N8N Full Tutorial: Building AI Agents in 2025 for Beginners!

today I'm going to teach you how to build AI Agents from start to finish in N we're going to start with the basics and then work our way up to some of the more Advanced Techniques before we start jumping in and learning how to build this we need to understand what it is and why it's important first so what are agentic systems agentic systems are environments built up of agents and workflows so what are these agents and workflows workflows are automations or systems where the output is predefined by the large language model and other tools

you can think of a workflow like a post product purchase automation when someone purchases a product maybe you create a workflow automation that sends them an email with their order details that's a workflow now agents on the other hand work a little bit differently these are automations or systems where the large language model dynamically decides which tools and output is necessary so you can think of this like a customer support agent that has access to specific tools maybe these tools could be things like creating support tickets pulling in order history or even issuing refunds you

basically give the large language model or the agent access to use those tools if that conversation ever comes up they can decide which one is best used for that scenario so there's no one way this automation could go the agent has to decide which tools to use in order to accomplish requests made by a user so visually this would be a workflow right you have an input whether it be from a user on a schedule uh from a certain application and then that input goes through an automation workflow and gets you a consistent output now

the data could be changing in the output but you know that the output is going to go to the same spot you know where it's going to go because you've predefined it within building your automation while the agent example looks a little different so here the large language model would be receiving an input in this large language model depending on the input has a couple of tools or multiple different tools that it can use in order to reach a dynamic output so an agent is like a large language model with tools and we can even

put agents nested within other agents so that one agent can call another agent which has access to a certain amount of tools as well and we're going to get into that concept a little bit later in the video but it was important to understand what are agentic systems and to visually see it for yourself now if you enjoy this video and you want to keep building building automations and agents with us then I recommend joining our community for AI foundations we have this community hosted on the school platform and in this community we focus on

the three C's number one courses number two calls and number three saving the best for last Community the people that make up this community are better than any content within but let me tell you if you want to stay ahead of the curve on agents automations the fundamentals of AI then this is a spot to be we make AI easy to understand here we have a full nadn agent building course we have a calendar with a bunch of live calls and of course the lovely members that make up this community we just dropped this course

about 20 hours ago and people are loving it Anthony here says this community is a gold mine for anyone who finds it best investment of 2025 John here agrees with Anthony this community is truly a gold mine for anyone who finds it and it's true it's kind of a hole in the wall Community we've got 580 members we'd love to see you inside I'll leave a link to join below so once you create your nadn workspace on the cloud you're going to be at a screen something like this except you probably won't have all of

these different automations in here this is basically like your homepage and here you can see workflows credentials and executions credentials are things like your API keys and your account access that you give within your automations and executions are actually the running of your agentic workflow so whenever you hit run it tracks it as an execution or test step or whenever it runs through then it's an execution you have your domain which is the workspace that gets created for you upon creating an account and then your workflows over here are organized into projects over on the

left so I could select into specific projects and I have different workflows different automations within each one as you can see in order to create a workflow it's very simple all you have to do is in the upper right hand corner hit this Orange button that says create a workflow and then you are going to be within the editor UI what you can also do is select a specific project and hit create workflow if you want this workflow to be within a specific project now that you're in your project workflow what you want to do

is understand the node types before building AI agents within nadn nadn comes with different node types now personally I like putting node types into five different categories now nodes are anything that you put out here on the screen so I could have a node that starts on a chat message okay so anytime I chat this automation will start this would be an example of one of those five node categories so the five node categories in nadn are triggers these are the nodes that start your automations and this determines where when and how your agentic framework

will begin we then have action nodes so the these nodes allow you to do something within an app or a service for example Google Sheets air table notion WhatsApp telegram you could import Gmail Outlook these are your action nodes you have thousands of different actions you can take between all of the apps and services available to you on n8n next we have utility nodes these are those little boxes that will be able to transform your data things like if statements filters data conversion tools data storing tools these tools are native to nadn and they can

help you organize modify or send data elsewhere we also have code nodes so you can run code make HTTP requests set web hooks JavaScript and more now this takes your agent to the next level we also have the advanced AI agent node and this is what makes your agent autonomous this is what turns your workflow into an agent within your agentic system you can add search and retrieval functionality store memory start large language model chains perform sentiment analysis and so much more so these are the five node categories and pretty much anytime you add in

