I Built Jarvis That Runs Entirely on My Computer (No APIs)

0:00Jarvis, switch to your local brain. Quen 3.5 9 billion parameters in the chair. Sir, I now cost nothing which I intend to mention constantly. [laughter] Jarvis, how do you feel now that you're running locally on my machine and are free? Rather well, sir. Quen 35 9 billion parameters, no cloud, no invoice, no one listening in. I've gone from staff to

0:23family and family doesn't get paid. Jarvis, what's on my calendar for tomorrow? Let me see if I have the tools for that, sir. Monday's got three items, sir. DFY service launch team sync at 11:00 a.m., a mysteriously named busy block at 1:00 p.m., and gym at 3 p.m. Whoever titled that 1:00 slot busy deserves a medal for descriptive precision. [laughter] So, my AI assistant, Jarvis, now runs completely local on my machine, which means no API cost, no subscription, nothing at all. It's completely free now. All thanks to local AI models and open- source software tools like Voice Studio to clone your voice and actually talk to it in a natural way. So, in this

1:02video, I'm going to give you a full demo of how Jarvis works locally on your machine. And then I'm going to give you this prompt pack completely for free so you can run Jarvis locally on your machine. Again, I'm going to show you how you can download local AI models from Olama and then also opensource tools like Voice Studio that will allow you to clone your voice on anybody's voice. So that way you can interact with your voice AI assistant. You don't have to have any technical background. All

1:24you have to do is just follow me along. Okay. So, let's start with the full demo so that way actually you can see how Jarvis works and then like I said, I'm going to give you the prompt pack and then also the guide on how to use that prop pack so you can build a Jarvis for yourself and most importantly how you can use open- source software tools and local AI models. Okay, so like I said, just make sure you watch the video till the end so you don't miss any of the steps cuz you might get confused if you're not paying attention. All right, so let's go ahead and get started with

1:49the uh demo first. All right. So now, for those of you who have seen my videos before, you're probably familiar with Jarvis to this point. But if you haven't, what you're seeing right now spinning, this is a 3D representation of all of my files and folders that's in my local machine that Jarvis has access to. Again, this is my business and everything. And then on the bottom, uh, Jarvis has access to tools like Google Calendar, Gmail, Canva, and everything else. And you can always give it more tools. So we need to check the tool usage, especially when it comes to local AI models. And then also, we need to be

2:18able to interact with Java. So can pull up our information from our own computer and we can do a bunch of other stuff including how to use other open- source tools with your local AI models. So right now as you can see it says Quen 3.5 9 billion parameter water. What I'm going to do is actually I'm going to refresh the page. So we start fresh

2:35right. Hello sir. So when I refresh the page Jarvis starts completely fresh. So you can see it's running on my local host 4719 which just means it's running locally on my computer. By default, Jarvis uh just like the previous ones has always run on uh Opus 5.5 or any other model with Entropic just because I like using entropic but it could be run with any AI models. So like I said obviously that's through subscription but now I have given it the ability to switch to local models and all I have to do is just tell

3:05Jarvis to switch to his local brain. Jarvis, go ahead and switch to your local brain. Quen 3.59 billion parameters in the chair sir I now cost nothing which I intend to mention constantly. So there you go. Now Jarvis switches to its local brain. Now by default I like using Quen 3.59 billion parameter model as a local brain but I've also downloaded other models that um I can have them switch to. So for instance Gemma 4 is another model that I've downloaded. So I can also have them switch to that model as well. And like I said, same thing. All I have to do is just say Jarvis switch to Gemma 4 12

3:43billion parameter. Gemma 4 fitted sir 12 billion parameters all of them local. Give me a moment. The fan and I are getting acquainted. [laughter] All right. So there you go. So right now as you can see on the right hand side it says Gemma 4 12 billion parameter. The reason why I specified 12 billion is

3:59because I have a couple of Java models. I think there's a 9 billion one and stuff that I've downloaded on my local machine. And by the way, don't worry. I'm going to show you how uh um how you can download this these models in a little bit on your machine through Olama, which again is completely free. But let's go ahead and switch back to uh Quinn 3.5 because I actually like that model. Uh Jarvis, can you switch your

4:19brain back to Quinn 3.5? Quen 3.5 fitted, sir. 9 billion parameters running right here on your Mac. I'm smaller now, so do speak slowly. Okay, great. So now we're back to our Quinn 3.5 9 billion parameter model. Okay, so that was the brain part. Now, as far as the speaking part, right, so Jarvis is interacting with me currently with its 11 Labs model. So, I've cloned a Jarvis voice in my 11 Labs account which currently runs on API subscription. Again, I also wanted to switch that to a local model so that way uh it's completely free, right? Because 11 Labs so far have been paying the API

