00:00build this Grock 4 AI agent, it never forgets

00:02Do you ever wish your AI would remember everything that you told it so you don't have to keep giving it more and more context

00:08But with the brand new Gro for model, which is literally the most powerful AI model on the planet now, we can build this incredible memory agent system that will remember everything that you tell it

00:17So in this video I'm going to show you step by step how we can use grock for lovable and nitn to create this beautiful WhatsApp style chat interface in which you can talk to gro for this AI agent and it will never forget what you tell it

00:31Meaning you get a beautiful personal productivity assistant remembers everything you say your team say or your clients

00:36With that in mind grab that coffee let's dive straight in

00:37And a very quick tail

00:39Why do we care about Grog 4? Well it is absolutely smoked all the other language models and outperforms PhD students in every category

00:47And Grogall costs exactly the same amount as Claude, but it's five times better per dollar

00:50So let me show you how this system works very quickly

00:54So the way this system, this agent works, okay, is connected by web hooks to liveable

00:57So let me just show you very briefly how this works

00:59We'll ask this beautiful chat interface a question

01:02For example, how old Robert died? It then comes through to the scenario

01:05It will speak to grock

01:05It will store the memory

01:08It has the ability to search for internet

01:09Now crucially, what I've done is given its own functional database, which is perfect for projects such as this

01:14Of course, we could use rag, but we're actually going to build a specific dynamic thing where we can give it information daily

01:20It has two functionalities

01:22One is to search its memory database, and the second one is to create a record

01:25So, I'm going to give it a fictitious memory here

01:27And this memory could be the fact that I don't know, Jack really loves the color purple

01:32Please mention this when he talks about colors

01:35All right, cool

01:35So, when you're actually having this conversation, you can say things like this

01:39Hey, dude

01:39Um, I would love to purchase a new kitchen

01:42What colors would be great? So, let's trigger the scenario

01:47Come over here and ask this a question

01:49Cool

01:49So, we ask it a question

01:50It's now checking

01:50We come back over to our scenario

01:51It's running

01:53It's now checking Grock, which means it's actually using the system we want

01:56And as you can see, guys, it's also searched the memory, which means what this does in the system instructions, it goes over actually finds out what memory

02:02So, before it even responds, it will check your list of memory

02:04So, the idea of this is that you could have a conversation with it every day and actually it knows what you've said every single time

02:10So, you don't have to keep on giving it all the context

02:12And guys, just like that, if I come down, it's given me a beautiful answer

02:15It's also, by the way, research the internet to get some additional context

02:18So, it's also powered by the internet, which is ace

02:21But if I come over and have a look, guys, I can see it says, "Hey, you mentioned love in the color of purple

02:25I'll definitely weave that into my suggestions." We didn't have to mention it

02:27It's check this memory

02:27So, we're going to build this entire thing together step by step

02:30What we need to do first of all, guys, I'm going to grab over now the actual blueprints that we need, and I'm going to show you why this is super relevant in a second, right? I'm come over to every automation and grab this

02:39Cool

02:39So, we're going to download these two things

02:41We're also going to build these together

02:42the step by step

02:43So, we've got the Grock memory agent that we can just plug, play, and upload, which is great

02:45We just download that

02:47And then, crucially, what we also need to do is give a prompt to lovable

02:51Now, Lovable's really cool

02:51It's in the same category as Bolt

02:53We talk a lot about these AI operating systems

02:55What I wanted to do is simplify this

02:57So, let's just copy all the text in this document

03:01And what we can do is head over to Lovable

03:03Great

03:03And we can just paste it in

03:06If you want to create your own um specific thing, you can do that straightforward

03:10What's really cool is you can come down, give it the prompt

03:12What I've done is given it a load of coding for how I kind of want it to look

03:15But you can also these things down here, which is remix projects, which is fine

03:18One that you like, remix it, and then just adapt it for what you like

03:21Cool

03:21So, we give it a big prompt

03:21We're basically saying, "Hey, I want you to create me a beautiful WhatsApp style chat interface." We hit enter

