Build Your First No-Code AI Agent | Full Relevance AI Tutorial

hey guys in this video I'm going to show you how to set up no code AI agents with relevance AI I'm going to show you step by step how to set up your first AI agent and everything I've learned while building these AI agents for my clients over the last 6 months if you don't know me yet I'm Ben I started an AI agency last year and since then I've delivered around 15 AI agents to clients now I'm not pretending to be a worldclass expert I'm still learning a lot myself too but I have been

spending quite some time in relevance AI and even though it's a no code platform it can get a bit tricky when starting out so this video really serves as a resource I wish I had back in the day when I started building these agents and as the best way to show you how to set up these AI agents is by actually building one we're going to build an AI inbound lead manager agent that I recently delivered to a client but before doing that I do want to give you a brief overview of what are AI

agents when do you use them and a little bit of an overview of relevance AI if you're new to it but feel free to skip to the part that makes most sense to you using the the time stamps below so I've broken this video down into three parts the first part we'll go over what are AI agents what's the difference between AI agents and AI automations and when to use them in the second part I'll go over some key Concepts to know inside of relevance Ai and in the last part we'll actually be building our

inbound lead agent and I'll be go be going over the agent prompting the settings tool building and how to integrate your AI agent with other softwares and with the API so what is an AI agent an AI agent is a software that automates processes makes decisions and intelligently interacts with their environment to achieve a goal so if you can see my really badly done diagram there we have an AI agent which we prompt with a agent prompt or a system prompt it usually has access to a knowledge base too and we give it a a

specific goal there uh we also give the AI agent access to tools and those tools can interact with softwares can perform functions so the AI agent can access these tools to achieve its ultimate goal and automate a process now a common question I get asked a lot is what's the difference between an AI automation setup and an AI agent setup uh they can be very similar actually and the outcome can also be very similar but the biggest difference is an AI agent has decision- making capabilities so we give it a set of tools and a

prompt and the AI agent can actually decide which tools to use to achieve a certain goal while an AI automation is more a rule-based sequence of tasks to perform so when to use when in my experience although you can achieve uh many of these these automations with both in my experience the best use cases for AI automations in a platform like make.com for example uh are more for more these repetitive rule based tasks and workflows less complex systems in general and more predictable workflows and AI agents because of the decision- making capability you can handle

more complex complex tasks and workflows you can work with less predictable workflows and another big thing is you can have a human in the loop uh because a lot of these clients don't actually want to completely uh automate all of these processes always they do some of these these clients actually want to be there in the process to review evaluate before taking the next step I'll give you some examples after but that's big uh advantage of these AI agents that I've seen so here's an example of a more complex system uh where we have an

AI agent versus an AI automation they're both still possible through both ways either relevance AI with AI agent or make.com or zapier with the AI automation but in my experience for these more complex systems these AI automations become extremely complex very prone to error uh if we have to build out these large workflows and processes and in this case I usually prefer going to an AI agent which I give a system prompt I give some tools and then he has more decision-making capabilities to take his own decisions so in my experience uh the best use

cases for AI agents what I've seen are customer service agents sales agents also because a lot of sales reps actually want to be in the loop they want to make sure for example Enterprise sales reps want to make sure that the research and the email that's being sent out is perfectly crafted before actually sending them so it's actually good for them to not completely automate it then we have marketing agents uh also in terms of the content production many people don't want to automatically post AI generated content they do want to review or check so

those are good use cases for AI agents too and in general any other workflow where they have high Precision requirements because in those cases it's very important to have sort of that human in a loop to verify to review that everything's going the right way to do the next and to take the next action so they're basically two types of Agents we have co-pilot agents and autopilot agents so co-pilot agents is basically what I said before it's where we have a human in the loop so an example could be the content repurposing agent so they

want to produce content but with human supervision before actually posting a blog post on their uh blog Etc and then we have autopilot agent for example one I've built is the bdr agent which finds prospects researchers send personalized emails answers emails and books meetings all on autopilot so there's no human involvement here but it depends right because we have Enterprise clients that one client is extremely important so they really want to be in the loop but maybe for SBS they can autopilots are better it really depends on the client and the use case now what's

the opportunity I think the opportunity is gigantic for AI agents uh I think for businesses it can improve efficiency so much it can set them apart from competition it can automate entire workflows and complex workflows and tasks and I just heard Elan musk talk about this yesterday he says AI adopted businesses will out compete others in the very near future and for entrepreneurs of course AI Specialists will become extremely valuable to these companies as they will have to adapt uh and also I think maybe for the first time ever now non-tech people can now build

really powerful business applications with these no code Solutions and if you're an employe of course you can really set yourself apart and become way more productive by automating your own tasks and workflows through AI agents and AI automations so now I want to give you a brief overview of relevance AI why I like it so much why I prefer it over other Solutions and some of the key Concepts I think you should know before starting out first of all it's a no code AI agent Builder which is important to me because I'm not a coder

