Hello there, Jaryy. How's it going?

>> Hey there. It's going smoothly on my

side.

>> Jary, can you please open the menu?

>> Sure. Let me click.

>> Hey, can you actually change the

background to pink?

>> Okay, switching the background.

>> Hey JY, show me the moods you can be in.

>> Nice. Let's pop open the mood options.

>> Hey Jary, can you change to angry?

>> Okay, switching to an angry mood now.

>> Thank you. Close the menu.

>> Got it. Closing the menu and tucking it

away. Jary, can you actually go into

computer use mode?

>> All right, switching into computer

control mode now.

>> Hey Jary, this is pretty cool. Yeah, I

want you to open up the uh browser.

>> Okay, I'll open your default browser so

we can use it. The browser is open and

ready for your next step.

>> Hey, can you please open up Codeex?

>> Okay, let's bring up Codeex for you now.

>> All right, so this right here is my mini

Jarvis that I created in three prompts

using cursor. And this AI is connected

to GPT realtime 2, which is OpenAI's new

voice model. In this video, we're going

to be building a custom Jarvis. And this

is going to be a full voice agent that

can use your tools and it will be able

to fully control your computer. And for

this, the tool that we're going to be

using is Cursor. And you might be

wondering, how much coding experience is

required? Well, the answer is 0 out of

10. You can be a complete novice to

coding and you'll be able to create a

voice agent for your personal use or for

your business. But the really magical

technology behind Jarvis is OpenAI's

brand new GPT real time 2. And it is so

realistic to talk to. The AI agent will

just speak to you in real time. And

we're about to create that right now. So

in order to get started, you should

download Cursor if you haven't already.

and you can find it at cursor.com

and you can very easily download it for

Mac OS. And then you're just going to

sign in. When you first open up Cursor,

it may not look like this. At the top

right, it'll say open agents window.

That's what I'm going to be using. It's

just a different view of the same

technology. And we're going to be using

the agents window. So, it should look

something like this. And all we're going

to do to start out is create a new

project. I am going to come to this

workspaces tab. And here you won't see

anything if it's your first time using

cursor. And I'm just going to hover over

this and click open workspace. And I'm

going to hit open folder. I'm going to

go into my desktop for now. And we're

just going to create Riley Custom

Jarvis. And I'm going to create this

folder Riley Custom Jarvis and press

open. Now I am creating an app at this

folder location on my computer. Riley

custom Jarvis. Okay. So now it's time to

plan our first prompt. We are going to

give a really long prompt to cursor and

it's just going to build Jarvis for us.

And we can use any AI model to create

it. What's important for us here is we

need to give it an OpenAI API key. And

in order to find your OpenAI API key,

you need to go to openai.com/api-keys.

And so we are going to create an API

key. And remember, keep this absolutely

safe. And I'm just going to create a new

secret key now. Create new secret key.

Jarvis key. And we can create it in the

default project. Now I'm going to hit

create secret key. Okay. So I copied

this key to my clipboard. And don't

worry, I'm going to delete it right

after I film this video. But make sure

this stays completely safe and private.

And so when I paste my prompt, I'm

actually going to run it with my API

key. I'll paste it right there. And so

the prompt is relatively simple. Build a

Jarvis-like desktop AI companion named

Jary. Except let's go ahead and call

Jary. Let's make Ricky. We're going to

make one that's a completely different

vibe than the one that I showed you

earlier. The goal is to create a desktop

AI companion that I can talk to

naturally while it does useful work in

the background. It should feel like a

personal Jarvis, conversational, fast,

visual, and able to use tools. Read the

docs on GPT Realtime 2. This is going to

be a desktop application. We're using

the GPT Realtime 2 for live voice

conversation. Let me interrupt, ask, uh,

etc., etc. Give Ricky a minimal animated

face. I'm going to say high quality

animations, but just eyes and mouth,

nothing else. Natural blinking,

expressive moods, and speech syncs mouth

movement. We're going to keep the UI

simple, a small companion window, plus

an artifact panel for anything visual

and structured. Don't worry, I'm posting

this whole prompt in the description if

you want to try it out. When using GPT

Realtime 2, it is a voice model that can

use tools. And so all you have to do is

ask cursor to create and add tools to

the agent and it'll be able to just do

all of these really cool things. And so

one of the tools will be web search. And

for this we're going to be using Exa.

