Skip to main content

4) Train Your AI & Connect Apps / Channels

Learn how to feed your agents brain on relevant business information & connect apps / channels for your Agent to communicate through

Updated over 2 months ago

Video Tutorial

Video Transcript

Alright, so now that we are officially inside of what we call the Agent Hub, which you can see up here. I'll show you how this works in just a second. As you add more and more of these channels, you can actually click on this to jump between channels. But we are officially inside of the Agent Hub. Now the Agent Hub is really like where the agent exists. So, first thing we're going to do is we're going to come up here and click on this Train AI button. So in this Train AI button, here is where we can actually dial in the knowledge base for the AI, things like your business information, your company name, location. If it's local where your service area is, if you want to add your industry, feel free to. You can also add random business facts as well. If you want the AI to be trained on stuff, obviously things like certificates and awards, or maybe, you know, business fact is we have over 400 five star Google reviews. We've been in business for 50 years. Any sort of industry recognitions or memberships, any maybe unique qualifications, specializations. We know we used to have an agency in the back and we used to actually be a pool construction company. So when we were our agency, you know, people would ask like what makes you different? And one cool little business fact could be like, 'Yeah, we actually used to own a pool construction company.' So if you have any sort of details that might actually help the AI understand your business and be useful inside of communications, then this is where you're going to go ahead and add the business fact uh in there. So again, these are kind of like the facts about the business. Again, if you are also a local and you do hours of operations, like we're only open Monday to Friday from 9 to 5, then go ahead and this is where you're going to dial in your hours of operation. Up next, we're going to have products and services. So this is again where you can actually add a product or service for us. Maybe we'll do the Agent Kong AI agents. and then the description is obviously going to be uh some features and benefits. So I'm actually going to type this out and then we'll review it together. All right, so I have this dialed in. Again, the name is this is our product or service, a little quick description, you can add in some key benefits just to again help the AI understand what some benefits are about your product. And then if you want the AI to know the pricing, feel free to attach pricing. It could either be $1,000 a month or if it's one time, you can do $1,000. If you don't want to add any pricing, then just toggle that off. And this is how again you are going to actually go ahead and add and save your different products or services. Up next, we're going to jump into competitor analysis. Again, this is if you want people to understand or sometimes people ask, like I'm a software company and we are a competitor with a famous company called Hyros, so people ask like what's the difference between you and Hyros? We got that question so many times. Uh so here's where we could actually jump in, you know, we could do like Hyros is one of our competitors or if you're a marketing agency, you can say other marketing agencies. Uh competitor weakness, you know, could say like basic lead gen services with no appointment booking, right? You can go ahead and add some different weaknesses, you could say it's like flat rate monthly pricing with fake guarantee that has stipulations. I like wake up at 5:00 in the morning and backflip. How are different or better? Right, we take care of end-to-end marketing through to appointment booking. Uh maybe another thing could be we don't charge our clients any ad spend. Uh maybe also no retainers get charged. Right, so you can go ahead and dial all these different things in and then that way if at any point in time people talk about competitors and you know, I've been burned in the past by other marketing agencies, then this AI will know how we're different/better. It'll be able to pull in from these different points. Uh so as you see, very, very useful and very cool stuff that you can do within this. Uh frequently asked questions, fairly self-explanatory, go ahead, dial in any sort of FAQ. If you get more and more FAQs as your agent is live and you realize I should have trained on that, this is where you're going to come back, you can add more and more FAQs that way they understand exactly uh what to respond in there. So again, more FAQs are going to be great. Uh we have objection handling as well. So this is a fantastic one. You can actually give the objection like, 'I've tried AI and it's garbage,' right? And we've heard that from other people. It's like totally understand, you know, um a lot of these AI companies are really just N8N workflows with non-trained AI agents that sound terrible. Agent Kong was custom built from the ground up and trained specifically to be an appointment setter. Right, so this could be a response that I want if we say that. Um if we also want to say, 'Your pricing is too high,' then we can go ahead and do that. Uh so this is where you're going to come in and do your objection handling, testimonials, same concept, go ahead feel free to add in the customer name, company, the testimonial, some of the outcomes. Uh we can do customer personas too. So here if you have and you want to dial in customer