I can imagine you're not either if you're watching this video uh but that separates it for me from other popular AI tools like crew AI uh also I think it can help in terms of speed of delivery or speed of development um they're way more user friendly than working with gpts or the open AI assistant and you can actually build these tools which you can't do in the open AI platform uh they actually have many templates already available inside of their app but you can also easily build them from scratch uh they have many software

integration already inside of the app like uh some CRM Integrations uh LinkedIn in scrapers website scrapers and and some more uh you can also easily pretty easily connect them through the API with with your own softwares I'll show you that later when I'm building out my agent how you can connect these but it's quite easy and then very importantly they have a very nice UI for agent management which is especially important if you're managing it inside of your company or delivering it to clients you can actually um give them these these uis for their AG

management so if let's say they have a human in the loop system you can deliver this this agent system on their website or whatever wherever they want where they can manage their agents check what they're doing uh review steps before taking next actions and you know it's a you can sort of deliver it in a professional way to your clients they also have tracking usage and dashboards so again if you're managing multiple agents or multiple clients you can easily see the usage for each client or each agent and also they have the availability of working

with multi-agent systems now I'm not going to go too in depth in multi-agent systems in this video but you can imagine that you can already do quite complex things with one agent if you work with multi-agent systems you can do even more complex things so even though it's a no code platform you can really the the use cases are unlimited you can customize this any way you want and you can really do anything you want so no Cod code but still you can go extremely complex you can even add in code uh if you want

so the key Concepts in relevance Ai and any AI agent really are tools agents and multi-agent systems and you can really think of them as layers of complexity so first we have the tools which are more the rule-based sequence executions very similar to something like make.com if you've worked with it these tools you can make extremely complex but in general you want to sort of keep them simple again to not over complicate then we have our agents we work with these tools to achieve a more complex goal and then we can work with multi-agent systems

where we can either let one agent report to another agent or even report to a manager that manages multiple agents so how do these AI tools actually work I'm going to show you later in this video step by step how to set it up inside a relevance AI but I just want to give you a quick quick overview basically these tools again are rule based sequences of steps uh we have access to variables LNM softwares and apis and a know knowledge base inside of these tools and with that combination we can build quite powerful things

so if we take an example of building an email personalization tool we can first of all decide the user input of the tool now let's say we have only have access to a name of a person so we put our user input to a name and we store that name into a variable right which we'll use in all the next steps now the next step we're going to take is for example do a Google search AP call the Google search API where we're going to try and find the LinkedIn of this person uh because we're

going to personalize the email based on the LinkedIn so what we do is we feed the variable to the Google search API plus LinkedIn so the Google search API is going to look for the name plus LinkedIn and gets basically the entire Google search results back then we feed those Google search results to the next step which is an llm step an AI step where we say analyze these Google search results and please find find the LinkedIn URL of this person's name so let's say this this L&M outputs the LinkedIn URL we can then add

it to the next step which is our LinkedIn scraper which is going to scrape the data from this person's LinkedIn and then we can feed that to the last uh L&M step which is write a personalized email based on this person's LinkedIn data so this is just a simple tool but you can imagine you can make this very complex if you chain together these different steps and sort of make a combination between these softwares and nlms you can do quite some interesting things uh again in general I recommend to not make them extremely complex because

again this is very similar to make.com in general rule of thumb no more than 10 steps if it gets more than 10 steps it's better to actually create multiple tools and give it to an agent uh just in terms of complexity and it will be less prone to error that's in general what what I do so how do these AI agents work um in relevance AI in relevance AI we have some nice extra features for these AI agents we can trigger them from softwares with for example they have native Integrations with Gmail or some crms

so we can trigger these agents automatically for example when a new lead comes into a CRM then of course here we have the core instructions which is basically the system prompt then we have the flow Builder where we can basically give it some extra instructure on how to behave in certain situations which I've I've noticed helped a lot in terms of performance of these agents then we have abilities here where we can actually label so the agent will be able to label tasks he has done for example if his sequence of this agencies first research

the lead and then send an email when he has researched the lead he can label it uh so you can see in your UI research lead and when you send the email you can label it again email send uh so you have a nice overview in your UI of where this agent is in this process uh then we have sub agents where we can work with the multi-agent system and some other settings which I'll show uh later in this video and lastly have we have the multi-agent systems again I'm not going to go over that

in this video but you can imagine you can build very complex workflows and systems with multiple agents so now I'm going to go over the agent we'll be building today uh I'm going to show you step by step how to set this agent up but I just want to give you some context before we start building um this is an agent I recently delivered to a client of course it's not the exact uh setup because I can't show you that but it's a it's a replica of that so what was the client case in this

case it this company was a B2B financing solution from Denmark they're actually a startup but they're quite big and they had some issues mostly with managing inbound leads so their sales reps were wasting a lot of time on these inbound leads so what happens is this company their main target is really Enterprise clients right so their sales reps most of their time they actually spend uh trying to do call Outreach or trying to contact these Enterprise clients to get these in but the solution is also available for smbs and they did get a lot of

inbound leads in through their website through a contact form with these more smbs and these sales reps had to manage both the SMB leads and the the Enterprise leads now the process of handling these inbound leads was very inefficient because these these um sales reps were basically overloaded with inbound leads they were spending way too much time on bad leads too little time on on the actual good inbound leads they had no time to research these inbound leads they were really badly prepared for calls and in the end just a lot of time wasted low

