How to Automate ANYTHING with AI (N8N Tutorial)

So, I recently did a video on Chad GBD agents and a lot of people in the DM said, "Hey, Vun, I think you should do a video on NAT. " To be very honest, about five or six years ago when I was trying to automate some of the workflows at my company back then, we used to use this tool called Zapier that used to connect different tools and you'd be able to have something in one tool trigger something in another tool. And it used to be very useful, but today we're in a different era. Today, we're in the era of AI. And there is a tool called NAT that you can use to automate a bunch of your different workflows. Whether it be marketing workflows, whether it be tech workflows, whatever you want, NAT allows you to do. Now, everyone wanted sort of a tutorial in A10. So, I thought, let me do a tutorial, show you a simple workflow, show you how to get started, and then we'll also show you a slightly complicated workflow to kind of automate what we do with our AI shots. Just a brief version of it. I'll explain that to you. Ready? Cool. So, let me first

tell you what NA10 is. Naden is like a visual workflow automation platform. It lets you connect apps, services, and APIs. So for those of you that don't know what an API is, it stands for application programmer interface. So a very simplified way to talk about this is if you have another tool, let's say you want to do something on Google calendar, you want to add a new calendar invite, you can actually use Google calendar's API and sort of go and tell it, hey, do this for me. But with any 10, instead of writing code, you drag and drop nodes. Each of these nodes represents different actions into a visual canvas and you can connect them with lines to create automated workflows. The cool part is anyone with access can see your entire automation process at a glance. So it's very easy to understand. debug and if you want to modify it as somebody who you know doesn't write code, you want to go in and modify it, you can actually do that. So it allows teams to be in sync with each other a little bit easier compared to traditional coding. Anyway, before we dive into NA10, I want to explain what JSON is. Now, NA10 workflows are all JSON based. Think of it like a data format that allows you to represent data in a very easy to consume and easy to read format. Okay? Now, NA10 templates are actually all JSON. And the cool part is you can copy somebody else's NAN template and you can paste it as JSON and immediately get their workflow. In fact, community templates, which is templates for repeated workflows that you might use in your business, are available in many different places. You can find them on the NA10 website. You can also find them on GitHub. So if you find something clever, you can easily transmit it to other people. You can also use their workflows in your business. So I'm going to first start with explaining three core concepts of N810. The first one of course is the workflow. It's like a sequence of actions. So you have this visual canvas-based interface. Now workflows can be of two types. The first one is linear. If this happens, do this. The other one can be nonlinear, which is ask an AI agent or an AI in this case, hey, what should I do with this user has asked me for something? Make some decision. So it can be deterministic or non-deterministic. In this video

actually I wanted to show a hosted version of this. So in this video, we partnered with Hostinger to demonstrate the most cost effective way to run your automation server. So I'm going to start there. I'm going to show you how to set it up. To begin with, we simply go to Hostinger, okay, and select the KVM2 plan. And under applications, you'll just find NAT there. So I'm just going to select NAN. And then you start. And you have your automation server running as we have running here. I've already done this process. The KVM2 plan gives you two CPU cores and about 8GB of RAM, which is enough to run tons of workflows running simultaneously. Costwise, it's about 549 rupees per month. So, it's not really expensive, especially if you assume that the workflow can save you a lot of time and money. The best part is that unlike NA10's expensive costs and cloud execution limits, your self-hosted version can run unlimited workflows. So, we're going to run unlimited workflows with this. and you it's basically all you can eat. So, if you'd like to set this up yourself, visit hostinger. com/invarun and use the coupon codes listed in the description to receive a discount. Awesome. So, now we're going to assume you have your server setup, which is a one-click install of nat on Hostinger.