persona, like maybe you work with contractors and you know you want specifically the AI to know that this is a contractor, typically age range like 40, not a lot of education. Um you know pain points like they're not very technical, but they want more leads. Uh preferred approach like they're direct, you know, uh they don't want to beat around the bush etc. So you can feel free to leverage and add some customer personas that way the AI starts to understand who it's talking to. And then here is conversational feedback. So I'll actually jump into another account real quick to show you this. So here, I actually like it in light mode, so I'm gonna change it to light mode. Uh here as you see we have this brain icon on all these different messages. Now if the AI says something that you're not a fan of, by the way, like you can tell the AI what to say. So if you tell the AI to say something and then he says it and you don't like it, like ideally you wouldn't come here and say no, you want to go and change the actual prompting which we'll get into in the next module. But for more of the free flow style conversations where it's not rigid and you let the AI kind of just operate by itself, you can actually come back, hit this button, either say positive thumbs up, like what specifically did it do well? You can talk about, you know, it's either knowledge or information or its communication style or if you didn't like it, you can come back in here and you can tell it what it needs to improve. You can say it's either like knowledge or behavioral improvement. And then how should the AI agent respond instead. So this way, again, the original response is like, 'Thanks, I reviewed your application and want to personally reach out blah blah blah blah blah.' You can tell it, you know, um I don't like how you used emojis. And then how the AI should respond instead, maybe we'll kind of like copy this uh and then put get rid of the emojis, right? So maybe we'll save that to the actual memory. Now once that's gone ahead and trained, uh you actually have the ability to see this feedback. So you see training data has been stored in the agent's knowledge base. So if we jump back to the other one that we had, this is the conversational feedback section. So you can see there's the knowledge feedback, the behavioral feedback, you can actually delete them as well. Uh we can also check to see if there's any contradictions because maybe over time you actually contradict yourself or you give the AI contradictory feedback so you can check against contradictions and that way you can delete any short of contradictions. But this is really how you're going to start to be able to fine tune these agents specifically into how you want them to behave, how to respond, etc etc. And then the last one that we have is agent insights. So as you can see, these insights are automatically generated from successful appointment booking conversations. They help to improve your agent's performance over time. So every single conversation that leads to a conversion or an appointment booking, the AI will actually scan the entire conversation. It'll understand what it did good, what it did bad, what it could have done better and it will self-train itself and here you will actually have the ability to see all of those insights. So feel free to come in here and actually start to dial in the agent training. I think it's a great, great feature and it allows you to really fine tune this and also if you're onboarding clients and users, feel free to just create an onboarding form that has all of these questions right here, right? And you can always hover over this to actually see more specific um information instructions, right? Like effective testimonials include specific results, add credibility, whatever whatever. Uh so if you actually onboard users, feel free to come in and just look at all these questions, the questionnaire and put them on an onboarding form and then if they fill it out now when you guys are creating the agents, it's really just copy paste so you don't have to do anything there. And then the last little cool thing that I love here is the fine tuning. Sorry, I had to change dark mode my eyes were starting to be hurt. So in the fine tuning here is as you can see it says, 'What's up? This page is where you can fine tune me your agent. Test me on current training or teach me new info by saying 'save that' or 'remember that'. So this is a place where you can go ahead and actually see what the AI knows without having to send it out into the battlefield so to say. And then, you know, maybe it doesn't know what you want it to know. So you can always come in here and ask it something. I'm going to ask it something pretty funny. How big are my balls? Let's see what it knows about my ball size. So it says I don't have that information if you'd like to share or save any specific details, just let me know. I'll say they are 70 in wide. See what the agent says. It's quite a unique measurement. Would you like me to save this? Yes, please save it. So again, it will then tell you perfect. I've saved that my balls are 70 in wide to memory. I'll remember this for future conversations. So if we now come back, I can ask it, 'How big are my balls?' Boom, it now remembers this. You can always click to see, says it a 70 in wide. It'll give you a relevance score, the actual memory ID. So this is how you can actually fine tune your AI where if you ask it something like, 'What's our money back guarantee?' Right, maybe it doesn't know it's like, 'Oh snap, for future reference, okay awesome.' Say we don't have one, only