03:26It starts building

03:28Then we come over to our brand new workflow

03:29I'll show you how you import it

03:31Then again, we will build this together step by step

03:32Let me just come down over to import from file

03:34You select the file and the whole thing populates

03:37Really freaking cool

03:37So what we're going to do though first of all building this step by step is come down here

03:42I'm going to go for web hook over here

03:44Okay

03:44Beautiful

03:44Okay

03:44And you're going to get this thing again

03:46This is what connects to Lable

03:47We copy this web hook just by clicking it

03:49We come off

03:51Of course, we're going to give this a name

03:52We'll call this one Grock memory which is ace

03:55I don't I'm so excited

03:55We get these big quantum leaps

03:57And the fact crushes it I think is really freaking cool

04:01So we come over to this

04:02Lovable is doing it

04:02And what we're going to say is something like this

04:05Hey, I would like you to send the chat messages to this web hook

04:07Cool

04:07And all you're doing, guys, is enter in that web hook

04:10And when it's done, we're going to hit enter

04:12And whilst that's waiting, guys, what we first of all then need to do, if I come over to our beautiful scenario over here, is we want to add in the AI agent and connected to Grock

04:19So to do that, we come over

04:20Let's go over here

04:22AI agent

04:22Beautiful

04:22We need to add a chat model

04:24So click on chat model

04:26Type in XAI

04:26Fantastic

04:26And you're going to need to add an account

04:30Now, where do I do that? You're going to head over to console.x.ai

04:34AI

04:34You come over to your dashboard

04:36Really cool

04:36Go over to your API keys and then you just create an API key like so

04:40Boom

04:40We hit this

04:40It'll pop up

04:43Give it a name like ABC or whatever you like

04:45So see remember what it looks like

04:46And then you can simply copy this to clipboard

04:47Come back over to your scenario here

04:49Uh click on this

04:49Click on this bit here

04:53And then we just simply let's just create a brand new connection

04:56Um let's let's do a new one here, shall we? Create a new credential

05:00There we go

05:00Fantastic

05:00And we're in

05:02Superb

05:04That's now all people connected and we can chat with Grock which is cool

05:06So let's just check that that works

05:07Let's just remove this and let's add in a little chat interface real quick

05:10Chat trigger

05:12Great

05:12Bring it over and we should be able to have a conversation with Grock in our new scenario which is great

05:18So come down in chat and we'll say hey man who was the tallest man to ever live and you can see now that is working perfectly

05:27And then we've got the information back immediately like lightning fast which is very very cool

05:32Beautiful

05:32So, what we want to do next is once we've done that is probably going to start giving it some tools

05:35So, I'm going to bring this guy over here for a moment and let's go check in on how Lovable Bull's getting on

05:39Cool

05:39So, now Lo's built this thing, we're just going to say, hey, I wanted to send it to this information here

05:43So, just send that one off so it comes through to the right web hook

05:46In the meantime, let's get the actual scenario set up and working over here

05:49So, for the meantime, if I just do it based off a chat and then we can connect to deliverable afterwards

05:54What's the stuff that we want? So the first tool we're going to add guys is Tavi which you can add by HTTP request tool and filling out this information

06:03Really really cool

06:03And basically you just do it with the headers

06:05If I click on this for example, hey use it to search internet

06:07Obviously you can just add a key for this

06:09I've just done this very quickly

06:10Uh value here

06:12You put your API key in from tab

06:14Again just come over to taby.com sign up get a free API key

06:17Very very cool

06:17Uh and then we've got the query to be defined by the model results 5

06:21So basically it'll pull down the five top results from Google and then you're ready to rock and roll

06:26So the next thing we want to do is give it the ability to search memories and also create memories

06:30So how do we do that? We're basically going to add in a table which obviously you can do here the same module then we just change it