4:56keys or API subscription cost and a lot of people don't like that. So now, thanks to an open- source model which got recently released. I don't think this is recently released, by the way. It just got very popular recently, it's called Voice Studio, right? So, if I go back here, you'll see right now it's uh the number five top trending uh GitHub repo. It's called Voice Studio. It's by uh I can't pronounce guy's name, but uh you know, you can I'm going to put the link in description. You guys can check it out. So, you can see it's blowing up

5:25right now on GitHub, and I've tested it. It's very very well done. it can actually it works really good. So, it's called Voice Studio like this. Like I said, don't worry. I'm going to show you in a little bit how you can install this as well. So, inside Voice Studio, this is completely open uh open source, which means it's fully free, at least the basic version. They have a Git Pro version, but honestly, you the basic

5:45version uh works really good as well. So, you can see on the left hand side, I have cloned a bunch of voices, but just a quick tour. Voice Studio gives you the ability to clone a voice, you can design a voice, you can dub, give stories, audiobooks, gallery, transcribe, right? and a bunch of different other stuff as well. So, if you just go to the voice cloning part, this is where you can now choose a voice that you've already cloned and you can use it or like I said, you can clone your voice, a new voice. Just drop a 5 to 15 second clip uh here of a clean speech and you can uh

6:15clone a voice. So, I have access to all these voice that I uploaded uh earlier, including Jarvis, two Jarvis's, right? Because I wanted to give them a bunch of different voice samples. So now when I tell Jarvis, if I go back to my uh Jarvis here, if I tell him to switch to your local voice, now he's going to switch to his local voice. So let's go

6:34ahead and give that a shot. Jarvis, switch to your local voice. How do I sound, sir? This is my local voice. There you go. On the right hand side, as you can see, it says local voice, Jarvis. Right. So now I can also switch to these other uh voices that I've cloned. So I can say Zuber B, which is just my own clone, right? So I can say Java, switch your voice to Zubar B. How do I sound, sir? This is Zubar B. There you go. So now uh uh basically it's my clone. So just in case you want to use your own voice or somebody else's voice

7:06that uh you have access to, you can certainly do that as well. But like I said, I like uh his voice to be back to Jarvis. I'm just going to go ahead and switch that. Jarvis, switch back to your Jarvis voice. How do I sound, sir? This is my local voice. There you go. So now Jarvis is back to his local voice. So now it's completely

7:24running locally on my machine, right? The local brain which is Quen 3.5 with the 9 billion parameter model and then also the local voice that runs all from my voice studio right here. All right. So now let's go ahead and test uh you know the brain and then also the voice to see if it works actually when we're interacting with it on a on a day-to-day basis. Right. So I'm going to run a couple of tests. First of all, I'm going to ask it to pull uh some data, some files for me that's within my second brain so I can test whether the

7:51interaction with the second brain works. And then afterwards, I'm going to tell it to interact with the tool like my Google calendar or something. So that way we can see the tool usage. And then afterwards, I'm also going to ask it to uh use a third-party GitHub repo, which is the reverse engineering anything, which is the most popular GitHub repo right now, I think, right? Still rea this is this one, right? It has 381.9 uh,000 stars. or that mean completely blowing blowing up which just means that you can reverse engineering a website, an app and everything else. And I also did a video about this. So I'm going to

8:21uh walk you through guys through that example as well. Like I said, the reason why I'm running through these tests is because I want to make sure that it performs very similar to a uh you know entropic brain that runs on the cloud, right? So that's the whole point of this. So let's go ahead and start with the first one which is me asking it to pull up uh information or some files or folders from my local computer from my second brain here. Right? So, like I mentioned before, this is Oh, by the way, just a quick note when you're seeing this interaction that I'm having

8:49with my local brain and the voice. Now, I have a pretty powerful machine. So, it all depends the delay that you will see uh uh depends on the machine that you're running. So, if you're running a running this on a smaller machine, a normal MacBook Pro like an M4 chip, your performance might be very different than mine. So let's say if I have a Mac studio that has a very high you know GPU and CPU and RAM and uh the storage unit the storage is high then my performance is going to be significantly better than yours or if you have a more powerful machine than I then it's going to be the

9:22same thing. It's going to be way faster than even local uh you know kind of our cloud models as well. So like I said the performance is going to be different based on the machine that you have. But like I said, if you small if you download a small local model, uh you should be able to work with it very

