Free YouTube Transcribe

Video transcript

How to Create Your Own AI Assistant (It's So Easy)

Riley Brown · 4,315 words · 20 min read

Want to search this transcript, jump the video from any line, or download it as TXT, SRT, or VTT?

Open in the transcript tool

Full transcript

Introduction

0:00Hello there, Jaryy. How's it going?

0:03>> Hey there. It's going smoothly on my

0:05side.

0:05>> Jary, can you please open the menu?

0:08>> Sure. Let me click.

0:09>> Hey, can you actually change the

0:10background to pink?

0:13>> Okay, switching the background.

0:14>> Hey JY, show me the moods you can be in.

0:17>> Nice. Let's pop open the mood options.

0:19>> Hey Jary, can you change to angry?

0:22>> Okay, switching to an angry mood now.

0:24>> Thank you. Close the menu.

0:26>> Got it. Closing the menu and tucking it

0:28away. Jary, can you actually go into

0:30computer use mode?

0:32>> All right, switching into computer

0:33control mode now.

0:34>> Hey Jary, this is pretty cool. Yeah, I

0:36want you to open up the uh browser.

0:39>> Okay, I'll open your default browser so

0:41we can use it. The browser is open and

0:43ready for your next step.

0:45>> Hey, can you please open up Codeex?

0:47>> Okay, let's bring up Codeex for you now.

0:49>> All right, so this right here is my mini

0:51Jarvis that I created in three prompts

0:55using cursor. And this AI is connected

0:58to GPT realtime 2, which is OpenAI's new

1:02voice model. In this video, we're going

1:03to be building a custom Jarvis. And this

1:05is going to be a full voice agent that

1:08can use your tools and it will be able

1:10to fully control your computer. And for

1:13this, the tool that we're going to be

1:15using is Cursor. And you might be

1:17wondering, how much coding experience is

1:19required? Well, the answer is 0 out of

1:2210. You can be a complete novice to

1:24coding and you'll be able to create a

1:27voice agent for your personal use or for

1:30your business. But the really magical

1:32technology behind Jarvis is OpenAI's

1:36brand new GPT real time 2. And it is so

1:40realistic to talk to. The AI agent will

1:43just speak to you in real time. And

1:44we're about to create that right now. So

Step 1: Download Cursor & Set Up a New Project

1:47in order to get started, you should

1:49download Cursor if you haven't already.

1:51and you can find it at cursor.com

1:54and you can very easily download it for

1:56Mac OS. And then you're just going to

1:58sign in. When you first open up Cursor,

2:01it may not look like this. At the top

2:03right, it'll say open agents window.

2:05That's what I'm going to be using. It's

2:06just a different view of the same

2:08technology. And we're going to be using

2:10the agents window. So, it should look

2:12something like this. And all we're going

2:14to do to start out is create a new

2:16project. I am going to come to this

2:19workspaces tab. And here you won't see

2:22anything if it's your first time using

2:23cursor. And I'm just going to hover over

2:25this and click open workspace. And I'm

2:27going to hit open folder. I'm going to

2:29go into my desktop for now. And we're

2:31just going to create Riley Custom

2:35Jarvis. And I'm going to create this

2:37folder Riley Custom Jarvis and press

2:40open. Now I am creating an app at this

2:44folder location on my computer. Riley

2:47custom Jarvis. Okay. So now it's time to

Step 2 – First Prompt in Cursor and OpenAI API Key

2:50plan our first prompt. We are going to

2:52give a really long prompt to cursor and

2:55it's just going to build Jarvis for us.

2:57And we can use any AI model to create

2:59it. What's important for us here is we

3:02need to give it an OpenAI API key. And

3:05in order to find your OpenAI API key,

3:07you need to go to openai.com/api-keys.

3:11And so we are going to create an API

3:14key. And remember, keep this absolutely

3:16safe. And I'm just going to create a new

3:17secret key now. Create new secret key.

3:21Jarvis key. And we can create it in the

3:24default project. Now I'm going to hit

3:26create secret key. Okay. So I copied

3:29this key to my clipboard. And don't

3:30worry, I'm going to delete it right

3:31after I film this video. But make sure

3:33this stays completely safe and private.