Now that you have it set up, let's start with setting up our first workflow. Okay. So, we are going to make a meetings. So first, okay, we are going to set up a telegram trigger. So I've already made this workflow here and I want to explain each and every part of it. Okay, so this should trigger when the bot gets a text message or an audio message on Telegram. So as you can see there's a Telegram trigger and I wanted to connect to Telegram account 2 via the Telegram API and I wanted to trigger on message. Okay, so to set up the API and add in your credential and stuff. I'll show you in a second, but I just wanted to show you what the Telegrams trigger looks like. So now if you go to settings, you can see, you know, if you wanted to always output data, execute just once, you can do all of those cool things. And there's a lot of trial and error for you to figure out what works best. But for our current use case, we just needed to trigger when a message appears on Telegram account, too. Cool. So I've opened this switch. Okay. And switch makes some decisions for you. So we're going to put it in the mode called rules which is based on a matching rule. We'll do something. There's another way to use it which is expression but I'll leave that for a different time. So let's set select rules. And as you can see there are some routing rules. Okay there's that if the message dovoice dome type contains audio then you want to rename the output to audio. If it's a text text. So anyway now we're going to click on execute workflow to see where that came from. I'm going to go to Telegram, this thing that I've set up, this bot, and I we'll teach you how to do that. Hey, I want to meet Varun Maya at 10 p. m. at teamvm subscriptions@gmail. com. Okay, so I've set that up and now if you see n10, the workflow is executing, an event is set up, and the workflow is executed successfully. Okay, so if you go back to the Telegram bot, it says your meeting is scheduled. Here is the meet link, which is pretty cool. Now

I'll tell you where those variables came from. You see this thing that says JSON dossage type. As you can see it's now red in color which means this data didn't arrive. And this other one JSON dossage. ext is green because it arrived. What are these variables? As you can see it's all on your left. Everything that came in with your message is here. Okay. So what is the ID? What is the message ID? Uh whether it's a bot or not. So it says here is bot is false. That's because we are the user and we sent a message. So it's not a bot sending the message. is a human sending a message. The account setup is Shya Mishra's account. So as you can see the first name, last name, username, uh language, the ID, the again the first name, last name, username of the chat. And then you can see that the date came in and the date has come in a very different format. It's actually Unix format. We don't need that for now. But if you see the text, it says, "Hey, I want to meet Vunaya at 10 p. m. at teamvmsubscriptions@gmail. com. " So I'm actually going to delete this here. Okay. And say that, bro, this value, I'm going to drag this text from here. and bring it here which is where that variable came from. So it's part of the payload that's coming in from that message. So where do we get that voice data from? How did it even come there? Why is it read and why are we not able to edit it? Well, to do that, we first need to execute the workflow. Okay, so I'm executing the workflow waiting for us to create an event in Telegram. Let's go to Telegram and let's put on a voice note saying, "Hey, I want to set up a meeting to meet Vunaya at 5:00 p. m. at teamvmsubscriptions@gmail. com. " and I hit send. Now that data is sent, let's go back over to NA10. Now, as you can see, we got an error. Remember, all these platforms require debugging things. So, it says problem in node create an event where it says bad request. Please check your parameters. So, I'm going to click on this. I'm going to find out what is wrong exactly. Okay. And it says invalid attendee email. But wait, I said teamvm superficienc. com. Why didn't it show? So, I'm going to see what is it. What is the out? What is the thing that it sent? And as you can see here, okay, the AI agent has got the output team space VM space subscriptions space at spacegmail. com. Wow, that's quite some pathetic text to audio. So let's go diagnose this at the switch level. When you double click the switch, you'll see that if you go all the way here, you have a file that's come in here. Okay, so the audio is sort of data. And now you can see that the actual audio here is green. The variable here is JSON dossage type right and as you can see I can just drag this from here to here and you can actually set that up and that's how that became green but the translation isn't doing well some step but either way this is how you do voice and text and all I'm doing is renaming the outputs from the switch to either audio or text so we don't have to deal with like long strings okay so now back to canvas now we're going to download the audio okay so if the switch is coming out from uh audio it's if it said bro we've detected audio uh we are actually getting the voice file ID so in the last one we're checking whether what's the type is it audio if it is audio then bro take out the data from the audio so now again we're just dragging the file ID so we're actually getting the file ID here okay so we're just dragging that out and remember this goes out from the audio part so remember these are the two outputs audio and text I've dragged out the audio and I'm saying audio file but with text don't do any of that go straight to the text message part and we'll explain what that is. Now that we've got the audio, we're going to transcribe it. Remember, we have an audio file. We got all that information. We need to transcribe this information into text. So for this, we're using Google Gemini's Palm API account and this comes from Vertex AI. So we are saying audio resource operation please transcribe a recording. Okay, I want you to use Gemini 2. 