a new part of your work flow you can kind of think within these types of categories now when you are beginning an automation you always start with a trigger so that's what we need to go do now so the cool thing about the trigger nodes is you can do multiple you don't just have to have one way you start an automation you can have 10 different ways you start an automation if you would like but we need to add our first step and what you're going to notice is when we do add our first step

it automatically asks us what triggers your workflow this is the where what when how of your automation the interrogative of your automation so what we can do for Simplicity is we could just do on a chat message there's multiple different node types you can select from especially when you get into onapp events you know within these applications you can start automations on certain events happening if we go to Google Drive for example you could start an automation on a specific file change or when changes involv in a specific folder so there's different ways to start

your automation but for now we can just do on a chat message M I'm going to make this chat publicly available and this is whether the chat should be publicly available or only accessible through the manual chat interface for now we can just make this publicly available and we can have this test URL right here I'm going to delete the initial messages and this is just preliminary stuff this is kind of cool you have the ability to change this from hosted chat to embedded chat which is kind of nice I'm going to go back to

Canvas and now we have our first node in here which is great great we have our first trigger and this lightning bolt right here is signaling hey this is what's starting the Automation and what you will notice is at the bottom of the screen we have a chat button so when we select this chat button we can type in anything we want which is cool and this is how we're going to be communicating with our automation uh for the sake of this tutorial this is actually a very useful node to have let's add our next

node in here and in order to get this started I like just adding the AI agent node so I'm going to open the nodes panel and then I'm going to go to Advanced Ai and then I'm going to add the AI agent and this is the secret sauce of n8n this is what makes it so amazing so we can leave everything the same right now uh because right now the prompt or the user message is just looking at this previous chat node so this is like the brain of your workflow your automation or your agentic

system there's so many things that we can do from this AI agent node we could even just talk to it like it's chat GPT and we could have our own version of chat GPT within n8n let me show you an example now in order to power this AI agent node you always are going to need to add a chat model you need something to power the brain right you need something pumping blood to the brain so we can just use a chat model and I'm just going to do the open AI chat model you can

select a model in here and they've got all the open AI models which is very cool I'll just do GPT 40 and then I'll click out of here but there's so many different models you can select it doesn't just have to be uh this chat GPT model you can select anthropic AWS Bedrock grock a Lama I mean there's so many different things in here you can select from but I like keeping it simple I've already got my API key connected with open Ai and I'm just going to power it with a chat model now chances

are you might not have your credentials connected to chat GPT in order to do this or whatever application you want to use in order to power your AI agent so if that's the case what you can do is go to platform. open.com and then you can go to this settings button right here and within the settings button what you can do is go to API keys and this is the screen that you want to create a new secret key and once you create a secret key it'll take about 30 seconds to create one and once

you do you copy your secret key you come back here and you have the ability to create a new credential and just paste it in right here and then you're all hooked up it's that simple you can load up5 to $10 in your open AI account and it will be plenty for testing purposes so once you have a model connected to your agent and you have a way to start your automation for example right now we're chatting with this agent in order to begin this automation process what you can do is you could just talk

to this thing like it's chat GPT or any other large language model so I could ask it hey how are you and then I could send it off and now it's just using chat GPT and it brings me a response it says hello I'm here and ready to help how can I assist you today but the problem is right now this AI agent can only understand this query and then it doesn't maintain any previous context in the chat so in order to fix that we need to add memory so that it can understand what was

previously said so that we can actually have a conversation with it and it can use our conversation as context in order to help us for example without memory things like this happen I can say continue the see quence for me each time I reply 1 2 3 and then I'll will put dot dot dot so it should say four 5 six so if I send that off what it will do is say four and maybe it's just going to go one number at a time so if I copy it and I go five what it's

going to do is not understand the previous context so it's not going to keep on going it just says my message looks like it's incomplete but if it had memory it would be able to do this so we can add temporary memory to this chat by selecting this memory plus icon and then selecting window buffer memory this stores memory in N there's no credentials required it's the easiest way to go so here's your context window length and this is how many past interactions the model receives his context for now it's connected from the chat node