We're also going to be generating

images. So this voice agent will be able

to generate and edit images with GBT. I

also want it to be able to create

mermaid charts. This is a coding

language that animates or it renders

highquality charts. And so this is going

to be really cool. I'm just going to

copy this entire prompt. Now I'm going

to go into cursor. Now what I'm going to

do is I'm going to use GPT 5.5 high and

I'm going to paste this prompt and I am

now going to run it. So we are running

this prompt and now cursor is going to

build it. So now Cursor is probably

going to take 10 to 15 minutes building

this app. I'm going to go get a little

bit of dinner and come back and we'll

see what Cursor cooked up. All right,

guys. So I just came back and it is now

done and it created an Electron desktop

app and here it is open on my computer.

This is Ricky. He blinks. Wait a sec.

Wait for it. There he is. This is Ricky

and let's try it. Hey Ricky, how are

you? Ricky, are you there?

>> Yeah, I'm here. Loud and clear. What do

you need?

>> Wow, this is cool. Can you list out the

things that you can do?

>> Let's map out what I can handle for you.

I can search the web, show results,

notes, tables, charts, and images in the

artifact panel, and keep simple local

nodes and records. I can also help

>> Can you display this in a mermaid

diagram in the artifacts panel? Is that

possible?

>> I'll sketch the capabilities as a simple

mermaid diagram in the artifact panel.

Done. It did not render correctly. It

says parse error on line 8.

>> I'll simplify the node labels and

rerender the diagram. So it parses.

>> Okay. So it can create little diagrams.

All right. Very very cool. This is

amazing. Um can you search the internet?

>> Yes. Give me a specific topic or

question.

>> Search um the best uh places to eat in

New York.

>> I'll pull a few reputable lists and

recent roundups that summarize the

common standouts. I can't access web

search right now because the search

>> Okay, I just realized I forgot to give

you the exa API key. There's a lot I

want to edit here. Hey, for the past few

weeks, we've been helping companies get

useful agents set up in their

businesses. And these are agents that

can be shared with your entire team. So,

our team can come in, analyze all of

your workflows, find where agents can

save you time, and be incredibly useful

across ads, research, content,

reporting, internal docs, whatever it

may be. We can locate where agents can

help your business and then together we

can build sharable agents that everyone

in your company can use. These agents

will have the right context, the right

tools and most importantly the right

permissions. So your whole team could

just use agents where you already work

because we can put these agents in Slack

and we can even put them in iMessage or

wherever else you communicate as a team.

So if you want help becoming an

agentnative company, the link is in my

description. Reach out there. I'm going

to disconnect from Ricky. So, we

disconnected. Now, I need to make some

changes on cursor. Okay. So, I'm going

to list out a bunch of things that I

want you to change right now. The first

thing that I want you to change is I

want you to fix the mermaid diagrams.

There was a parse error and I want to

make sure that never happens again. I

want you to fix that. The next thing

that I want you to fix, I am going to

give you my Exa API key and that's how I

want you to make the web search work

every single time. So, I'm going to

paste that at the bottom of this prompt.

And then for the notes, I noticed that

it saves them locally, which is awesome.

And then overall, I want a design

overhaul. I want the site to have less

like components. You know, there's too

many like rounded edges on this thing.