5 Flash, which is not the best audio transcriber, but fine, whatever it is, we're using it for now. And in the last one, remember we called the file name data. We're saying, please use the data component. Okay, so what's happened? We've downloaded audio. We got the file payload. We have sent it to Google Gemini's 2. 5 flash. We've asked you to transcribe it. Now the transcribed data is flowing through the nodes. Then we go to the transcribed audio. Now if you see the next thing it's actually called a set node the entire you know pen pencil whatever you see there. So if I press plus you can see that I can just type in set. Okay so I can just set fields right I have added the set node and I've said bro this entire thing that you got this data payload please remove the text and you can see here from Google Gemini 2. 5 flash one of the outputs we got is text. So I can drag it and drop it here which is what is there. Okay. So now we've got the text and again the text that we got was I'm going to say hey uh please uh set up a meeting to meet Vonaya at 5 p. m. at teamvm subsubscriptions@gmail. com. Okay. So that data we have set it and we're calling that field text. We can name it anything we want but we're calling it text for now just to make it easy for us. Okay. Now that is if it came via audio. Now let's see what we should do if it came via text. If only a text message came. Okay. Now again we are going to just set it. Okay. And this one's easier. There's no transcription. There's nothing. We just say field to set text. Okay. And the message we we've pulled over from the telegram trigger is the text. So wherever the text is what now in this one there's no text. That's why this is grayed out. But if there was a text message this so that's the thing, right? It's going along the top path because we sent audio. It'll go along the bottom path if we had sent text via the bot. Now it goes to an AI agent. The AI agent will make a decision. Remember I told you there's linear workflows. Till now we've been doing linear workflows. Now we're going to ask AI to be the brain and make a decision for us. So I'm going to double click this as AI agent and I'm going to say bro take that text whatever we had. So you can see right after the field was set the transcribed audio. I pinged it down and you can see this text. So I've dragged it here and now I'm adding a system message. And here there is some system message which is you are a helpful assistant. Your task is to schedule zoom meetings. You'll receive a prompt from the user which includes timings in full datetime format people to add etc etc. You need to output data accordingly. And then I add another line which is do not ask any follow-up questions otherwise sometimes the workflow will break if it asks you follow-up questions. Another thing is maybe it should have been to schedule meetings Google meet meetings right why should it be zoom we don't have zoom subscription can't afford it. Okay. The other thing is remember that it says that includes timings in full datetime format which is date time blah blah blah blah. Remember when we pinged it, we just said 5 p. m. We didn't give it full date time. This is where these non-deterministic AI systems come in handy. You don't need to perfectly match the format. Even if the user is somewhat there, it's okay because the AI will be like, "Okay, this person means this time. If he says 5 p. m., okay, we'll take the closest 5 p. m. and make a decision. " That's why I said do not ask follow-ups because they ask for follow-ups then the AI agent will reply and then breaks the workflow because then you'll need to have the telegram bot continuously asking questions to the user which we don't want to do. Cool. So now we have had the AI agent do this and as you can see it's asking what chat model. Remember the AI agent now is taking some inputs. It's taking some things from uh it's asking for some things. The first one is the chat model. What is the chat model to use? And here I am using Google Gemini Pal API account. I'm using a Gemini 2. 5 flash. Remember 2. 5 flash is not the smartest model, but it's good enough for some use cases. Okay, so I've not added any memory and in the parser I've added something called a structured output parser. I'm telling the AI, please output with this structure. Okay, so I've said please output start time, please output end time, please output candidates number one, number two, whatever it is. Okay, I want this to be exactly this type of output. Okay, so as you can see the output coming out from here are teamvm subscriptions@gmail. com which it butchered the email. It's added spaces and stuff because it got the audio wrong and it perfectly gave a time which is actually its best guess because I just said 5 p. m. it could have been on any day and as you can see the output here is nicely formatted into start time, end time and the candidate and the first candidate