3:35And so when I paste my prompt, I'm

3:37actually going to run it with my API

3:39key. I'll paste it right there. And so

3:40the prompt is relatively simple. Build a

3:43Jarvis-like desktop AI companion named

3:46Jary. Except let's go ahead and call

3:48Jary. Let's make Ricky. We're going to

3:50make one that's a completely different

3:51vibe than the one that I showed you

3:52earlier. The goal is to create a desktop

3:55AI companion that I can talk to

3:56naturally while it does useful work in

3:58the background. It should feel like a

4:00personal Jarvis, conversational, fast,

4:03visual, and able to use tools. Read the

4:06docs on GPT Realtime 2. This is going to

4:09be a desktop application. We're using

4:11the GPT Realtime 2 for live voice

4:14conversation. Let me interrupt, ask, uh,

4:17etc., etc. Give Ricky a minimal animated

4:20face. I'm going to say high quality

4:23animations, but just eyes and mouth,

4:28nothing else. Natural blinking,

4:30expressive moods, and speech syncs mouth

4:33movement. We're going to keep the UI

4:35simple, a small companion window, plus

4:37an artifact panel for anything visual

4:39and structured. Don't worry, I'm posting

4:41this whole prompt in the description if

4:43you want to try it out. When using GPT

4:45Realtime 2, it is a voice model that can

4:47use tools. And so all you have to do is

4:49ask cursor to create and add tools to

4:53the agent and it'll be able to just do

4:55all of these really cool things. And so

4:57one of the tools will be web search. And

4:59for this we're going to be using Exa.

5:01We're also going to be generating

5:03images. So this voice agent will be able

5:05to generate and edit images with GBT. I

5:09also want it to be able to create

5:10mermaid charts. This is a coding

5:13language that animates or it renders

5:16highquality charts. And so this is going

5:18to be really cool. I'm just going to

5:20copy this entire prompt. Now I'm going

5:22to go into cursor. Now what I'm going to

5:25do is I'm going to use GPT 5.5 high and

5:30I'm going to paste this prompt and I am

5:33now going to run it. So we are running

5:36this prompt and now cursor is going to

5:38build it. So now Cursor is probably

5:41going to take 10 to 15 minutes building

5:44this app. I'm going to go get a little

5:46bit of dinner and come back and we'll

5:48see what Cursor cooked up. All right,

Step 3 – Meet Ricky

5:50guys. So I just came back and it is now

5:53done and it created an Electron desktop

5:57app and here it is open on my computer.

6:00This is Ricky. He blinks. Wait a sec.

6:03Wait for it. There he is. This is Ricky

6:06and let's try it. Hey Ricky, how are

6:10you? Ricky, are you there?

6:11>> Yeah, I'm here. Loud and clear. What do

6:13you need?

6:14>> Wow, this is cool. Can you list out the

6:16things that you can do?

6:17>> Let's map out what I can handle for you.

6:19I can search the web, show results,

6:21notes, tables, charts, and images in the

6:23artifact panel, and keep simple local

6:25nodes and records. I can also help

6:27>> Can you display this in a mermaid

6:29diagram in the artifacts panel? Is that

6:31possible?

6:32>> I'll sketch the capabilities as a simple

6:34mermaid diagram in the artifact panel.

6:35Done. It did not render correctly. It

6:38says parse error on line 8.

6:40>> I'll simplify the node labels and

6:42rerender the diagram. So it parses.

6:44>> Okay. So it can create little diagrams.

6:47All right. Very very cool. This is

6:49amazing. Um can you search the internet?

6:52>> Yes. Give me a specific topic or

6:54question.

6:54>> Search um the best uh places to eat in

6:57New York.