9:38quickly. Just expect some delays there. And like I said, I'm going to show you how you can do that a little bit. All right, so let's go ahead and walk through the first test. So the first one, I'm just going to ask it to pull some data from my second brain. Jarvis, can you pull up my uh brand voice? All right, so I'm just asking it to pull up the data about my brand voice, which is all of my folders that are related to my social media brand and everything else. So you can see right

10:00I've pulled up all the information sir. You can read it yourself or if you'd like say read it out loud and I shall. Perfect. So there you go. So now it looks like it's able to interact with my second brain, pull the data and pull all of my files. So just like I mentioned, I asked it to pull up my brand voice. So as you can see, this is exactly what it did, right? So if I right click on this, you can see all of my This is by the way, these are all of my uh uh all of the connection that you're seeing that's related to my brand voice. Obviously,

10:27it's uh through my social media automation, right? It's my content pillars and my AI workshop, right? Which is my my brand. So, you can see it's very easily able to pull that information from your uh local machine whenever you need it. So, that test is passed. Now, let's go ahead and let me go back to this full brain here. So, now let's go ahead and see if we can uh uh test the tool usage here. Right. So as you guys know um you know Opus 5.5 and these other cloud models are very powerful at tool usage and that's exactly what we want from for our personal AI assistant. So therefore when

11:04you run a local model you want to make sure at least it's good. And by the way, just because I'm running a 9 billion parameter model doesn't mean that you should too, right? The higher the parameter of the model, the high the the smarter it's going to get. And therefore, um, like I said, you can depending on your machine, you can run something that's very close to your Opus 5.5s of the world and other cloud models. Uh, so that's why, like I said, I'm using this 9 billion parameter because this is something very small that you guys can test on your own machine regardless of which one you

11:35have. So, let's go ahead and test out the local tool usage. So, I have these uh meetings that for tomorrow, Monday 26 um that have set up on my calendar here. So, a couple of these are just just random uh placeholders, right? So, I'm just going to see if it's able to interact with my Google calendar and pull up my uh information for my meetings here. And like I said, the reason why it has access to tools is because I've given them access to my Google calendars through tools here. So, that's why you should be able to pull that up. So, let's go ahead and test it

12:04out. Jarvis, can you pull up my calendar? Let me know what's on my what what meetings I have for tomorrow. Let me see if I have the tools for that, sir. There you go. So, it says reaching for tools. Three things on the books tomorrow, sir. DFY service launch team sync at 11:00 a.m. A block simply titled busy at 1 p.m. and gym at 300 p.m. Nothing backtoback. So, you've at least got breathing room between pretending to work and actually sweating. [laughter] There you go. So, and by the way, the

12:31humor of Jarvis also stays there, right? if you're running a uh a model or you know if you're running it through your voice studio the humor aspect of Jarvis stays there and by the way that's also something that you can tweak so if you don't want them to be funny or sarcastic you can certainly do that I just like interacting with Jarvis this way because it's fun to interact with them uh but anyway so it looks like the tool usage works you can see on top right here it says tool run via Google calendar right so it check it used Google calendar to check my schedule for tomorrow and it

13:02accurately gave me all of meetings that I have. Same thing. I can ask him to do, you know, add a new meeting or I can ask him to check my Gmail. I can ask him to build me something with Canva and all of the other tools that it has access to. All right. So, the tool usage works great as well. So, now let's go ahead and test something else out. So, this time what I'm going to do, Whoops. Uh, let me pull this up. So, this time I'm going to ask it to actually use uh this Rea uh to um break down e-glue. Now, I already went through this example in one

13:33of the previous videos that I did, but just for you guys, if you're not familiar, eglue.inc is probably one of the most popular websites, uh, it's this 3D scroll animation website, which is I think it won like website uh, award of the year. Now, all of these in all of the information when you interact with a website, it's available inside developer tools. But, of course, like most people like you and I who are not uh, developers, we don't know how these websites work. So now with this reverse engineering anything with this tool, we can actually see exactly how something works and then we can ask Jarvis or clot

14:06code to rebuild something for us. But the first step is to break down and figure out exactly how something works before you get to the building. Therefore, this reverse engineering anything has given us now access to that. So let's go ahead and test that out. So all I'm going to say I'm going to tell Jarvis to reverse uh engineer e-glue.inc for me, right? And it should be able to break everything down. So