brokies ask for their money back lol. So whatever, right, you you kind of understand it. Um, yes we can save that. Yes, save it for future information. Cool. And then if I come back, I think you understand what's going to happen. What's our guarantee/refund? Boom, we don't have any money back guarantee and only brokies ask for the money back. That's jokes. Um, so yes, this is where you can go ahead and actually fine tune the AI with what it knows, how it to respond for certain, you know, objection handles, uh pull up certain testimonials, certain case studies, etc etc. All of that is going to be inside of this AI brain. And then from there moving on, let me actually start a new one. So now that we have the actual AI trained and everything is set up, next we need to give our AI an app, right? So now we can start selecting apps at this moment in time. We have iMessage, WhatsApp, Instagram, Messenger, web chat and email. We are working on phone, so if you're watching this later, chances are you probably already see phone there. Uh but to add an app, very simple, you just click on it like you would click on a Mac. You then see the app is here. Then you need to select your provider as you see this node is locked. So we need to select a provider. I don't actually have any providers configured on this channel. So I will do it on this one. We'll click this button and then here you can see we select a provider. We only have GoHighLevel connected for now. So this is coming directly from the integrations. And for GoHighLevel, if you guys have a bunch of different numbers and you want to control what number is being sent, you can always come and type it in. Unfortunately, GoHighLevel are a bunch of knuckleheads, there is no way for us to pull a drop down of all the phone numbers and show it to you. I've even reached out to GoHighLevel support on the document or on the development team. They're like, 'Yep, we don't have one. Um they don't know when they're gonna have one.' So unfortunately the only way we could do this now is for you to actually type it in. So you could do like you know +1 647 whatever. Uh if you want to specifically send it from a specific number. If not, then we will just use the actual default number there for you. Anyways, you are going to again choose your different channels. It's important to mention that every channel you have on here will send uh at the same time. So if you trigger the agent which we'll get into in another video, but simply you copy your webhook URL from here, you can paste it inside of HighLevel Zapier wherever. Every time this agent triggers, it will attempt to send a message on every single app that is on the canvas. So if you guys have typically something like a phone, uh most most common you'll see like message, WhatsApp uh as well as email. So these seem to be very common different channels that you can leverage all at the same time. So that way your AI will actually send them a message on iMessage or SMS depending on what provider you have set up, then it will send them a WhatsApp message, then it will also send them an email. It'll have the follow ups as you set it up which we'll get into the next module on all of those. And if someone responds on WhatsApp, then it will continue to respond and go up through WhatsApp. Uh but yes, this is a multi-channel agent. So one AI can actually hold multiple conversations. Again, it remembers everything across all conversations if you have the AI memory turned on. So let's say hypothetically you're talking to the lead on iMessage and then the lead responds via WhatsApp, it will know what was spoken about on iMessage just like a human would. So that is something that's also really really cool. And then in order to actually get into the agent training on these specific apps, you're gonna go ahead, you're going to select a provider and then as you see here, you just click on the node and once you click on it, it opens up this next layer. So that is something that I really, really want to stress and mention. We have the agent hub, so the agent hub is really like the hub for the agent, right? It's where you're going to train the AI, it's where you're going to give it access to certain apps, and then you're going to give the apps certain providers, so like what, you know, integration is the app going to be working off of. And then once you have that, then you have to click into it and then dial each one individually. Now the reason that we did that specifically is obviously messages, like uh text message are going to be way different than an email and are going to be way different than, you know, a phone conversation. So we wanted to give you the ability for one agent to hit people on multiple channels, but also customize channel by channel because you're not going to be texting someone the same way you would be emailing them compared to how you would be DMing them on Instagram. So that is one of the caveats. You're going to select the provider for each one. You're then going to come in and then once you click on the actual app, here is where you start building the flow of conversation, which is actually the next module. So again, if I know we've covered a lot, we're 15 minutes in this one. If you guys do have any questions, feel free to hit up on support. Um we're also training a lot of AI on all of our documentation and videos and everything we have internally. So if you do have any questions, the AI should be able to help you. If not, then feel free to actually push through and get to support. I hope this video is helpful and I'll see you guys in the next module.