so it already has our session key stored from this usually with this memory what you need to do is you need to map a specific ID so if you connect something like telegram if you connect something like WhatsApp you need to store that chat ID right here so it knows where to pull context from but right now it's mapped it for us from this node right here that's great so now if we try this again and we say continue the sequence for me each time I reply remember we have a window buffer memory of five

now I can save this and then if I paste that in here and we go through this it will say four and when I type out five this time it will keep on going in the sequence because it remembers the past conversations because we gave it that context window to do so so now we can have full conversations with it four five 6 7 8 that's what window buffer memory does and you don't just have to use it with this chat node but you could use it with other nodes as well now that we know

we can chat with this thing how about we start making it a little bit more powerful let's add our first tool to this AI agent now when we add a tool the AI agent is going to be able to decide should I use this tool or not depending on what I say from the messages so whatever I type in it's going to determine whether it needs to use that tool or not so what we can do is we can hit this little plus button underneath tool and within these tools we can do some pretty cool

things we can call other workflows so right now we are in an nadn workflow we could add an entirely new workflow as a tool we could write code make an HTTP request we could call all sorts of vector stores and then we could use other tools as well like we could add Gmail Google Calendar we could give it a calculator Wikipedia so it can search Wikipedia woocommerce Wolfram slack so many different tools that we can use but for now what I want to do is I want to select air table let's say we have a

home Inventory where we have all of our grocery items our household items all in a list with an order threshold so that we know when we need to order more I'm going to select air table and what we can actually do is we can search Air table databases within this tool right here so if I click into this tool as you can see we have a lot of different options here that we are going to get to but I want to point your attention to operation because within this air table tool we can do so

many cool things we can create records we can update them delete records get records and we can search records which is what we want to do right now so I'm going to select search and now this will be ready but first I want to show you the database so here's my home Inventory database now if you're not familiar with air table it's like Google Sheets on steroids you can run automations very easily you have the ability to view your data in multiple different ways it's like if Google Sheets and notion had a baby uh you

would get air table here but as you can see I just have some item names like toilet paper shampoo hand soap toothpaste trash bag Ziploc bag so it's basically like an entire home Inventory we even have some food items down here we have the current quantity like how much we currently have of that specific item and then we have inventory threshold so when it gets to that level we know that we need to order more so if we get only four sodas left let's just say then I guess we need to order more soda or

if trash bags if we get down to three boxes of trash bags left then we need to order more in order to increase that quantity so this is our database that we're going to be uh giving n access to so that this AI agent when we're talking with it can actually use this tool so what we can do is within air table here we can just go down and hit create new credential and here we need to give an access token with the following Scopes in the air table Builder Hub you can create personal access

tokens I'm going to hit create new token and I'm going to give it a name of test home Inventory I'm going to add some Scopes and this is basically just saying what can can n8n be allowed to do well we want it to be able to read the data we want it to be able to write in the data and we also want it to be able to read our database structure and see things like table names and field types so we can select schema basis read and now we need to give it access to

that home Inventory base we need to give it access to this so we can hit add a base and then scroll down to home Inventory test and select that and now we can hit create token we're then given a secret access token that you will only be able to see once so when you hit copy make sure you uh store this in a safe place where nobody else can see it and now we can go connect our credential back in n8n so I can paste in that access token and then we can hit save and

now our air table connection has been tested successfully so now it has access with those parameters to our air table database beautiful so now we need to give a tool description this is going to describe what this tool is now anytime you add a new tool to your AI agent you want to let your AI agent know hey this is what's going on with this tool and that's what the tool description is for so we're going to set this manually and we can just say searches home Inventory in air table so we're kind of giving

it these words like home Inventory air table so that it knows that if I ask about my home Inventory that it can search it within this air table tool node you kind of have to to be specific within the description you want to explain to the llm what this tool does a good specific description would allow the large language model to produce expected results more often that's the idea of it now we can keep the resource on record operation is search because we want to search our database then we can select our base down here

so I can select home Inventory and then we can select our table as well so I'm going to select table and then I'm going to do the goods the table is within home Inventory because you can create multiple tables within a certain base we don't need any filter and we can just return all that is fine so now this is ready to go so now I could have a conversation about my database because it's now connected as a tool so before we could just talk to it like it was chat GPT we could still talk