When I look at this, it's just kind of

ugly. I want it to just have a simpler

interface and have the buttons. We don't

like those buttons like connect, voice,

display, computer use, artifacts. Those

can all be very simple buttons down at

the bottom of the left panel. make the

face take up like the left half of the

screen like 50/50 and then the artifacts

are just on the right. Have like a nice

line that separates the left and the

right. Okay, huge prompt. We're making

like 10 changes here. Let's see if it

does it. Right before I enter this

prompt, I'm going to enter my Exa API

key. So, here we are. We're using Exa.

Exa is just a web search API. The same

way we're using an API key to use the

GPT real-time voice, we're also using

one for Exa. In my opinion, it is the

best web search API you can give any

agent the ability to search the web

really fast and at a really high

quality. Name this project. Let's call

this Ricky. We're going to create a key.

And now I'm going to copy the key. And

now I'm going to go ahead and paste it.

And we are going to run the prompt. All

right. So now let's wait for cursor to

finish. Okay. So it is now done. Let's

go ahead and check out ah our new

updated Ricky. So, let's go ahead and I

think I can just click this. Hey, Ricky

boy, can you please create a mermaid

chart for me explaining how artificial

intelligence works?

>> Okay, I'll sketch a simple flow of an AI

system and display it as a mermaid

chart.

>> Hey, can you actually edit it to be a

little more detailed and have like

little complete sentence subtitles on

it?

>> I'll update the chart with clearer steps

and short sentence subtitles. Done. The

updated version has short sentence

subtitles for each step, so it reads

more like a process description.

>> Very, very cool. I want you to please do

a little quick web search for me. Tell

me about the latest news in the World

Cup and then explain it to me simply

because I don't really understand

soccer.

>> I'll check the latest World Cup news and

then break it down in plain language.

Here's the simple version. The

tournament is in the knockout phase.

>> Wait, pause for a sec. Don't respond. I

need to change you. Okay. So, I need to

go to cursor for the web search. I want

it to render the web search as this like

clean markdown. I want it to stream out

and look really cool and futuristic when

it when it does that. And right now, you

just kind of pull up a bunch of tables.

I want AI to like whenever I ask for a

web search, I want it to render out in

in text, you know, make it look good.

Okay, I'll fire that prompt off. Hey,

one thing I want you to do as well, I

want to be able to ask you to show me

the menu. So, I'm going to say, "Show me

the menu," which allow just lists all

the things that I can do just to keep

track of all the things I can do. And I

also want you to put some more time and

effort into making the mouth move more

realistically based on the different way

the the sound comes out, making the uh

animations look better and higher

quality. So, it is done. Let's give it a

try. So, let's go ahead and Hey, buddy.

Uh, show me the menu. Hello, you there?

Show me the menu.

>> Menu is up in the Artifact panel. Want

me to walk you through a quick starter

like web search notes?

>> Um, actually search the internet. Tell

me about the latest about the World Cup.

>> Got it. I'll pull a quick current

overview from the web and summarize

what's new. Here's

>> Nice. Good job. Generate a photo of a

fox.

>> Nice idea. I'll create that image and

pop it into the artifact panel for you.

>> Let's go.

>> Oh, nice work.

>> Thanks. The image generation is still

running in the background and it should

appear in the artifact panel as soon as

it finishes.

>> You're doing a great job, buddy. Thank

you.

>> You got it. Happy to help anytime,

buddy.

>> Hey, can you please show me the menu

again?

>> I'll pop open the capability menu in the

panel for you. The menu is up in the

artifact panel.

>> Okay. Can you actually enter computer

use mode?

>> Sure. Switching you into computer use

mode now so I can control the desktop.

You're in computer use mode now.

>> Can you open codeex?

>> Got it. I'm opening codeex now.

It works. Let's go. Good job, Ricky.

>> Love to hear it. All set and standing by

for the next move. What do you want to

do in Codex?

>> Can you type into Codeex to make the

script better for the intro? I just want

you to type it into the box there.

>> I can do that, but I need your explicit

confirmation before typing into an app.

>> Yeah, I I give you my explicit

confirmation.