6:58>> I'll pull a few reputable lists and

7:00recent roundups that summarize the

7:01common standouts. I can't access web

7:03search right now because the search

7:04>> Okay, I just realized I forgot to give

7:07you the exa API key. There's a lot I

7:10want to edit here. Hey, for the past few

7:11weeks, we've been helping companies get

7:13useful agents set up in their

7:15businesses. And these are agents that

7:17can be shared with your entire team. So,

7:19our team can come in, analyze all of

7:21your workflows, find where agents can

7:23save you time, and be incredibly useful

7:25across ads, research, content,

7:28reporting, internal docs, whatever it

7:30may be. We can locate where agents can

7:33help your business and then together we

7:35can build sharable agents that everyone

7:37in your company can use. These agents

7:39will have the right context, the right

7:40tools and most importantly the right

7:42permissions. So your whole team could

7:44just use agents where you already work

7:46because we can put these agents in Slack

7:48and we can even put them in iMessage or

7:50wherever else you communicate as a team.

7:53So if you want help becoming an

7:54agentnative company, the link is in my

7:56description. Reach out there. I'm going

7:58to disconnect from Ricky. So, we

8:00disconnected. Now, I need to make some

Step 4 – Second Prompt

8:02changes on cursor. Okay. So, I'm going

8:04to list out a bunch of things that I

8:05want you to change right now. The first

8:07thing that I want you to change is I

8:09want you to fix the mermaid diagrams.

8:11There was a parse error and I want to

8:13make sure that never happens again. I

8:15want you to fix that. The next thing

8:16that I want you to fix, I am going to

8:19give you my Exa API key and that's how I

8:22want you to make the web search work

8:24every single time. So, I'm going to

8:25paste that at the bottom of this prompt.

8:27And then for the notes, I noticed that

8:29it saves them locally, which is awesome.

8:31And then overall, I want a design

8:33overhaul. I want the site to have less

8:35like components. You know, there's too

8:37many like rounded edges on this thing.

8:40When I look at this, it's just kind of

8:42ugly. I want it to just have a simpler

8:44interface and have the buttons. We don't

8:46like those buttons like connect, voice,

8:48display, computer use, artifacts. Those

8:50can all be very simple buttons down at

8:53the bottom of the left panel. make the

8:55face take up like the left half of the

8:57screen like 50/50 and then the artifacts

8:59are just on the right. Have like a nice

9:01line that separates the left and the

9:03right. Okay, huge prompt. We're making

9:05like 10 changes here. Let's see if it

Step 5 – Add EXA API Key for Web Search

9:08does it. Right before I enter this

9:10prompt, I'm going to enter my Exa API

9:12key. So, here we are. We're using Exa.

9:15Exa is just a web search API. The same

9:17way we're using an API key to use the

9:18GPT real-time voice, we're also using

9:21one for Exa. In my opinion, it is the

9:23best web search API you can give any

9:25agent the ability to search the web

9:26really fast and at a really high

9:28quality. Name this project. Let's call

9:30this Ricky. We're going to create a key.

9:33And now I'm going to copy the key. And

9:35now I'm going to go ahead and paste it.

9:37And we are going to run the prompt. All

9:39right. So now let's wait for cursor to

9:41finish. Okay. So it is now done. Let's

9:45go ahead and check out ah our new

9:48updated Ricky. So, let's go ahead and I

9:51think I can just click this. Hey, Ricky

9:54boy, can you please create a mermaid

9:56chart for me explaining how artificial

9:59intelligence works?

10:00>> Okay, I'll sketch a simple flow of an AI

10:03system and display it as a mermaid

10:04chart.

10:05>> Hey, can you actually edit it to be a

10:07little more detailed and have like

10:08little complete sentence subtitles on

10:10it?

10:10>> I'll update the chart with clearer steps

10:12and short sentence subtitles. Done. The

10:15updated version has short sentence

10:17subtitles for each step, so it reads

10:19more like a process description.

10:20>> Very, very cool. I want you to please do

10:23a little quick web search for me. Tell

10:25me about the latest news in the World

10:28Cup and then explain it to me simply

10:30because I don't really understand

10:32soccer.

10:32>> I'll check the latest World Cup news and

10:35then break it down in plain language.

10:37Here's the simple version. The

10:38tournament is in the knockout phase.

10:40>> Wait, pause for a sec. Don't respond. I

10:42need to change you. Okay. So, I need to

Step 6 – Third Prompt

10:45go to cursor for the web search. I want

10:47it to render the web search as this like

10:51clean markdown. I want it to stream out

10:53and look really cool and futuristic when

10:55it when it does that. And right now, you

10:57just kind of pull up a bunch of tables.