to it like it was chat GPT I could ask it to give me 20 facts about Paris I could ask it to create me a business plan I could do whatever but now it has access to a tool that knows deeper information about me so now it's acting in that more agentic type way so I'm going to save this Automation and then I'm going to test it out let's ask it questions about my home Inventory first of all I'm going to come in here and I'm going to make something out of stock so I'm going

to make butter out of stock I'm going to say inventory the current quantity I have is zero now that we have that is zero we could ask a question like hey is anything out of stock within my home Inventory I don't even have to say home Inventory the AI agent is really good at picking up on context even without any instructions I could say is anything out of stock in my house and I could send that off and as you can see it's searching our air table for us and then returning back to me what's

out of stock it says currently you are out of stock on the following item in your house butter current quantity zero and if you remember that's the only thing that we changed to set as zero within our database I could ask is anything getting close to the order threshold and I could send that off to see if we're getting close to needing to order anything again and it will search my database search my home Inventory that we just added as a tool and then it gives me everything I need right here it's even pulling in

the record ID kind of weird but you know these are things we could adjust with instructions it says glass cleaner aluminum foil mouthwash and honey are all getting close to their order threshold so what you're going to notice is that this AI agent I'm having a conversation with it about things that it didn't previously know it didn't have these items as context within the conversation and since it doesn't it uses my tools in order to go find those items and then bring them back into the conversation so this is great we now are searching our

own database having a conversation about it how about we add another tool let's get into adding a tool where we can actually update these records right now all we can do is search our database what happens if we want wanted to update something maybe we bought something we were on the go we say to our chatbot you know we can eventually hook up voice to it and connect it to our iPhones or our Androids and we could we could have a conversation with this chatbot so in the future maybe I'm out shopping and I'm saying

hey I just picked up two more boxes of pasta my laundry detergent I just picked up X Y and Z this agent needs to be able to not only search the records in order to find the record ID that needs updated which it will do but it also needs to be able to update those records so we can just add another air table tool I'm going to hit this plus button under tool and then I'm going to scroll down and select air table here it's a very similar process we already have our credential with air

table connected so we don't have to do that again but what we can do is set a tool description and this tool description needs to be a very accurate description of what this tool does I can say updates inventory items from my my air table beautiful now we can leave it to record and we need to change operation from get to update and after we have it on update we need to select our base once again which is the home Inventory and then we need to select our list which is the goods and now we

have a couple of fields that we can update if we want to so this is where the fun part happens because how does AI know what to put Within These Fields well it really doesn't until we message something and and the output's going to be different every single time that's where nadn adds this awesome expression from AI we have the ability to map a dynamic expression into each one of these fields and ask AI to fill it in for US based on the chat that we have with it in here so based on our chat

it's going to fill in the fields for us without any instructions now adding instructions really beefs it up right it makes it much more consistent but id id is the ID of the record the item name each item has its own ID and the ID is found within the URL so just like with any item the same thing with your emails the same thing with notion the same thing with pretty much anything that has multiple rows records tables everything has an ID attached to it so what we can do is we can use this expression

I copied it from up here and we're basically asking AI to find the ID for us so I'm going to make this an expression a fixed field will allow for you to type in fixed data while an expression allows for dynamic changing data so now we can give this ID a name we're telling AI to find it for us and it goes in this order key description type and default value they're all separated by commas and each one needs to be put in quotes so you're basically just defining your own key what do you want

to call this well since we're updating records in air table and we need to find the ID for that record we could just name this something like record ID and then we must give a solid description describing what this field is that we want pulled in so this is the thing we want pulled in now let's describe what we want pulled in we can put a comma and then two apostrophes or quotes that would work as well and I can say the air table ID of the item needing updated so now it knows okay we

need to find the item that needs updated and we need to use that record ID and now ai is going to find all that stuff for us again we don't even have any instructions we're just telling AI with key description values what we want it to do for us and so now I can copy this expression and do the same thing for all of these fields that I want to edit so let's say I just want to edit quantity I don't really want to change item name I don't even really care to Change inventory threshold

maybe I just want to be able to update the quantity of a specific item from my database maybe I go to the store and I buy something I can delete those other fields and then I can make an expression here and within this expression I can paste in that placeholder name again and then I can say for the key something like new quantity then I can delete this description that I currently have and I can say the new quantity of the item being up updated beautiful now what we need to do here is we need