conversion rates and this is where we really tried to solve this pain point for for this company with an AI agent so the goal of this AI agent was really to help sales reps prior prioritizing about leads research and in the end uh convert more so what does our agent do it really completely manages sort of the the workflow of these inbound leads so it automatically researches these leads when they come in it updates the CRM with these new data points it scores the lead so these sales reps know which leads to call first uh

it finds the decision makers in the in these companies which is very important for these guys too because they always have to talk to the right person and uh it also prepared a personalized call script for them which up which is updated inside of the CRM so they could literally see okay this is a high priority lead this is this is a little summary of the lead and here's a personalized call script uh so you can imagine it made quite a big impact in this company because these sales reps can just focus on the right

leads and be very well prepared for them so I'm going to show you in a second exactly how we set up so I'm going to show you in a second exactly how I set up this agent but I just want to give you a brief overview basically we have given this agent eight different tools uh first of all it's triggered through hops spots when a new lead comes in then what we do is we automatically lead use the lead researcher tool to find relevant information on the lead and the company then based on that information

we score the lead uh basically prioritizing lead for the sales reps how good or how bad is this lead then we update that information into their CRM we then try to find the decision maker because it's very important for the sales reps when they're in the call that they know who they have to get to in in the end and it will really help them perform better in these sales calls so we'll try to find a decision M maker this is not always possible with public information but if we find a decision maker we're also

going to do some research on this decision maker we're going to update the CRM with the decision maker info and then we're going to generate uh uh personalized call script on all this research data and again update the CRM with this personalized call script if it didn't find the decision maker we'll generate the call script without that information so this is a simple layout of how this tools set up now I'm going to show you the relevance ey dashboard and how we set up this agent so now I'll show you a quick demo of how

it actually works and then we dive into how you actually set this up by the way the template is available for free in the description below so you can literally copy and paste this entire AI agent setup into your own relevance AI um and adapt it to your use case so here we are on the uh relevance AI dashboard in the agent dashboard agent management dashboard so the way this works um is the agent gets triggered as soon as a new contact is is formed or made in um hopspot uh now of course we can't

use their actual template so I just set up a quick contact for myself so the way this works is if we have a new contact here this is just an example and someone signs up we get an inbound lead this triggers a make.com automation the make.com sends the sends the data to the hotspots and creates a new contact and that triggers our AI agent so if we go back right now we can see a new one was triggered and here we can see our AI inbound lead agent is working so on the in the background

you can see what it's doing right so it's first using the lead researcher tool so it's researching the lead for example this past one you can see it labeled what it has done right lead research and scored so if we go back to this one you can see five steps performed in the background so he's researched the tool he scored the lead he's updated CRM with the lead info he has tagged now the conversation with lead researched and scored so you can see this always here and in the sidebar where he is in the process

now he's trying to find a decision maker you can see in the end it will also give you a summary of everything he's done so he's find found the decision maker you can see new decision maker found he's doing the the research on the decision maker he's updated the CRM with the decision maker info and now it's generating a personalized call script so this this is a autopilot agent right so we literally don't have to do anything and he's done so the process for managing has been successfully completed right this is everything is been doing

lead research lead scoring CRM update decision maker if you need any further assistance feel free to ask so let's go back into uh hopspot and see what happened there so we can see a new contact was created the lead score we have now here where where the sales reps can prioritize which are the important leads in this case it's not a very good lead we have the LinkedIn account we have the job title of the lead we have the lead summary we have a website URL with the company LinkedIn we have the company size the

industry company summary the decision maker of the company decision maker LinkedIn and a decision maker summary and finally we have a cold call script a personalized cold called Script so you can see this one is personalized to them hi Christina this is B from in this case I personalized it to my own company because of course I can't use the actual company and then it it generated spin selling questions which is a sales technique uh so we can see situation qu questions I wanted to ask how does clate currently manage lead enrichment and outreaching your

sales process and what tools or systems are you using to automate sales to as a clate to have a personalized skill script so this is how it's actually working if we go back to the lead agent we can also check here every tool what has been done for example leas research told the input was this the output was this I'll show you in more detail later before I want to give you a brief overview of the relevant AI dashboard if we go back here here we are inside relevance AI if you signed up for a

free account about the pricing by the way um you will see that the first paid plan is $200 don't be scared by that because you don't actually need that one you can go with the free version and where you get 100 credits per day and then you can buy extra th000 credits for $2 now I can tell you especially if you're starting out you can do quite a lot with a th000 credits um so don't worry about the the the pricing if you want to learn more about the pricing exactly let me know in the

comments and I'll give you some more information but it's not expensive to build your first AI agents um so here we are in my agents the these are all the agents I've built and you can see we have tools here we have knowledge so we can upload knowledge bases we have templates so here they have actually have many templates available already so they they even have some agent templates and but mostly they have like tool templates for example you can see extract data from PDF Google search analyze CSV so there's some very useful ones that

will save you a lot of time if you need to do specific things but in general I noticed that I mostly build tools from scratch because you always need to modify them to each specific use case so and then lastly here we have the activity center analytics where we can track these dashboards Etc and we have Integrations uh you can see they have some native Integrations with Google hotspot slack Outreach here you can put in your API key so if you haven't uh used relevance AI yet I'd highly recommend you put in your open AI