which in an array will be number zero is teamvm subscriptions atgmail. com. Awesome. And then it goes and creates an event. I'm using a create event node. Okay. And I'm using Google Meet for this. So I've connected Google calendar account. You can connect your Google calendar account. And I've said I want to create an event and you can create many different types of things. You can create an entire calendar if you want to. And I'm saying from list which is from email which it needs to set this from is teamvm subscription@gmail. com. So what is the start time? I drag the start time from that output parser. I drag the end time and then I'm saying attendees is this first candidate attendee. And as you can see, this is where it failed because an attendee can't have spaces in their email. Maybe the AI agent should have been smarter. If we had used a smarter AI agent, maybe it would have said, "Okay, fine. There can't be spaces in email. " And we can also prompt them. We can say, "Bro, if you find something that looks like an email, please convert it to an actual email, even if it's the wrong email. " But that's a trade-off. Anyway, now that it says set, I want to set a send a message on Telegram, which is, you know, I've said connect to Telegram account 2. This time I want a message. Okay, I want to send a message. Then as you can see in the chat ID, this thing is gray and there's nothing on the left because it says error running node create an event. Why? Because the previous node failed. Every node is dependent on the previous node cuz the previous node failed. We literally don't have an output available here. So we are not able to continue with this step. So let's go back and this time I'm going to show you the text workflow. Okay? Because the audio workflow sometimes the transcription can come wrong. It really depends on how good your mic is and if it's long emails it'll butcher it. But let's show you text. So I'm going to click on execute workflow again. Okay, it's waiting now. Let me go back to Telegram. Please schedule a meeting with varuna at 5:00 p. m. Thursday at tmvm subscriptions@gmail. com. This time because it's text I have control. So I'm putting the exact email. I'm pressing execute. Now let's see it execute. And as you can see it executed pretty fast. It created an event and it sent a telegram text message. Now if you open the last node, we can see in the last node how now it is active, right? You can see that the chat ID now has become green. We were able to drag the chat ID from here. So now you can see from the telegram trigger if I drop this down I can just drag the chat ID from here into this and therefore it'll become green and as you can see the text is also there and I can type in my own text. I can say your meeting is scheduled. Here's the meet link and I can actually drag the meet link that we got from Google calendar. So as you can see the meet link is here. It's clearly here. So I can just drag it from here to here and it will be done. If I want to change the output message etc. I can do that. Now if I go back to telegram it would have sent a message saying your meeting is scheduled. Here is the meet link. So as you can see we just executed a full workflow. And unlike other people running through workflow, I want to show you a couple of errors as well. Where people can sometimes show this perfect version. And I want to show you that it's not always perfect. It does require debugging and diagnosis, but it does make life easier. Now, one last question

you have, which is where are all these set? Where do I set the Telegram API key? Google calendar? Where do I set all of these things? And that is the final question I'm going to answer now. So if we go to the telegram trigger, I can double click this and I open this and I say create new credential. So now it's asking me for an access token and ask me for a base URL. The way we made this telegram bot is through a service called botfather. So you can actually go to Google and search for botfather telegram. Okay? So just go there and you can click on botfather start bot. The minute you do that, especially if you do it on your phone, botfather responds to you and it says what can this bot do? Botfather is the one bot to rule them all. Use it to create new bot accounts and manage your existing bots. So, you know, it gives you a bunch of instructions on how you can start. If you type /start, it'll allow you to create a bot with just text, right? So, I just created a new bot, okay? And I've called it something, okay? And then it'll give me a link to that bot. And it will also give me a token to access the HTTP API. All I have to do is copy this thing that you see here and paste it into N8. So all I have to do now is paste that access token here and the bot is up and running. Anyone who texts that bot, remember it gave you a bot link. Botfather gave you a link. Botfather is the easiest way to create bots on Telegram. It also gives you this nice little access token which you just dump in here and then you have Telegram up and running. Now if you ask me where do I get the Google Gemini chat API? All you have to do is go to Google AI Studio, which is Google's at this point, that's Google's central hub for developers, especially developers developing on AI, and you click on get API key. You can actually click here, create an API key, and as you can see, and I'm showing you on the previous API key, you can just copy the number, and it'll give you all the data that you need to just plug