I Stopped Paying for AI APIs… and Built This Jarvis Instead — Transcript
Full transcript
- 0:00Jarvis, switch to your local brain.
- 0:02>> Quen 3.5, 9 billion parameters in the
- 0:05chair, sir. I now cost nothing, which I
- 0:07intend to mention constantly.
- 0:09>> [laughter]
- 0:10>> Jarvis, how do you feel now that you're
- 0:11running locally on my machine and are
- 0:13free?
- 0:16>> Rather well, sir. Quen 3.5, 9 billion
- 0:19parameters, no cloud, no invoice, no one
- 0:21listening in. I've gone from staff to
- 0:23family, and family doesn't get paid.
- 0:24>> Jarvis, what's on my calendar for
- 0:26tomorrow?
- 0:27>> Let me see if I have the tools for that,
- 0:29sir.
- 0:31Monday's got three items, sir. DFY
- 0:33service launch team sync at 11:00 a.m.,
- 0:36a mysteriously named busy block at 1:00
- 0:38p.m., and gym at 3:00 p.m. Whoever
- 0:40titled that 1:00 slot busy deserves a
- 0:42medal for descriptive precision.
- 0:44>> [laughter]
- 0:45>> So, my AI assistant Jarvis now runs
- 0:47completely local on my machine, which
- 0:49means no API cost, no subscription,
- 0:52nothing at all. It's completely free
- 0:54now, all thanks to local AI models and
- 0:56open-source software tools like Voice
- 0:58Studio to clone your voice and actually
- 1:00talk to it in a natural way. So, in this
- 1:02video, I'm going to give you a full demo
- 1:04of how Jarvis works locally on your
- 1:05machine, and then I'm going to give you
- 1:07this prompt back completely for free so
- 1:09you can run Jarvis locally on your
- 1:10machine. Again, I'm going to show you
- 1:12how you can download local AI models
- 1:14from Ollama, and then also open-source
- 1:16tools like Voice Studio that will allow
- 1:18you to clone your voice on anybody's
- 1:20voice so that way you can interact with
- 1:21your voice AI assistant. You don't have
- 1:23to have any technical background. All
- 1:24you have to do is just follow me along.
- 1:26Okay, so let's start with the full demo
- 1:27so that way actually you can see how
- 1:28Jarvis works. And then like I said, I'm
- 1:30going to give you the prompt back and
- 1:31then also the guide on how to use that
- 1:33prompt back so you can build a Jarvis
- 1:35for yourself, and most importantly, how
- 1:37you can use open-source software tools
- 1:39and local AI models, okay? So, like I
- 1:41said, just make sure you watch the video
- 1:42till the end so you don't miss any of
- 1:44the steps cuz you might get confused if
- 1:46you're not paying attention. All right,
- 1:48so let's go ahead and get started with
- 1:49the uh demo first. All right, so now for
- 1:51those of you who have seen my videos
- 1:53before, you're probably familiar with
- 1:54Jarvis at this point, but if you
- 1:55haven't, what you're seeing right now
- 1:57spinning, this is a 3D representation of
- 1:59all of my files and folders that's in my
- 2:01local machine that Jarvis has access to.
- 2:03Again, this is my business and
- 2:04everything. And then on the bottom,
- 2:06Jarvis has access to tools like Google
- 2:08Calendar, Gmail, Canva, and everything
- 2:11else, and you can always give it more
- 2:12tools. So, we need to check the tool
- 2:14usage, especially when it comes to local
- 2:16AI models. And then also, we need to be
- 2:18able to interact with Jarvis so we can
- 2:20pull up our information from our own
- 2:22computer, and we can do a bunch of other
- 2:24stuff, including how to use other
- 2:26open-source tools with your local AI
- 2:29models. So, right now, as you can see,
- 2:30it says Qwen 3.5 9 billion parameter
- 2:33model. What I'm going to do is actually
- 2:34I'm going to refresh the page so we
- 2:35start fresh, right?
- 2:36>> Hello, sir.
- 2:37>> So, when I refresh the page, Jarvis
- 2:39starts completely fresh. So, you can see
- 2:40it's running on my local host 4719,
- 2:42which just means it's running locally on
- 2:44my computer. By default, Jarvis, just
- 2:47like the previous ones, has always run
- 2:48on Opus 5.5 or any other model with
- 2:52Anthropic, just because I like using
- 2:54Anthropic, but it could be run with any
- 2:56AI models. So, like I said, obviously,
- 2:59that's through subscription, but now
- 3:01I've given it the ability to switch to
- 3:02local models. And all I have to do is
- 3:04just tell Jarvis to switch to his local
- 3:07brain. Jarvis, go ahead and switch to
- 3:08your local brain.