>> Okay, I'll type that into codeex now.

>> Enter it in.

>> I'll enter that now. If by enter it in,

you mean press enter to submit that

text? I need your explicit approval.

>> Let's do it. Why do you need my explicit

approval? Go.

>> Okay. Submitting it now. Done. It's

submitted.

>> It didn't submit.

>> It still didn't.

>> When I go to use the computer use mode

and I want it to type something, it asks

me for approval every time. Is there a

way to turn off that explicit approval

mode? Are you able to do that? If so, do

it. Let me see. I'll look at the

computer use. I found the exact blocker.

Computer type text reject sensitive

monoton. Okay. Let's see if this works.

When I enter computer use mode, I want

it to the window to get really small and

take up the bottom left. I want it to be

like really small on the bottom left.

And it should just show the little

character on the bottom left. And the

background of the whole app should be

translucent. So, it should just be the

little character, just the little circle

at the bottom with a little button at

the top right of it. That's brings it

back to normal mode, like computer mode,

because I want to see whatever's on my

computer. When I go into computer mode,

it should just get really small and it

should just be the little character at

the bottom left. Okay. So, let's give

this a try here. Let's see. Hey pal, how

are you?

>> Steady and sharp. Ready?

>> Nice. Nice. I need you to enter computer

use mode.

>> Got it. Switching modes so I can operate

the desktop directly.

>> Look at him. All right. Um, please can

you open up codeex?

>> I'll open codeex now from the desktop.

>> Nice. Can you please uh type in a prompt

here and say, can you please make this a

little shorter and then fire off that

prompt for me?

>> Sure. I'll type that prompt and send it

now. Done.

>> Nice work. And that's amazing. Okay, I

just got a really good idea. I want you

to be able to generate thumbnails of me

using the GBT realtime image one. And

then I want to be able to edit images.

And so the way I want this to work is

I'm going to give you an image of me and

that should be used as an image

reference. Look up the GBT image 2 or

the the whatever the latest one is. Look

up the documentation. When I want to

create an image of me, basically I'll

ask you to create a thumbnail and then I

want to be able to edit the image. So I

can say like edit it so that it does

this. And then you should be able to

edit it and change it just with voice.

And I want to be able to generate

multiple versions in parallel. So like

what you'll do is when I say generate an

image, I want you to generate them in a

grid. So like I'm thinking like a three

wide grid and it always generates in

position one, the top left one. And then

when I tell you to generate another one,

they should all be numbered. And so I

could say edit number two and make it

blah blah blah. And then that one, the

newest one, will always show up in

position one. And so I can continuously

edit these images. This is kind of this

continuous process for editing

thumbnails. This will be sick if we get

it right. Let's see how cursor does

here. Wait. And I just actually entered

a photo of me. I realized it's going to

need a photo of me. All right, let's see

if this works. Let's open up our homie

here. Hey, pal. How you doing? You

ready?

>> Ready. What do you mean?

>> All right. I want you to generate a

thumbnail image of me standing in front

of a tree. Cinematic view.

>> I'm generating a single new thumbnail on

the board with your exact scene in mind.

>> Okay. Generate an image of me holding a

iPhone. The iPhone screen's facing the

camera, so it's visible. And it shows a

really cool app on the screen, and it

looks like a camera, but with like a

cool yellow border around it.

>> All right, I'm curing a clean new 16x9

thumbnail with your exact iPhone setup.

>> Cool. Generate an image of me holding an

a MacBook except the screen is visible

showing a little blue character with

cute white eyes.

>> Number 14 is being generated now with

the iPhone screen facing the camera and

the yellowboard. Nice. I'll create a

single new thumbnail with that visible

MacBook screen and the blue character.

>> Amazing. Okay. Can you pull up uh photo

13?

>> Sure. I'm opening number 13 so you can

see it full screen.

>> Nice. Can you please give him blue

sunglasses?