10:58I want AI to like whenever I ask for a

11:00web search, I want it to render out in

11:03in text, you know, make it look good.

11:05Okay, I'll fire that prompt off. Hey,

11:07one thing I want you to do as well, I

11:09want to be able to ask you to show me

11:11the menu. So, I'm going to say, "Show me

11:13the menu," which allow just lists all

11:16the things that I can do just to keep

11:18track of all the things I can do. And I

11:19also want you to put some more time and

11:20effort into making the mouth move more

11:22realistically based on the different way

11:24the the sound comes out, making the uh

11:28animations look better and higher

11:29quality. So, it is done. Let's give it a

11:32try. So, let's go ahead and Hey, buddy.

11:35Uh, show me the menu. Hello, you there?

11:38Show me the menu.

11:38>> Menu is up in the Artifact panel. Want

11:40me to walk you through a quick starter

11:42like web search notes?

11:43>> Um, actually search the internet. Tell

11:45me about the latest about the World Cup.

11:48>> Got it. I'll pull a quick current

11:49overview from the web and summarize

11:51what's new. Here's

11:53>> Nice. Good job. Generate a photo of a

11:55fox.

11:56>> Nice idea. I'll create that image and

11:58pop it into the artifact panel for you.

12:00>> Let's go.

12:03>> Oh, nice work.

12:05>> Thanks. The image generation is still

12:07running in the background and it should

12:09appear in the artifact panel as soon as

12:11it finishes.

12:11>> You're doing a great job, buddy. Thank

12:13you.

12:14>> You got it. Happy to help anytime,

Step 7 – Add Computer Use Mode

12:16buddy.

12:16>> Hey, can you please show me the menu

12:17again?

12:18>> I'll pop open the capability menu in the

12:20panel for you. The menu is up in the

12:21artifact panel.

12:22>> Okay. Can you actually enter computer

12:24use mode?

12:25>> Sure. Switching you into computer use

12:27mode now so I can control the desktop.

12:29You're in computer use mode now.

12:31>> Can you open codeex?

12:33>> Got it. I'm opening codeex now.

12:36It works. Let's go. Good job, Ricky.

12:39>> Love to hear it. All set and standing by

12:41for the next move. What do you want to

12:42do in Codex?

12:43>> Can you type into Codeex to make the

12:46script better for the intro? I just want

12:48you to type it into the box there.

12:50>> I can do that, but I need your explicit

12:52confirmation before typing into an app.

12:54>> Yeah, I I give you my explicit

12:55confirmation.

12:56>> Okay, I'll type that into codeex now.

12:59>> Enter it in.

13:00>> I'll enter that now. If by enter it in,

13:02you mean press enter to submit that

13:04text? I need your explicit approval.

13:06>> Let's do it. Why do you need my explicit

13:07approval? Go.

13:09>> Okay. Submitting it now. Done. It's

13:11submitted.

13:12>> It didn't submit.

13:13>> It still didn't.

13:14>> When I go to use the computer use mode

13:16and I want it to type something, it asks

13:18me for approval every time. Is there a

13:20way to turn off that explicit approval

13:22mode? Are you able to do that? If so, do

13:24it. Let me see. I'll look at the

13:25computer use. I found the exact blocker.

13:27Computer type text reject sensitive

13:29monoton. Okay. Let's see if this works.

13:32When I enter computer use mode, I want

13:34it to the window to get really small and

13:36take up the bottom left. I want it to be

13:38like really small on the bottom left.

13:40And it should just show the little

13:42character on the bottom left. And the

13:43background of the whole app should be

13:45translucent. So, it should just be the

13:46little character, just the little circle

13:48at the bottom with a little button at

13:49the top right of it. That's brings it

13:51back to normal mode, like computer mode,

13:54because I want to see whatever's on my

13:56computer. When I go into computer mode,

13:57it should just get really small and it

13:59should just be the little character at

14:00the bottom left. Okay. So, let's give

14:02this a try here. Let's see. Hey pal, how

14:05are you?

14:06>> Steady and sharp. Ready?

14:08>> Nice. Nice. I need you to enter computer

14:10use mode.