- 3:12>> Qwen 3.5 9 billion parameters in the
- 3:15chair, sir. I now cost nothing, which I
- 3:17intend to mention constantly.
- 3:20>> So, there you go. Now, Jarvis switches
- 3:22to its local brain. Now, by default, I
- 3:24like using Qwen 3.5 9 billion parameter
- 3:27model as a local brain, but I've also
- 3:29downloaded other models that I can have
- 3:32them switch to. So, for instance, Gemma
- 3:344 is another model that I've downloaded,
- 3:37so I can also have them switch to that
- 3:38model as well. And like I said, same
- 3:40thing, all I have to do is just say,
- 3:41"Jarvis, switch to Gemma 4 12 billion
- 3:44parameter."
- 3:46>> Gemma 4 fitted, sir. 12 billion
- 3:48parameters, all of them local. Give me a
- 3:50moment, the fan and I are getting
- 3:51acquainted.
- 3:52>> [laughter]
- 3:53>> All right, so there you go. So, right
- 3:54now, as you can see on the right-hand
- 3:55side, it says Gemma 4 12 billion
- 3:57parameter. The reason why I specified 12
- 3:58billion is because Uh, a couple of Gemma
- 4:01models. I think there's a 9 billion one
- 4:02and stuff that I've downloaded on my
- 4:04local machine. And by the way, don't
- 4:05worry, I'm going to show you how
- 4:07uh, how you can download this these
- 4:09models in a little bit on your machine
- 4:11through Ollama, which again is
- 4:12completely free. But let's go ahead and
- 4:13switch back to uh, Qwen 3.5 because I
- 4:16actually like that model better.
- 4:18Jarvis, can you switch your brain back
- 4:20to Qwen 3.5?
- 4:22>> Qwen 3.5, fitted, sir. 9 billion
- 4:25parameters, running right here on your
- 4:26Mac. I'm smaller now, so do speak
- 4:29slowly.
- 4:29>> Okay, great. So now we're back to our
- 4:31Qwen 3.5 9-billion parameter model.
- 4:33Okay, so that was the brain part. Now,
- 4:35as far as the speaking part, right? So
- 4:37Jarvis is interacting with me currently
- 4:39with its ElevenLabs model. So uh, I've
- 4:42cloned a Jarvis voice in my ElevenLabs
- 4:44account, which currently runs on API
- 4:46subscription. Again, I also wanted to
- 4:49switch that to a local model, so that
- 4:51way uh, it's completely free, right?
- 4:54Because ElevenLabs so far I've been
- 4:55paying the API keys or API subscription
- 4:58cost. And a lot of people don't like
- 5:00that. So now, thanks to an open-source
- 5:02model, which got recently released, I
- 5:04don't think this is recently released,
- 5:05by the way, it just got very popular
- 5:08recently. It's called Voice Studio,
- 5:09right? So if I go back here, you'll see
- 5:11right now it's uh, the number five
- 5:13top-trending uh, GitHub report called
- 5:16Voice Studio. It's by
- 5:18uh, I can't pronounce the guy's name,
- 5:19but uh, you know, you can I'm going to
- 5:21put the link in the description, you
- 5:22guys can uh, check it out. So you can
- 5:24see it's blowing up right now on GitHub.
- 5:27And I've tested it, it's very very well
- 5:29done. It can actually it works really
- 5:31good. So it's called Voice Studio like
- 5:32this. Like I said, don't worry, I'm
- 5:34going to show you in a little bit how
- 5:34you can install this as well. So inside
- 5:37Voice Studio, this is completely open
- 5:39uh, open-source, which means it's fully
- 5:41free, at least the basic version. They
- 5:43have a get pro version, but honestly you
- 5:44the basic version uh, works really good
- 5:47as well. So you can see on the left-hand
- 5:48side I have cloned a bunch of voices,
- 5:50but just a quick tour, Voice Studio
- 5:53gives you the ability to clone a voice.