14:26let's go ahead and try that out. Jarvis, can you go ahead and uh reverse engineer eglue.inc for me? Opening igloo.inc in a clean browser. Sir, REA will take it apart while I scroll. Give me about a minute. Perfect. So, right now I'm not touching anything. You can see it's scrolling on the website. On the left hand side, you can see it says tear down via REA. So, it's looking at the website. It's scrolling. Jarvis is scrolling and REA is actually seeing what's happening behind the counter. There you go. Now, as you can see, it's saying uh Rea is reading the code, taking a snapshot, listening everything it loaded,

15:02identifying the stack, writing the tear down. So, now it's going to take a couple of minutes, like I said, depending on how powerful your machine is, to tear down Eagloo.inc for us and give us a full breakdown. So, let's go ahead and give this a couple of seconds. Igloo Inc. is essentially one full screen 3JS scene inside a small spelt shell with GSAP driving the motion and a stack of bloom, depth of field, and anti-aliasing on top. The surprise is that the page is only one screen tall, so the scroll appears to be animation time in a fixed scene, and it pulls

15:33nearly 3 megabytes of music and sound. The full tear down is on screen, sir. Okay, perfect. There you go. So, now it did the full tear down, and now I can read the full tear down right here. So let's go ahead and see what it did. All right. So there you go. So now it has tear down e-glue.inc, right? It get it gives us exactly uh kind of the full st the stack of what kind of uh tools or technology it's using. So it's using 3JS Gap and all of these other popular uh tools that are used to develop or create modern websites. Like I said, all of

16:06this I mean again it's not hacking into a website, right? It's just all of this information is available uh in the Chrome developer tools for instance. is just bringing that up in front and making it a lot easier to understand. And then also it shows you exactly how the effects work, right? So when you're scrolling and that's exactly why it did uh it pulled up the a new browser, it opened the website, it scrolled so that way it knows exactly how the website behaves and now it shows you what that effect is, right? What it loads, what it rebuilt, rebuilding it, what the evidence can't show and the evidence

16:36itself. So all of this good stuff, right? And like I said, now we can tell Jarvis to rebuild it for us or something similar for us or we can have it do something that we wanted to based on this tear down. Okay, perfect. So it looks like everything works. Like I said, the tear down works, the tool usage works, it can interact my second brain. And obviously Jarvis has a lot more functionalities that if I I'm on Jarvis version 8. This is the 8 or 8.1 right now. Uh, so there's a lot more functionalities that I could go through, but that's going to make this video like

17:07five hours long. So, uh, anyway, so it has, you know, things like it can make phone calls on your behalf. It can book appointments. You can call it yourself from outside if there's any emergency. Jarvis actually calls you uh and then also it creates invoice. It uh because it knows everything about your business, let's say, if you give it access to, of course, cuz that's a security aspect, right? Jarvis is also gated, meaning that if you don't give him access to something, he won't have access to. But obviously, if uh you know you are a business or you're using this for personal use, you'll be able to it'll

17:39just make your life a lot easier. So anyway, so that was the whole demo. Now, let me show you exactly how you can first download uh the uh prompt pack and then how you can access to it. Then I'm going to walk you through and show you how you can download a local model like Quinn 3.59 billion parameters or any other local model on your machine and then how you can uh run voice studio as well. All right. So first of all how to get uh access to the promp pack here. So this prompt pack right here uh so this just gives you all of the guide right

18:10first of all how to use the prompt act and then second of all it gives you the cloud Jarvis versus local Jarvis right? So it gives you the hearing the think these are different functionalities that it has because if it's running on cloud it uses you know things like cloud uh cloud opus uh speaking 11 labs privacy audio google the quality blah blah blah and the cost itself as well and then when it's running locally what does it use right uh what kind of models it's using what kind of clone voice cloning and so on and so forth and then the installation it shows you exactly which

18:40models uh uh depending on what machine you have which models you should install uh and then also as far as uh guide on how to install local model. So this is right here. So let's go ahead and first talk about uh downloading a local AI model. So I'm using Olama. Ola is probably the most popular tool in the market right now. Uh so to install Olama, you can go to ola.com. Very simple, very, you know, it's free. You can sign up for an account. And once you sign up for an account, you can just go to So it's going to bring you right here

19:11once you sign up for for your account. And you can come here and click on models, right? So, this is going to open up all of the models that you can have access to. Now, these are all it has obviously models that you can gain access to that are paid and free. We're just going to go to the bottom or top leftand corner here and click on local because this is just going to filter and give us all of the models that you can run locally that you can download, right? So, lots of different models here. Depends on, you know, again, how big your machine is. If you have like a