API key uh cuz then you'll get charged through open AI instead of relevance relevance AI which will put a small mark up on it so use your own API keys or if you like other um L&M they have them all available here so this is a little overview of the dashboard we go back to the agent so we have our inbound lead agent and here's really where we go to the agent setup so we have the agent name the agent description the agent description is more for you or if you start working with multi-agent systems

this actually describes to other agent what this agent does which sometimes can be useful but for now not very important and then here we have Integrations and triggers so they have some native Integrations right but I'm I actually integrated it and triggered it through a mate.com automation because I I want to show you guys later how you set that up because basically if you work with any other tool than this this will be the easiest way to trigger your age so that's why I set it up not directly with hopspot in this case but through

make.com because it allows us to basically integrated with any software we want then we have the core instructions which is the system prompt right now the system prompt is very important for this agent uh I use a structure I actually have a tool also available in the description below that is an agent prompting tool so all you will need to do is fill out some boxes and it generate the perfect prompt for for you I got this from another YouTuber uh Liam odley which is uh very good at the these things so that will help

a lot in terms of writing these prompts so in this case I have instructed it in the following way right you're an expert inbound lead manager efficiently researching incoming leads and companies you'll score the leads find all decision makers find decision makers in company and update our CRM with newly found information you also gener generate a personalized call script based on the research results you found you will you will be triggered right so I give it some context when a new contact is formed in hopspot and this is sort of your sop right the things

you have to do right so I give it the breakdown of all the things he has to do here are some extra rules right always start with this tool always end with this tool and some other rules for example this one you will only do research on the decision maker with the decision maker research tool if the if a decision maker was found and if the decision maker is different than the orig original lead because it was if it's the same lead if it's the same as the lead we already have done research on on

that person so this part in my experience is very important in agent prompting which is the SOP and tools so giving a good description of what each tool does and how it should use it it's going to be very important for the performance of your agent so in my case I've just described what each tool does and what it sort of returns back to the agent what information does it give back to the agent so this is just a description of all these tools and lastly notes notes are very good for if the agent if

you're trying the agent and it doesn't do exactly what you want for example it skips a step or whatever add it add it in a note because actually these language models there's been a study there uh done that showed that instructions that are given in the beginning or at the end of a prompt are being taken into account more than others so if you have a specific problem that's hard to solve try to add it always in the last part at notes and usually usually you will see that it resolves it pretty quickly now here's

another very good uh feature I think which is the flow Builder because as I said the most important thing for these agents is that it follows sort of this flow the right way right and although we giving that information already here in the in the system prompt like sop and tools we can sort of double down on it here with the with the flow Builder so as you can see what I've done here is you can choose between instruction so for example when you receive a lead always start with the lead researcher tool then always

score the lead with the lead scoring tool so you can actually fill this in and use uh a tool here so notice exactly what what what you want to do um you give it the next instruction C then update CRM with lead and Company info using update CRM with lead info then try and find the decision maker using uh find decision maker so in this case there can be multiple outcomes right so he can either not find the tool cannot find the decision maker or he can find and based on that the flow of this

agent is going to be different so we can add a condition here right so I've added a condition here if decision maker was not found or is the same as the original incoming lead continue here right where we basically go ahead and generate the personalized call script right away if the decision maker was found and is different from the original lead then research decision maker using this tool then update the CRM and the decision maker info and lastly we can get them together into the updates uh generate the personalized call script so I like this

feature a lot in my experience it improves the out uh the output a lot I can see actually have a mistake here because I need I need another uh step of generating the call script but as you saw it actually did it correctly so uh the the the system prompt was good enough then we have another interesting one here which is uh labeling tasks which is if you're really using the UI of these it can be helpful because you can have that list here on the side right where you see the tags of what it

has done right so if you have loads of of different tasks you can see exactly where each one is what it has done what it has not done Etc so basically we prompt that too right so we can here give it the name lead researched and scored whenever the lead was successfully scored by the lead scoring tool then it should label it like that so you can literally give it just an instruction on when to label it what then we have the tool section um so here we can add tools I'll show you later how

to actually build these tools I'm going to show you three or four of these tools how I've set them up um so the interesting here thing here is we can choose between three options right we can do auto run which I've done now with all my with with all my tools in this specific agent they're on auto run but you have do have the possibility of saying approval required so this is the human in the loop aspect that I mentioned before so you can decide he can never run this tool without previous approval now I

can show you maybe how this works after but it's for example in in some certain cases for example I had an an an AI agent for uh an Enterprise client who who sends personalized emails to very important clients and he didn't want to send those personalized emails automatically he did want to check before it was actually sent that everything was uh perfectly right so we could still automate a lot of his process like the lead research the writing of the email Etc but for him it's good to evaluate make some small changes before actually sending

it out so there can be very good use cases for this and then a third option is let the agent decide so you can even then prompt it to sort of instruct it when when it can use it by himself and when not for example in this case we could say maybe if if the lead score is less than 10 uh you can send the personalized email automatically because this lead is not extremely important but if the lead score is 100 and this is a great client then then uh we need approval from a human