14:11>> Got it. Switching modes so I can operate

14:13the desktop directly.

14:15>> Look at him. All right. Um, please can

14:18you open up codeex?

14:20>> I'll open codeex now from the desktop.

14:22>> Nice. Can you please uh type in a prompt

14:24here and say, can you please make this a

14:26little shorter and then fire off that

14:28prompt for me?

14:29>> Sure. I'll type that prompt and send it

14:31now. Done.

14:32>> Nice work. And that's amazing. Okay, I

Step 8 – Add AI Image & Thumbnail Generator

14:37just got a really good idea. I want you

14:39to be able to generate thumbnails of me

14:41using the GBT realtime image one. And

14:44then I want to be able to edit images.

14:46And so the way I want this to work is

14:48I'm going to give you an image of me and

14:49that should be used as an image

14:51reference. Look up the GBT image 2 or

14:53the the whatever the latest one is. Look

14:56up the documentation. When I want to

14:57create an image of me, basically I'll

14:59ask you to create a thumbnail and then I

15:00want to be able to edit the image. So I

15:02can say like edit it so that it does

15:03this. And then you should be able to

15:05edit it and change it just with voice.

15:07And I want to be able to generate

15:09multiple versions in parallel. So like

15:12what you'll do is when I say generate an

15:14image, I want you to generate them in a

15:16grid. So like I'm thinking like a three

15:19wide grid and it always generates in

15:21position one, the top left one. And then

15:22when I tell you to generate another one,

15:25they should all be numbered. And so I

15:26could say edit number two and make it

15:29blah blah blah. And then that one, the

15:31newest one, will always show up in

15:33position one. And so I can continuously

15:35edit these images. This is kind of this

15:37continuous process for editing

15:39thumbnails. This will be sick if we get

15:41it right. Let's see how cursor does

15:43here. Wait. And I just actually entered

15:46a photo of me. I realized it's going to

15:47need a photo of me. All right, let's see

15:50if this works. Let's open up our homie

15:53here. Hey, pal. How you doing? You

15:56ready?

15:57>> Ready. What do you mean?

15:58>> All right. I want you to generate a

15:59thumbnail image of me standing in front

16:02of a tree. Cinematic view.

16:04>> I'm generating a single new thumbnail on

16:06the board with your exact scene in mind.

16:07>> Okay. Generate an image of me holding a

16:12iPhone. The iPhone screen's facing the

16:15camera, so it's visible. And it shows a

16:18really cool app on the screen, and it

16:22looks like a camera, but with like a

16:27cool yellow border around it.

16:28>> All right, I'm curing a clean new 16x9

16:31thumbnail with your exact iPhone setup.

16:33>> Cool. Generate an image of me holding an

16:36a MacBook except the screen is visible

16:40showing a little blue character with

16:43cute white eyes.

16:45>> Number 14 is being generated now with

16:47the iPhone screen facing the camera and

16:49the yellowboard. Nice. I'll create a

16:51single new thumbnail with that visible

16:52MacBook screen and the blue character.

16:54>> Amazing. Okay. Can you pull up uh photo

16:5713?

16:58>> Sure. I'm opening number 13 so you can

17:00see it full screen.

17:01>> Nice. Can you please give him blue

17:05sunglasses?

17:06>> Got it. I'll make a quick edit to that

17:08selected thumbnail with blue sunglasses.

17:10>> Amazing. Hey, can you please pull up

17:12photo 15?

17:14>> Got it. I'll pull up number 15 full.

17:16>> Can you add text here that says cute

17:19character?

17:20>> Got it. I'll add that exact text to the

17:22selected thumbnail.

17:23>> Can you please pull up number 14?

17:25>> Sure. I'll open number 14 so you can

17:27view it full screen.

17:28>> Amazing. Can you please make the

17:30background outside like in the forest

17:32instead of like the blue to orange

17:35gradient?

17:35>> I'll update the background to a forest

17:37while keeping everything else

17:38consistent.

17:39>> Can you pull up 16?

17:40>> The thumbnails are still 16 by

17:42>> pull up 16.

17:43>> Got it. Pulling up number 16 full screen

17:46so you can see it clearly.