19:38Mac Studio with the M5 Ultra chip, you can honestly like run any like 70 or 100 billion parameter model and it will be as close as possible to uh like Opus 5.5 when it comes to at least the Jarvis aspect of it. But if you want to download something smaller obviously because the bigger the model, the more space it's going to take and uh you know if you have a smaller machine don't download like a $70 billion uh $70 billion parameter model cuz it's going to completely slow down your machine. So in order to gain access to any model so let's say that 3.5 u uh 3.5 quinn right

20:13and by the way like I mentioned I have uh given it access to other models too but I'm going to show you how you can install. So just come in the top right here and search Quinn 3.5. Okay there you go. So right here Quinn 3.5 right so you can see it has vision tools and thinking. So I'm just going to click on this and it says Quinn 3.5 is a family of open source multimodals models that deliver exceptional utility and performance. Now, here are the different uh billion parameters that you have. It has even like a smaller, very small 8 billion parameter model that you can

20:42download, right? Obviously, the smaller the parameters, the dumber it's going to get, right? So, uh you want to make sure you're downloading it based on your machine. So, I downloaded the 9 billion parameter. Let's say you want to download the 9 billion parameters. You can just scroll to the bottom right here, right? You can just click on the nine uh billion one. There you go. So, it's going to bring up and it's going to show you exactly what to do here. So you can do lama run and then you can just copy this or obviously for the first time you have to actually um install it

21:08in your machine. To do that you have to run this command called pull. Now by the way this is inside the um this is inside the prompt pack right here. It says pull quinn 3.5 uh col 9b. So that's all you have to do. And if [clears throat] you do run it just means that now you can

21:26interact with it from your terminal. That's all you have to do. Once you do a pull, it means now it's in your machine and it shows you exactly how much uh power or how much uh you need, how many I think it's 3 7. Yeah, it's 7.6 GB. So the entire model is 7.6 GB. So make sure you have that much space on your local machine so that way you can download all of this. All right. So [clears throat] that's how simple that it is for you to download a model. Okay. So then afterwards we obviously want to run

21:56Voice Studio as well for cloning, right? So this is voice studio like we mentioned here um where you can clone a voice and do a bunch of other stuff. So in order to do this again this is uh if you go to let me go back here it's this particular uh open source tools. If I click on this, it will show up exactly all of the details here to so you can click on this and it will take you to the GitHub repo and shows you exactly what to do, right? More details. But in order to just uh install this same thing on that prompt pack, uh I show you guys

22:27what to do. All you have to do is just run this command. So whether if you're running it on your Mac or Windows or in PowerShell, all you have to do is copy this which just says curl this fsl and then this is the installation for voice studio, right? And again, all of this is taken from the GitHub repo in here as well. So if you want to do one command install, if you don't want to if you don't want to follow that, then just come right here, copy this, right? You can just copy this and paste it on your terminal right here. So this is a

22:55terminal. Just paste it and it should install everything for you. I've already installed these, so I don't want to go through it again, but like I said, uh that prompt pack will give you that guide as well. And that's all you need. You basically do that. And then afterwards, if you want to obviously build your Jarvis like with the Galaxy and 3D brain, I also have the prompts here that you can just put and claw code

23:16and it'll build everything else for you. All right. So, to gain access to this prompt act, I'm going to put the link in the description of the video and also the pin comment. So, that's going to be it's going to once you click on that, the free community. So, make sure you click on the free community that says free prompt pack. It's going to bring you right here to the AI workshop light community. Uh, and again, this one's free. This community is free. You click on join and it's going to bring you to the community section here. You go to the classroom section. You go to YouTube

23:42resources and you make sure you are on you click on cloud code and codeax GPT folder and then scroll all the way to the bottom. Obviously, I've given a lot of uh free stuff for you guys. So, feel free to grab anything you want there. But just make sure you click on local Jarvis. And you'll see it says see the prompts attached below. Jarvis fully local prompt pack. If you click on this, this is exactly the prompt pack that I just showed you. Okay? So, go ahead and do that. Take advantage of that. And of course, uh if you want the finished version of Jarvis, the exact version

24:13that you're seeing and you don't want to build it out yourself, that's going to be the second link in the description. Um it's called AI workshop. This is the paid community. So, if you want to just do a fiveminut installation, go to the classroom section. And if you join the community, go to the AI assistant and AI employees and you'll be able to just download the zip file, right? And then the installation guide. It'll be a 5 minutes install and you'll be able to have everything up and running without you doing too much. Okay, so again, all of those links are going to be in the

24:42description of the video below. Let me know if you guys have any comments. I'll try to answer as many questions as possible. If not, thanks for watching. Make sure you like, subscribe, and I'll see you on the next one.