before we actually send this personalized email so that's that's Tools in this case I've added quite some tools in general what I've seen is don't add too many tools to an agent either just like I said with the make.com the AI automations if there's too many conditions Etc it's better to give it to an agent now if you have too many tools inside of an agent an agent can get a little bit confused so my my experience also the 10 around 10 tools is becoming sort of on the edge for AI agents uh because it

will be harder and harder for him to understand these workflows and and to to make the right decisions so if you need have an even more complex process where you need even more tools then you would go to multi-agents right where we can see here sub agents so for example we can have ADD in another agent here and for example whenever has done this whole process we instruct the next agent to write a personalized uh sales proposal I don't know but you can see we can add in these sub agents to make even more complex

workflows and to automate entire sales process processes or or or workflows now lastly there's one more thing I want to show you is how you can actually deploy these agents if we go back to the dashboard here we can click on share where we have an option of for allowing sharing and embedding of your agent so we can either get a public link where we just open a link and you get the agent interface which if you deliver it to clients you could deliver it like this and they can manage their agents here themselves or

we can embed it on a website right so we have a an HTML code here which we can embed onto our website or their website so here have a quick example on my own website so you can see we have the agent interface here so that's really it on agents at the end of the video I'll show you how to actually trigger this right and how to work with the API how to trigger these Agents from make.com but first we're going to build uh our tools and because tools is really where the magic happens and

where you'll spend most of your time uh so if we go back to our dashboard we can go to tools so I'm going to show you three or four of the tools that I've set up for this specific agent so you get a better idea of how to build these so we can start with the lead researcher tool which is the first tool our agent uses so we can go here on edit right if you're creating a new one just click here on new and build from scratch for now we're just going to go to

edit again this is available for you in the template so you will see all of this and you can you can follow along if you if you want so here we have our lead researcher tool so we have the name and the Sub sub uh title now this is actually important right because this is this text the agent's going to read and by reading it it knows what this tool does so this subtitle is actually important so that the description of what this agent does it could be a little bit more in depth my my

explanation but it's pretty straightforward so it researches the lead in the company so this agent will understand okay this is what the tool does then we have our knowledge option so we can actually add a knowledge base to a tool um now in this case we're not using that but maybe after I can give you an example where we would use that for example if we create a personalized email uh email generator tool we can for example upload a knowledge base of our past emails uh and maybe some information on our company and we can

then feed that to the tool to make sure it mimics our style of writing and also has some extra information on my company but I'll show you that uh in another video I think in this case we're not using a knowledge base then the second step is defining our user inputs now user inputs is sort of the variables what are tool the variables that our to that our tool is going to use so in this case this company only has a lead name and a lead email right so we can only do lead research based

on those two data points so what I've done is created two user inputs lead name again we have to describe what this is in this case it's pretty straightforward but again the the agent is going to fill this in with information it received from hopspot so it has to Define okay name of lead and the agent will fill out this user input now the important part here is we save this um data into a variable so we call this variable contact name so we can change the name if we want but we've called it contact

name and also we have the lead email right and we save store that in a variable that's called Contact email now these variables are very important because those are those are the variables we're going to use in all of the next steps of this tool now we can decide to put in more or different types of user inputs so we can say long text input we have we can do files we can do tables uh Json checkboxes anything's possible but for now this is all we need and here you can also Define if it's required

or not so for example in this case we need both of these data points to actually do the research so they are required and we know we have them right if if they're not always in the next door you'll see one where we have an optional one where I can explain it so how does this work again it's a uh a task a rule based sequence of tasks right so what we do here is we you can see we have sort of a chain of tasks that this tool has to perform right so I divided

it up here I added in some notes so first we research the company and then we research the lead right so how do we actually do this so in this case we started with an L&M right an AI which is basically just a prompt right so in this case we've used GPD 3.5 because it's a very easy prompt it's not an difficult action and we want to save some money so what does this prompt do basically what we're trying to do in this prompt is trying to get the company website from uh this lead now

we already know we already have the email right and by far the majority of these emails that come through are work emails so we know that the domain part of the email is most likely the website so we have a simple prompt that says please extract the website of the following email so here's the the key thing right we've prompted it and put in a variable right with a variable put in but uh using these double brackets then we can just choose the variable so here we'll fill out the email and then I instructed to

only take the domain name and add https in front don't output anything else so here we get the company website as an output so I can show you this in an example so we can see step by step what this whole dos so I've just put in the example or we used earlier and I'm running the tool right now just to give you a better overview of how this actually works the tool is running right now and doing everything so we go step by step so that first step you can see the output of that

first step is the company website right uh so again we're going to store this in a variable Right company website so this link now is stored in this variable which we can then use in our next step so in the next step we're going to try and find their LinkedIn why because in their LinkedIn we have a lot of data available that is very useful to this company so we're going to do a lot of research on this company based on the leads LinkedIn and the company LinkedIn but first the tool has to actually find

those LinkedIn so what we do is we use a Google search step so this is already integrated into relevance AI it's basically a Google search API where we can Define the query so what kind of Google search does it have to do well in this case it has to look for LinkedIn Plus plus the website and we use the website by putting in the variable so basically it's just doing a Google search with LinkedIn for the website so the output of this tool of this step is basically a Google search result you can see we