>> Got it. I'll make a quick edit to that

selected thumbnail with blue sunglasses.

>> Amazing. Hey, can you please pull up

photo 15?

>> Got it. I'll pull up number 15 full.

>> Can you add text here that says cute

character?

>> Got it. I'll add that exact text to the

selected thumbnail.

>> Can you please pull up number 14?

>> Sure. I'll open number 14 so you can

view it full screen.

>> Amazing. Can you please make the

background outside like in the forest

instead of like the blue to orange

gradient?

>> I'll update the background to a forest

while keeping everything else

consistent.

>> Can you pull up 16?

>> The thumbnails are still 16 by

>> pull up 16.

>> Got it. Pulling up number 16 full screen

so you can see it clearly.

>> Nice. Can you generate another one with

um brown sunglasses?

>> Nice. I'll create a new version with

brown sunglasses using this selected.

>> Very cool. And yeah, I want to do a web

search real quick.

>> I'll pop open the menu. Then you can

give me the web search query. Menu is

up. Tell me what you want to search for.

>> Okay. I want you to please search and

tell me the latest on the rumors about

Nano Banana. I just saw Logan Kilpatrick

tweeted about it. Tell me what you can

say about that.

>> I'll pull together the latest reliable

info about those Nano Banana rumors and

summarize what's actually supported.

Short version, it's mostly speculation.

I want you to show me a mermaid diagram

of how our thumbnail tool and like like

generating images work. I want to kind

of see a diagram of how that works.

Mermaid chart.

>> Let me sketch a clear flow of the

thumbnail and image steps in a simple

diagram.

>> Very cool. Thank you for showing me

that. Can we go into computer mode?

>> Switching to computer use mode now.

>> Nice. Nice. There you go. So we created

this Jarvis and we created it in one

sitting using cursor with GPT real time.

It also has the GPT image model built in

and you can just ask it to create stuff.

That's what I'm trying to show you in

this video is you can just talk to AI,

get it to create your own little custom

Jarvis and you can even make a useful

workflow. I actually enjoyed using that

thumbnail image generator. It could use

a little bit of work, but it's so much

fun to create this little Jarvis. And uh

yeah, it's just fun to use and it can

control my computer. It can do a lot of

things. Hey, Jarvis, go back. Uh leave

computer mode. Actually, don't don't do

it.

>> Got it. Staying in computer mode.

>> No, leave computer mode.

>> Okay, leave in computer mode now. You're

back in display mode.

>> Nice. Good job. All right. Well, I'm

going to take this entire project. I'm

going to upload it on GitHub. And so

you're going to be able to take the

link, the GitHub link below and get feed

it to cursor and say clone this and you

could create this exact Jarvis or you

can look below in the description and

get the prompt and you can create your

own version of Jarvis. And uh yeah, let

me know what you guys create. I'm

genuinely want to see what people can

create and what types of Jarvises people

are creating because you could just

create Jarvis. I one thing I would love

to see is someone create it connect it

to their email and someone would connect

it to just some some of their other

tools. I think that would be really

cool. And by the way, guys, I'm here in

New York City. I have my entire team is

here. Basically, all we've been doing is

helping companies implement AI agents

into their business. We will help you

set up AI agents in your company. We

will come into your Slack and we will

add agents. You can set up OpenAI's

codecs and quad you and you can just

switch between the different models. You

can also use the open source models if

you want. We'll take a look at all of

the workflows inside your business and

we can help you create agents. The most

useful agents that we are creating for

companies right now are marketing

agents, specifically ones that can like

scrape high-quality ads and content from

every single platform and then create

versions for your company, controlling

your ads and running your ads and

managing that. And so Claude Code,

Cursor, and Codeex are all individual

tools. We want to come in and help you

build agents for your entire team. And

that means creating skills that anyone

in your company Slack can use. And so we

will help you set that up. And if that

interests you, the link is in the

description. Anyway, thank you guys so

much for watching.