hopefully for the

59:23remaining of the week, they can see how

59:24they can build on that skills that

59:26they've learned today.

59:28Yeah. Thank you Zerex for having me and

59:30again, thank you everyone for uh sitting

59:33through that and I hope it was valuable

59:35and um yeah, please feel free to reach

59:37out uh if you have any questions. Um

59:40yeah, I'm happy to chat with everyone.

59:43And um have a good one, guys.

59:45Awesome. See you tomorrow, guys. Have a

59:46good one. Thanks, man. Enjoy your uh I

59:49think it's 1:00 a.m. your time or

59:50something.

59:51It is Yeah, it is. I'm going to hit hit

59:53the sack pretty soon.

59:55>> good night, man. All right, good night.

This transcript was generated from the captions YouTube publishes for this video. Get the transcript of any YouTube video atfreeyoutubetranscribe.com: free, unlimited, no sign-up.