to describe what is the type of this value whenever we want a number being pulled into a field we need to describe and tell it hey this is a number so for type here we're going to need to add another comma two more apostrophes and then we're going to need to put number so it doesn't return a string value or text but rather a number and this is how to use the from AI expression it's very cool because AI is searching for these fields for us we don't even have to worry about it look at

this this is awesome so I can go back to Canvas now and we can actually try this out you can rename your Fields by right clicking on them and then hitting rename searches inventory or I can even just put search inventory and then I can name this one update inventory this is just for you to see the AI doesn't see this I'm going to clear my history real quick with this chat thread so we can start on something new and then what I can do is I can say I just bought two more things of

toothpaste or two more tubes of toothpaste let's see what it does I just bought two more tubes of toothpaste then I can say update my inventory and then I can send that off and as you can see what it's going to do is actually first to search my inventory and then go and update it because it understood okay I've got to First search the inventory before I go and update it and as you can see toothpaste goes from two to before so we actually didn't need the instructions there and it figured it out itself now

often times this will happen right it knows the tools it has to use but sometimes it can be inconsistent and that's where we need to start adding instructions for how we want this agent to act and how we want it to go through certain processes but we could keep on having conversations about our inventory now I could say can you add two things of honey as well and because it has memory it can pick up on like what I'm talking about so it it knows that I just bought two more tubes of toothpaste and also

probably two things of Honey because it's going to use this as context with my prompt so I can send that off and it will search my inventory to find that honey object and then also update my inventory so now instead of one thing of honey I should have three down here at the bottom so I want you to take what you just learned and think about this entire process as one for your workflow because when you're dealing with multiple tasks and you're truly trying to create an agentic ecosystem it's not all going to happen with

one AI agent node like we just created you're going to need to be able to learn how to create system instructions in order to classify specific tasks within areas of your life or areas of your business or areas of your work and then call these little Tools in order to help along the way so you could look at this like oh this is an AI agent that has access the two tools but what I like to look at this as is my inventory agent my inventory tool this can do everything I need within my home

Inventory and it's one tool because what we have the ability to do is we can add a trigger to this and then we can add this trigger when called by another workflow and what's going to happen is when another workflow calls this AI agent it can send the data or my user query to this input and then start running this AI agent so it doesn't even have to start on a chat message but rather when I chat to a different agent and the way that happens in another workflow is by me adding a tool like

this and then calling an n8n workflow so you can actually give entire workflows to your agent to call and once it calls it it can execute an entire workflow like we just built and then return the data back to the original workflow so that's really where the power begins is when you're calling other workflows with your agent based on your query so whatever question you ask maybe there's a better workflow for it an entire workflow like this with a completely different AI agent node involved and you can also chain these AI agent nodes together and

maybe you pass information through this one and say you know what this agent is better for the job this has been a beginner's guide and introduction to building AI agents within NAD you're now up to speed out everything you need to know in order to go much further with creating these agentic ecosystems so I hope you enjoyed this again if you want to go deeper right now automatically then I highly recommend joining our AI foundations Community where you're going to be able to do so we have a new NN agent building course in here with

25 plus modules and if I click in here what you're going to see is we have all sorts of different modules going over things like retrieval augmented gener within nadn how to set it up we go over the node types more in depth and build out plenty of different automations in here and we even have projects for you to complete so that you can really hammer in your skills with building agents and we want you to learn with us because listen Carter and I you know the two brothers that are running this community we learn

from the people within the community uh more than we probably teach them right the community is such a good spot to be everyone says like okay they finally found their group of people that thinks like them and that's why I want to tell you about it because when people join they're automatically like oh my gosh this is a gold mine how is this only $49 per month but the fact of the matter is what makes it a gold mine is the actual people in the community and when we treat it like a community that's when

beautiful things happen that's when we grow most and learn more and build more is when we do things together with people from hundreds of different Industries right all these different Industries focused on one thing and implementing AI is that big thing so I hope you enjoyed this video I hope to see you in the community highly recommend it if you're into agent building and I hope this was a good introduction for you if you did enjoy please like And subscribe I would highly highly appreciate it all right I'll see you in the next video