it in to the API key. The host is generative language. googleapis. com. And that is about it. Now the last thing I'm just going to close this but the last thing which is the actual calendar. Okay, you can see that we connected the calendar and as you can see with the Google calendar all you have to do is go there go to Google Vertex AI and this is the part that's a little irritating with Google which is you have to go to different places depending on whether it's a use case one or use case two. I'm just going to search for API keys. I'm going to go to manage API keys. Hover down to create an API key. So I'm going to go to credentials, APIs credentials. Okay, again it's a little bit of navigating through everything and I can click on create credentials, click on oath client ID, go through the process is just filling out a form and then dump that client secret into our NAT workflow and it'll give you all these it'll give you the oath redirect URL as well. Um now how do you find out all this? How do you know von whether to go here or there? The truth is when you look at other people's workflows or when you get the hang of using this, you'll know where to get the API keys from because everyone who provides an API key, anyone who has an API, they will all have their own different formats. Telegram has their own different format. Their botfather along with Telegram is slightly, you know, different workflow. Google Gemini is a different workflow. Uh Google calendar But you get the hang of it and there is a little bit of figuring out that you have to do to actually get the API key or the oath keys. And there's a little bit of figuring out to figure out the right URL and the API key. But you know, it's not something you can't figure out with five minutes of just fiddling around. But a simple way is just to go to Google, type the name of the API that you want, ask for API keys, and it'll send you to the right link. Now, to be very honest, the reason I showed you all of this is because this is one workflow that shows you pretty much everything that we think we could show you that is important enough for you to get started with NA10. If you

went out and now started using this in your day-to-day use cases, you'll find that almost every use case you can think of already has a workflow created. So, if I go to NA10 workflows, okay, I'm just going to search for that. And there are 5,469 workflows here. Personal life manager with Telegram, Google services, voice enabled AI. Talk to your Google Sheets using chat GBD5. Uh chat with database with AI. uh generate AI viral videos with nano banana and v3. So all of these are pretty much available. You just have to press one click. So if I click on this generate AI videos with nano banana and uh v3, I can just click on use for free. I can copy the JSON and all I have to do is go to n10 create a new workflow. Ctrl +V and there you have it. Here is the entire workflow that this person has created. And you can see how the entire thing flows. The value of this is now once you study each node one by one, you can change whatever you want. Either way, you have to change your credentials, but you can change one by one. It even has things like auto posting. I don't know if auto posting works. Is this is a little bit dangerous to have, you know, something automated posted. It's better to just get the video and post it manually because these platforms are looking at uh other people posting automated stuff like on YouTube and Instagram. I think YouTube has already banned a lot of this mass generated content. So, you have to be very careful with it. You have to know where to use AI. tools like NAN. It's not just to use this and just spam the internet with rubbish, right? I think it is important to have good ideas and know how to execute those ideas. So, study some of these workflows and within 3 or 4 days, I promise you, you'll become pretty good at this. Then you are only limited by ideas. So, like I promised

you, we're going to show you a very simple idea to avatar NA10 workflow. We don't use this because we do a lot of this process manually. We still need quality video editing. But let's say you're somebody who just wants to learn, right? So, I'm going to show you the workflow. We already have one we set up for you. And now you'll be able to understand it very quickly. So, I'll tell you how this works. The idea was if there's a nice Reddit post, we're able to take that Reddit post, put it in to, you know, this tool. You just send it on Telegram and add a small line below of our opinion, whatever our opinion is on that post. So, what we're doing is Telegram trigger. We already know how this works. We're setting a field. And as you can see, it would have just set the text. Okay, we already know how to do this. We've learned it in the last section of this video. We don't want to play around with audio because sometimes it gets the translation wrong. We use an AI agent, but this AI agent has no prompt. It has zero prompts. Okay? Instead, what's happening is it's using Gemini and I just want structured output. I'm asking for a link. What is the link that the person has sent me on Telegram? What is the subreddit? What is the post ID? And what is the text, which is, you know, what is the feedback or what is my opinion on that Reddit post? Okay. And then it goes all the way it extracts data from Reddit. So here you're using the Reddit API uh connecting a Reddit oath account getting uh the actual post. So it's getting many comments