have all the results here now what we do then in the next step is we do another AI step L&M step where we instruct uh the AI to find the company uh LinkedIn URL out of this Google search results so again we give LNM this variable inside of this prompt and we instructed to say please find the most likely company LinkedIn URL of this company of this company which again we put in the variable so if you want to look at this in depth I highly recommend uh downloading the the temp the free template because

then you can read these prompts in detail uh but you can see you're an expert researcher espe iiz analyzing Google search results and extracting the most likely company LinkedIn URL of the following company right where we put in the variable of the company and uh here below we give it the Google search result right so this basically stores all of this data so you can see the output of this tool is the company LinkedIn again we store it in a variable company LinkedIn URL and then we go to the next step which again builtin LinkedIn

uh is a built-in LinkedIn scraper already done by relevance AI so all we have to do here is um put in the URL of the LinkedIn now of course we have that stored in this variable so we all we do is add in this variable here and in this case it's a company we can choose between user profile user post or company profile in this case it's a company so you can see the output of this tool is a scraped data from a LinkedIn profile right so we have a lot of information here about the

company and here's basically the last step in our research on the company which is again we instruct the AI with all of this data to uh extract some specific data points like company industry company summary employees Etc so again here we FedEd the variable of the scraped LinkedIn profile now here's an important thing because in this case we want to have multiple different data points from this company for example in this case we have here company industry the company summary and the employees the amount of employees now because we have very multiple outputs in this

tool and we sort of want to separate these outputs we don't want to have one big sort of text that says this is the company industry because we can't use that inside of our CRM in in our CRM it's separated in different blocks so we want to separate our output so how do we do that this is a very important trick to know in relevance AI because it will help you a lot when you will come across a lot of these situations where you need to uh multiply the output and the best way to do

it is by instructing the L&M to say uh to Output your findings in a Json format you don't have to know what a Json is but if it does that you can see it's just basically a piece of code and it separates it like this now what we do then is we use a step that's already built in relevance a two which is convert a string to Json so we're going to convert this answer into a Json and you can see here we actually have multiple outputs so instead of one block of text that says

it services and IT consulting we have it separated so very easy here right how do you add that you just go convert string to Json right and there we put in the variable so that's a very useful feature um now now we have the company information but now we want also want to research the lead so we do a very similar process right we do a Google search but in this case LinkedIn with the name of the lead and the company website uh just to get it a little bit more clearer because sometimes you know

there's a lot of people called the same way so it will get confused we give it an extra data point and again similar prompt right to try and find the LinkedIn and you can see we found the LinkedIn here again we store it in a variable we scraped the LinkedIn and again we instruct an AI to extract the right data points from that LinkedIn profile we we instructed again to Output the findings in a Json format we add in the extra step to convert the string to Json to have it separated and that's it that's

our lead researcher tool so you can see here our tool output because we separated all of these um outputs is all of these data points company LinkedIn URL company industry number of employees lead summary company summary lead LinkedIn URL company website URL so we have everything here here we can actually decide do we put it manual or last step now we gathered a lot of information in the previous steps so that's why we're using manual because what we're doing here is defining our output variable meaning we need all of this information from the previous steps

into our final output we we don't want only this last step output because there we don't have the company data points for example so we do manual and then we can just select here so for example if I want to add another one I can show you quickly there a buggy H can't show you but yes that's that's how it works if you only need the output of the last you can just do last step so this is a a brief overview of how these tools work here you can easily navigate as you can see

and we also have some extra settings here we can do a run timeout for example if it runs longer than a certain amount of time uh it it times out we have our API keys and you have to play around with this a little to to get good at this uh you can see there's lots of things already in here but that's it like maybe I can show you some advanced settings of the L&M so here we have an the AI step right where we can actually Define which model we would use so this case

I've used GPT 40 we can we can decide exactly which model we use and also which provider so we can use a entropic we can use Google grock anything really now so you can see these tools can already do pretty powerful things and actually you can already you don't need agents you can also use only tools so if you maybe have a more simpler workflow uh you could also just connect a tool to a software or trigger trigger a tool from a soft software output the information we got from the tool back into another software

so we could set up in make.com for example I have one in my in one of my previous videos which is the lead researcher tool or uh lead lead qualification setup in those two setups I literally just connected uh a CRM with one of these tools and the information we found here we output back into the CRM so because it's a simple setup all we can do is build one to and that's it so how do you actually do that if you go back to use here we have two options which is running bulk this

is also an interesting tool I have another video on that how do you the how do you scrape LinkedIn in bulk we can simply add a spreadsheet here and and run our tool on a spreadsheet and basically output all the these these outputs into the into the spreadsheet and then we have the API where we can connect this tool to other softwares I'm not going to I'm going to show you how to set that up and also with the agent later in the end of this video but for now I want to show you two

or three other tools so we can go back and now I can show you maybe the how we update our CRM so after that tool run right and we have the research data we want to update that data with some lead info so in this in this tool what we've done is we've integrated with it with the h spot API so again we've instructed the agent with what this tool does right update CRM with lead info update hopspot CRM with the lead info right now in this case the user inputs we have many because all