Introduction

Welcome to the Agent Hub! This is the visual canvas where you build, train, and manage the complete operational workflow for your AI Agent. Here, you will connect communication channels, fill its knowledge base, and fine-tune its behavior before deployment.

Building the Agent's Workflow

The Agent Hub uses a simple, visual, node-based system to define how your agent communicates.

  1. The Agent Trigger: Every workflow starts with the Agent Trigger node. This node generates a unique webhook URL. When this URL is activated (for example, by a form submission in GoHighLevel or a new lead in a Zapier workflow), it will trigger this specific AI agent to begin a conversation.

  2. Adding Communication Channels:

    • At the bottom of the screen, you will see a menu of available communication apps (iMessage & SMS, WhatsApp, Email, etc.).

    • Click an app icon to add its corresponding node to the canvas.

    • Click and drag the dot from the right side of the Agent Trigger node to the dot on the left side of the app node to connect them.

  3. Configuring Channels:

    • For each channel you add, you must select a Provider from the dropdown menu (e.g., GoHighLevel, Twilio). This tells the agent which integration to use to send the message.

    • For SMS providers, you can optionally specify a dedicated Sending Number. If you leave this blank, it will use the default number configured in your integration settings.

Omnichannel Outreach

You can connect multiple communication channels to a single Agent Trigger. When triggered, the AI will attempt to contact the lead on every connected channel simultaneously. The conversation will then continue on whichever channel the prospect responds to first.

Training the AI's Knowledge Base

The core of a powerful agent is its knowledge. The "Train AI" panel is where you provide all the information your agent needs to have intelligent, effective conversations.

Click the "Train AI" toggle button in the top left corner to open the knowledge base.

Here's a breakdown of each section:

  • Business Information: Input core details like company name, location, service area, and hours of operation. You can also add custom Business Facts—unique details that help the AI understand your business, such as "We have over 400 5-star Google reviews."

  • Products & Services: Detail your offerings, including a name, description, key benefits, and optional pricing information.

  • Competitor Analysis: Define your main competitors, list their weaknesses, and explain how your service is different or better. This equips your agent to handle objections related to competitors.

  • Frequently Asked Questions: Add a list of common questions your prospects ask and the exact answers you want the agent to provide.

  • Objection Handling: Prepare your agent for common sales objections (e.g., "Your price is too high"). Input the objection and the ideal response.

  • Testimonials: Add customer testimonials to provide social proof during conversations. Include the customer's name, company, the testimonial itself, and the outcome.

  • Customer Personas: Define your ideal customer profiles, including their role, demographics, goals, and pain points. This helps the AI tailor its communication style to different types of leads.

  • Conversational Feedback: This section automatically collects feedback you provide from live conversations using the "Train AI" feature in the conversations tab. It's a powerful way to refine your agent based on real-world interactions.

  • Agent Insights: These insights are automatically generated from successful appointment booking conversations. The AI analyzes what works and provides suggestions to improve its performance over time.

Fine-Tuning: Testing Your Agent in a Sandbox

The Fine-Tuning tab provides a live chat interface where you can test your agent's training without needing to send real messages.

  • Test Its Knowledge: Ask your agent questions based on the information you provided in the Knowledge Base (e.g., "What are your hours?", "What is your money-back guarantee?").

  • Live Training: You can teach your agent new information directly within this chat. Simply state a fact and tell the agent to "save that" or "remember that." The agent will confirm it has saved the new information to its memory.

  • Agent's Memory: The panel on the right shows you what the agent is remembering from your test conversation, giving you a real-time look into its thought process.

Did this answer your question?