17:47>> Nice. Can you generate another one with

17:50um brown sunglasses?

17:52>> Nice. I'll create a new version with

17:54brown sunglasses using this selected.

17:56>> Very cool. And yeah, I want to do a web

17:58search real quick.

17:59>> I'll pop open the menu. Then you can

18:01give me the web search query. Menu is

18:03up. Tell me what you want to search for.

18:04>> Okay. I want you to please search and

18:07tell me the latest on the rumors about

18:10Nano Banana. I just saw Logan Kilpatrick

18:13tweeted about it. Tell me what you can

18:14say about that.

18:15>> I'll pull together the latest reliable

18:17info about those Nano Banana rumors and

18:20summarize what's actually supported.

18:22Short version, it's mostly speculation.

18:24I want you to show me a mermaid diagram

18:26of how our thumbnail tool and like like

18:28generating images work. I want to kind

18:30of see a diagram of how that works.

18:32Mermaid chart.

18:33>> Let me sketch a clear flow of the

18:35thumbnail and image steps in a simple

18:37diagram.

18:37>> Very cool. Thank you for showing me

18:39that. Can we go into computer mode?

18:41>> Switching to computer use mode now.

18:43>> Nice. Nice. There you go. So we created

Final Demo

18:47this Jarvis and we created it in one

18:51sitting using cursor with GPT real time.

18:54It also has the GPT image model built in

18:56and you can just ask it to create stuff.

18:58That's what I'm trying to show you in

18:59this video is you can just talk to AI,

19:01get it to create your own little custom

19:03Jarvis and you can even make a useful

19:05workflow. I actually enjoyed using that

19:07thumbnail image generator. It could use

19:09a little bit of work, but it's so much

19:11fun to create this little Jarvis. And uh

19:14yeah, it's just fun to use and it can

19:16control my computer. It can do a lot of

19:17things. Hey, Jarvis, go back. Uh leave

19:19computer mode. Actually, don't don't do

19:21it.

19:21>> Got it. Staying in computer mode.

19:23>> No, leave computer mode.

19:24>> Okay, leave in computer mode now. You're

19:27back in display mode.

19:28>> Nice. Good job. All right. Well, I'm

19:31going to take this entire project. I'm

19:33going to upload it on GitHub. And so

19:34you're going to be able to take the

19:36link, the GitHub link below and get feed

19:38it to cursor and say clone this and you

19:40could create this exact Jarvis or you

19:42can look below in the description and

19:44get the prompt and you can create your

19:45own version of Jarvis. And uh yeah, let

19:48me know what you guys create. I'm

19:49genuinely want to see what people can

19:51create and what types of Jarvises people

19:54are creating because you could just

19:55create Jarvis. I one thing I would love

19:57to see is someone create it connect it

19:59to their email and someone would connect

20:01it to just some some of their other

20:03tools. I think that would be really

20:04cool. And by the way, guys, I'm here in

20:06New York City. I have my entire team is

20:09here. Basically, all we've been doing is

20:10helping companies implement AI agents

20:12into their business. We will help you

20:15set up AI agents in your company. We

20:17will come into your Slack and we will

20:19add agents. You can set up OpenAI's

20:21codecs and quad you and you can just

20:23switch between the different models. You

20:25can also use the open source models if

20:26you want. We'll take a look at all of

20:28the workflows inside your business and

20:30we can help you create agents. The most

20:32useful agents that we are creating for

20:34companies right now are marketing

20:35agents, specifically ones that can like

20:37scrape high-quality ads and content from

20:40every single platform and then create

20:42versions for your company, controlling

20:44your ads and running your ads and

20:46managing that. And so Claude Code,

20:48Cursor, and Codeex are all individual

20:51tools. We want to come in and help you

20:53build agents for your entire team. And

20:55that means creating skills that anyone

20:57in your company Slack can use. And so we

21:00will help you set that up. And if that

21:02interests you, the link is in the

21:04description. Anyway, thank you guys so

21:05much for watching.

This transcript was generated from the captions YouTube publishes for this video. Get the transcript of any YouTube video atfreeyoutubetranscribe.com: free, unlimited, no sign-up.