of these data points we previously gathered from the research tool we're going to now put into this tool or the agent rather is going to put that into this tool to update our our hopspot so here we have our hopspot contact ID now that's important to identify which contact it has to update now this is the information the lead the the agent gets from from uh make Doom right in this case so this is what the information that the agent got in this case the hopspot ID so again the agent is going to fill that

out right so this one's required because it can only update that contact if it has the contact ID now the rest of them is optional why because sometimes the lead researcher tool not does not find all of this information Sometimes some information is lacking inside of their LinkedIn uh profile or sometimes they don't even have a LinkedIn profile so all of these are optional but the agent will fill it out if it has the information so we have the job title link lead LinkedIn URL lead summary company website URL company LinkedIn URL company size company

industry company summary and the lead score which was another tool we had but I'm going to show you right now so we have all of these uh all of these user inputs which we all again store in variables and then we're going to run our hopspot API call now how do you do that in this case hopspot has already has one right here but in general you can run you can trigger any API with the Run API request right so it's a very similar setup so in this case the method is patch because that's what

hopspot uses to update a contact we have our our account our hopspot account and the path this path you'll have to find and it will depend on the software how it's exactly set up in this case this is the path right CRM V3 objects contacts and then in the last part of the URL of the path we need to add in the Hop contact ID so in this case that's going to change for every every different lead so we put in the variable again here right hopspot contact ID variable so here we have the body

body is pretty straightforward here we have um the fields or the properties that are set up inside of hopspot so these properties you actually have to look up in in your own CRM or in in this case hopspot if you want to know how you do it in hopspot I can show you for example we have lead score here if we go edit columns we can go here to properties we're not going to this property but let's say we have lead score for we open it and here we can see the internal name right so

this is the this is the property name we have to use inside of our API call so if we go back so we have all our properties here and then we put in again our variables right which we get from here so this is all the information well it will update on this contact hpot contact ID so that's how it works very easy so that's how you sort of integrate an API if you want to learn more about apis and how to set them up let me know in the comments and maybe I'll make a

separate video on that now I want to show you one more tool which I think is interesting which is the find decision maker tool so the way this one is set up is in the following way so in this case we have set the user input only for company URL which is of course what our agent already found through the lead researcher tool so we have the company URL and with the company URL we're going to do a Google search for this company URL plus impressum so in this case this client operated in the German

market now in Germany you have uh legally you're required to have an impress page on your website where basically your decision makers and CEOs are listed so this is an advantage in Germany that almost all companies have this onto their website because they have to so basically what we're doing is a Google search so we can try it again with the with the same uh company so you can see how this works plate we can run it so you can see we get some Google search results back for that specific page and then again we

instruct an AI in The Next Step you're an expert researcher that specializes in analyzing Google search results and finding the impress and page right so you can see it found impress and page and then what we do is we use uh web scraper right also Building inside of relevance side already if you click here you see extract website content so what we do is of course we put in the variable of that page in here and we scrape the website and in this website information there should be a specific name which in German called G

shaft furer which is the decision maker of the company the most likely decision maker of the company so in the next step is like analyze this website and extract the name of the person who's mentioned as the gasa fer so you can see he found uh it found the decision maker and again now that we have the decision maker we basically rerun that sort of research Tool uh I showed you before on the decision maker to find his LinkedIn scrape the data of this LinkedIn give us a little bit of a summary Etc now these

are a few of these tools that we've set up for this specific agent but you can imagine the amount of possibilities you have by building these tools I can show you one more which is not related to this agent but where we actually use uh a knowledge base let's say this one so this is one tool I use in my inbox responder agent where I'm also going to make a a video on that one but in this case we've equipped it with two different knowledge bases so it's basically just a CSV file in this case

where one is uh FAQ about my company Force Factory so just some knowledge general knowledge and the second is a past email uh replies so I have two knowledge bases because what I do with the past email replies is basically let try and get the AI to mimic this style or tone of voice that I have when writing emails and in the in in the force Factory fa FAQ I just have some information about my company that if let's say someone has a specific example that's not publicly available it can use that knowledge base to

answer that question in this case our user inputs are email content some others tread ID message ID I'm not going to go in depth in this one I'll do that in the other video but you can see in this AI I use again these knowledge bases as a variable so past emails and knowledge right now you do have some extra settings here on how to use knowledge so we can go here how to handle too much content which is good because if you have a huge knowledge base we don't want to overload the AI and

also pay a lot of money to to get that information so we can go here and select most relevant data so it will do in in that case it will do a vector data search which basically tries to find the most similar relevant data but I'll give you more information on the knowledge base in another video so there's one more thing that I want to show you that is very important in these tool building which are conditions so you can basically set up conditions for when a step should run or when a step should not

run uh which is going to be very useful and I'm going to show you for example we can we have actually a variable here set up which says scrape the this is the step with where they where they scrape the LinkedIn profile right so in this case we have uh a variable set up which you do right here sorry right here and you do condition right so add condition will it will say normally so I already have the condition here so my condition basically says company LinkedIn URL which is this variable includes https so why