from the post. Now here's where it gets a little bit complicated. So in this video we didn't want to really you know go into code. We didn't want to actually show ourselves coding because we felt like it'll alienate a lot of the non- tech audience that's watching this. So what we did instead was we said we'll put in a node and the node is called code in JavaScript but we'll go to claude we'll ask claude or any model you can use any model to do this we'll dump in all the results we got from this get many comments in a post which also comes with a lot of rubbish data right the ID this that we don't need all of that I told claude just give me specific comments and replies right so that's all we pulled out we set that field we merged the two so both the post as well as the comments and then we're using an AI agent Okay. And we told it you are the best script writer. Your task is to write a script with the Reddit text and the opinion provided. So all the Reddit text packet of data below that are opinion. That's what we're sending it. Okay. And you can see here uh what is the text and what is the opinion. So there's text, there's comments and there's opinion. And then we're telling it we write a script for Instagram reel using a broad range of audiences blah blah. You can write whatever you want. Right? This is basically a Reddit reactor workflow. Once we got that output, we're converting text to speech. Okay? And we're using 11 labs to do this. There's a nice little node that allows you to use 11 Labs to convert text to speech. Again, you got to go to 11 Labs, do the API dance, get the keys, put it in. Now we have instead of just whatever output writing script it gets, now we have audio version. Now we're going to Hen and we're uploading that audio. So we've taken the audio, we've uploaded to Hen. Now it's uh video. We get Hen video. We already have trained an avatar. So you probably have to go train an avatar. Then we ask it to create an avatar. And remember, I'm just doing an overview here so you get an idea. You should experiment and play around a bit. So we uploaded the asset. We're asking it to create with Hen avatar creation. Now remember that while 11 Labs has a way to do text, it has its own node. Hun does not have its own node. So we have to go use the API in the old ways. So we actually have to use the HTTP node. And I'll just show you what it's like. So we're going to make an HTTP request. Okay. And in the request, we can actually upload asset. How do we know all this? This is all in the Hen API documentation. Then you wanted to wait for a while, right? Because remember the avatar process takes some time. So waiting 3 minutes, which is enough time to give the avatar time to cook. Then we're fetching hijen videos. So we're going back to the API key and then fetching uh the video that we want. Okay. So we're going back to hen fetching the video that we want. Then we're downloading video. Then we're uploading video to YouTube using the YouTube node. And then we're sending a text on Telegram saying done. So anyway, I'm just going to show you one of the outputs from this workflow after executing it. Here it is. Humanoid robots may have just crossed the uncanny valley. In Hanzu, Aia from isn't just building emotional androids. They're crafting replicas of future humans. They partner with artists to design faces that look naturally beautiful, then power them with character mind. This is

the process. Now, of course, you can get really complicated with this process. You can stitch together a little bit of hen video, B-roll. It'll be a very, very long workflow. The problem with this is your videos will start looking the same after a while. Your videos will be it's just going to be somebody else's post with a little bit of one line of opinion, right? But it's still written by AI. It doesn't have insight or thought too much insight or thought into it. That's why instead of automating everything that we do, we still have humans, but we do use a little bit of automation. Like I don't present on every video, right? We can use AI to present on some of the videos. So knowing when to use automation and automated workflows and when to use humans is probably a very important skill, right? Because if you go too much to the automation side it people won't engage with the content and you've seen some of those people who just generate just AI generated you know some oh this bell this whistle some someone jumping in m Minecraft that's slob right and you have to avoid doing that you have to know when to put humans in uh but you don't need to have as many humans as before you don't need you know an entire media row right you could have like team is maybe 10 people now so you could have just 10 people doing the actual editing process where that style that quality of edit really matters Anyway, I hope you

learned something and I feel like platforms like this, no matter how many tutorials on this you watch, the first time you do it, you'll have aha moments every time you click something new. And there's purposely parts, especially with the second workflow that I didn't explain because I want you to go explore it, right? Because the first time you do it or even once that you do it, you'll feel so much more empowered to do more, right? To try more and you'll have cool ideas for workflows. And I do think this is going to become an important role, right? To set up these workflows and make sure that you're running your company on autopilot. That's it for me. Bye.