06:35First one is to create a record

06:37So we come down

06:39So automatically you pick the list uh the base the table and all you need to do is just let the memory cuz this will pull through all the categories in air table which I've got memory

06:48It's the only one I've added in there

06:49Of course you can do this many different ways

06:50Uh, and I just said, hey, let the model define what the thing is

06:53Awesome

06:53That creates the memories

06:55Then we've got searching memories, which is the same thing, but operation is to search instead of create

07:00And it can do it from this bit here

07:03No need for a formula, and it will return all of the information, which is code for very, very cool news

07:07And now we've done that guys, we can basically just connect it together with the web hook

07:10So let's get rid of a chat message real quick

07:11Bring the web hook over

07:13Connect that guy up

07:15And then we need to send something back over to Liverpool to create our operating system

07:17So come down here to respond to web hook

07:19First incoming item is fine because it's just going to be the output from the agent

07:23Come over to the first web hook

07:25Instead of respond immediately, just put using respond to web hook node

07:29Beautiful

07:29What we need to do now guys is give the agent a prompt and instruct it on what we kind of want it to do

07:34So before we do that, I'm just going to test this workflow so we can actually get something from URL to make sure the data comes by

07:40To do that, I'm just going to say, hey, okay, message didn't go through

07:45Let's just make sure that that all is all fully active

07:48If I just stop the scenario real quick

07:50Come over

07:50Come on here and let's just listen for test event

07:52Come back over and we'll say one more time

07:54Hey

07:58Okay

07:58So, basically it's not sending the message to the web hook

08:00Very, very cool

08:01Let's just fix that very quick

08:03So, we just say, "Hey, information is coming through

08:05Make sure you send it to this address." Simple as that

08:06Cool guys

08:08I've just sent yo, and the message came back

08:09Little tip, by the way, when you're testing this, if you're isolating it, you need to make sure it's respond immediately

08:14But the news here is that we've actually got the message, which is cool

08:17So, we connect it like so

08:17Click on the AI agent and then we need to basically say hey this is the thing that you're responding to

08:23So all we do is we come down over here to message which is great and pop that in the prompt

08:30Fantastic

08:30Next thing I want to do here as well is just give it a bit of a memory

08:34So guys for the memory section we click on the simple memory and then define below I'm just going to put this JSON body date which basically means it'll remember everything we spoke about today and then we can actually store things into its long-term memory which is very very cool

08:46I will put context length as 10

08:47But and then if we come over to the agent just check out what his actual system message is

08:51So come down here which is basically it instructions comments expression just so I can full screen this real quick

08:57Beautiful

08:57Basically we just tell it everything we want it to do

08:59Uh what its job is it's called behavior protocol

09:02Basically every time it responds it needs to check its memory before it does that

09:07And if asked to it needs to create a new memory

09:08So, for example, now if I do execute workflow and I come back over to our beautiful new scenario, which is uh this one right here, right? I can say, "Hey, Bruce, what's my favorite color? Also, please record the fact that my favorite food is tacos

09:22Thanks." All right, hit enter

09:25Bam

09:25Now, this is working in the background

09:27We get this animation

09:28We come back over to our scenario

09:29We can see it's now checking in with Grock, which is awesome

09:31It's now searching the memory database to find out all the stuff that we said

09:34And then what it will do is create a memory for us

09:37So if I come over to our knowledge base for example, you can see here we've got Jack loves a cold purple

09:41Please mention it when it talks about it

09:42There we go

09:42Jack's favorite food is tacos

09:44This was mentioned on insert date

09:47So that's basically how it works

09:48And in the future, if I ask it for something like, hey, book a restaurant, it will know that Jack really likes tacos

09:54So let's go find Mexican restaurants instead

09:55And then we come back to the conversation here

09:56Basic conversation history

09:58I can see your favorite color is purple

09:59I've also recorded your favorite food is tacos

10:02You mentioned today

10:02Is there anything else that you would like to share or discuss? Now, really importantly, this is an AI operating system.