did I do this right basically it says only only run this tool if the output of this one and remember this tool was looking for the LinkedIn profile based on the Google search results so only output only run this tool if the output of this one includes https why if the this previous step doesn't find a URL of the LinkedIn it will probably say something like LinkedIn profile not found actually I instructed it here to say company LinkedIn not available right so if you can't find the LinkedIn URL of a company because it doesn't exist

uh or whatever it will output company LinkedIn not available and in that case the LinkedIn scraper will not run because the answer doesn't include htps right um so why do I do that because if it actually starts putting in let's say it didn't find the LinkedIn and it says company LinkedIn not available it will fill that that in into this variable and then there will be an error in my tool right because of course it can't run a LinkedIn Scraper on a non- LinkedIn URL or just some text so the tool will run into an

error and basically the system crashes right so these variables are very important uh to to know how to set up so they're done with JavaScript I think actually they they added in a new feature where you can do add conditions yes so we can select a variable contact name does not equal right so we can set it up like this or we can do it ourselves because in this case we don't have the includes option so that's why I wrote it myself but yes you can see you can do it without code here right now

lastly I want to show you how you can trigger your agent from any software you want really so for that we have to go back to the agent and here we go to agent setup where have where we have our triggers right so Integrations and triggers so in this case we go to API you have the information to connect our API so in this case we're going to use make to basically be able to connect it to to any software so in this case I made a scenario here and make so in this case my

scenario right gets triggered by contact form then it's being uh the contact is s to hopspot and whenever a new contact is created in hopspot my uh AI agent is triggered uh of course for the company this was not the same setup because they automatically got their clients their inbound leads into their CRM so all all the trigger was was hopspot to to uh to the AI agent so let's say you're working with a different CRM uh or different software all you do here is create a new one like let's say pipe drive another popular

CRM right I don't have a pipe Drive account so I can't I can't show you in detail but let's say uh whenever a new contact is created so we can say watch leads triggers when a new lead is added or updated right so we have that as the trigger right we set up the connection right and then basically triggers every time a new contact is created and then we have to connect it to our AI agent so in this case we're going to set up an HTTP model for relevance AI for our AI agent we're

going to go to the make request make a request and for relevance AI it's always going to be the method is always going to be post to trigger these agents so we can go back now to whatever sayi and the end point you find here we copy that we go back into make and we add that here in the URL and now we have to add two headers so we're going to add two headers the first header you you can find again here if you go back you can find here in the sample curl you'll

see content type application Json so this will be always the same so you can literally copy that so the first header is going to be content type and the value is going to be application Json then for the second header we go back and here you find authorization right that's the second header name and the value is going to be your API key so the API key you'll generate also from here you have here an option that's called generate API key if you click here an API key will be generated you copy and paste that

into the value of uh authorization right once youve done that you select here body type is raw the content type is going to be Json and here we're going to have to put in our request content or request body so we go back in here and we have our request body so that's where we're actually going to put in the message we're going to send to our agent when it gets triggered lastly we're going to have to put uh Parts response yes because I'm going to show you this request content in the module I actually

set up so you can understand it better so if we go here and we're in request content so as you can see here we have the same request body but I added in the variables from hopspot so first one we have role is user we can leave that as it is and the content is what we're going to send to um to our AI agent so in this case ouri agent you remember for our lead researcher tool what uh or what does our agent need it needs the hopspot ID now to update the data whenever

it run the the research Etc it needs the email and the lead name right so in this case I just typed in literally hopspot ID and then I added in the variable from hopspot ID right then I typed in email because this is this exact message will be sent to the agent and then I uh added in the variable of the email and Lead name with first and last name right so this message this is going to be sent to the AI agent I can show you here you can see it it it sends hopspot

ID it fills filled in the variable email and Lead name so you can even give some instructions to your agent in this request if you want um but we go back that's it so don't uh don't forget to add to click here on part response yes and now you have set it up uh it's pretty easy actually it's the same sort of principle for tools I can show you quickly if we go back to tools get a random tool we're going to have the same setup you can see we have the endpoint here we have

the request body the sample curl the only difference is with a tool you'll have to add in the variables exactly at these properties at these params but that's it it's the same same way of doing it so that's it for connecting it with an API with make you can really use any software you want there's one more thing you should know uh which is note that synchronous triggers only last for 30 seconds if your tool is running long running you'll need to make a call to the asyn synchronous end point so this is a little

bit of a different setup and it would be if you get an actual if you're waiting for an output of your tool to update or an agent to update a next step now in this case we're updating our CRM from the AI agent so we're not waiting for any output here inside make.com but if you do you have to set up um an asynchronous uh endpoint uh I will make another video on how to do that uh because it's a little bit different than this system all right guys that's everything for this video thank you

so much for watching please like And subscribe if you got any value out of this it took a lot of time to to set up and I'm planning to make a lot more of these types of videos where I show different use cases of AI agents and AI automations also if you have any suggestions on what I should build next please drop me a comment and uh I'll do my best also if you're looking for any Consulting my my link is in the description below or if you're looking for a bigger project for your company

uh you can go to my website forcea Factory ai.com and you can book in a a meeting with us uh thank you so much for watching and uh I'll see you in the next one