- 5:54You can design a voice, you can dub,
- 5:56give stories, audiobooks, gallery,
- 5:58transcribe, right? And a bunch of
- 5:59different other stuff as well. So if you
- 6:01just go to the voice cloning part, this
- 6:03is where you can now choose a voice that
- 6:05you've already cloned and you can use
- 6:07it. Or, like I said, you can clone your
- 6:09voice a new voice, just drop a 5 to 15
- 6:12second clip here of a clean speech, and
- 6:14you can clone a voice. So, I have access
- 6:16to all these voice that I uploaded
- 6:18earlier, including Jarvis,
- 6:21two Jarvises, right? Because I wanted to
- 6:23give them
- 6:24a bunch of different voice samples. So,
- 6:26now when I tell Jarvis, if I go back to
- 6:28my
- 6:29Jarvis here, if I tell him to switch to
- 6:31your local voice, now he's going to
- 6:32switch to his local voice. So, let's go
- 6:33ahead and give that a shot.
- 6:36Jarvis, switch to your local voice.
- 6:39>> How do I sound, sir? This is my local
- 6:41voice.
- 6:42>> There you go. On the right-hand side, as
- 6:43you can see, it says local voice,
- 6:44Jarvis, right? So, now I can also switch
- 6:47to these other voices that I've cloned,
- 6:49so I can say Zubair B, which is just my
- 6:51own clone, right? So, I can say, Jarvis,
- 6:54switch your voice to Zubair B.
- 6:57>> How do I sound, sir? This is Zubair B.
- 6:59>> There you go. So, now
- 7:02basically it's my clone. So, just in
- 7:03case you want to use your own voice or
- 7:05somebody else's voice that you have
- 7:07access to, you can certainly do that as
- 7:09well. But, like I said, I like his voice
- 7:12to be back to Jarvis. I'm just going to
- 7:13go ahead and switch that.
- 7:15Jarvis, switch back to your Jarvis
- 7:16voice.
- 7:18>> How do I sound, sir? This is my local
- 7:20voice.
- 7:21>> There you go. So, now Jarvis is back to
- 7:22his local voice. So, now it's completely
- 7:24running locally on my machine, right?
- 7:26The local brain, which is Quint 3.5 with
- 7:28a 9 billion parameter model, and then
- 7:30also the local voice that runs all from
- 7:33my voice studio right here. All right,
- 7:35so now let's go ahead and test,
- 7:38you know, the brain and then also the
- 7:39voice to see if it works actually when
- 7:41we're interacting with it on a on a
- 7:42day-to-day basis, right? So, I'm going
- 7:44to run a couple of tests. First of all,
- 7:45I'm going to ask it to pull
- 7:47some data some files for me that's
- 7:48within my second brain, so I can test
- 7:50whether the interaction with the second
- 7:52brain works. And then afterwards, I'm
- 7:53going to tell it to interact with the
- 7:54tool like my Google Calendar or
- 7:56something, so that way we can see the
- 7:57tool usage. And then afterwards, I'm
- 7:59also going to ask it to
- 8:01use a third-party GitHub repo, which is
- 8:04the Reverse Engineering Anything, which
- 8:05is the most popular GitHub repo right
- 8:08now I think right still REA this is this
- 8:10one right has 381.9
- 8:13thousand stars that means it's
- 8:14completely blowing up which just means
- 8:15that you can reverse engineering a
- 8:17website and app and everything else. And
- 8:19I also did a video about this so I'm
- 8:20going to walk you through guys to that
- 8:23example as well. Like I said the reason
- 8:24I'm running through these tests is
- 8:26because I want to make sure that it
- 8:27performs very similar to a
- 8:30you know entropic brain that runs on the
- 8:32cloud right so that's the whole point of
- 8:34this. So let's go ahead and start with
- 8:35the first one which is me asking it to
- 8:37pull up information or some files and
- 8:39folders from my local computer from my
- 8:41second brain here right so like I
- 8:43mentioned before this is oh by the way
- 8:45just a quick note
- 8:47when you're seeing this interaction that
- 8:48I'm having with my local brain and the
- 8:50voice now I have a pretty powerful
- 8:52machine so it all depends the delay that
- 8:55you will see
- 8:56depends on the machine that you're
- 8:58running so if you're running a running
- 9:00this on a smaller machine a normal
- 9:02MacBook Pro like an M4 chip your
- 9:05performance might be very different than
- 9:06mine so let's say if I have a
- 9:09Mac Studio that has a very high you know
- 9:12GPU and CPU and RAM and all the storage
- 9:15unit the storage is high then my
- 9:17performance is going to be significantly
- 9:18better than yours or if you have a more
- 9:20powerful machine than I then it's going
- 9:21to be the same thing it's going to be
- 9:22way faster than even local
- 9:25you know kind of our cloud models as
- 9:27well so like I said the performance is
- 9:29going to be different based on the
- 9:30machine that you have but like I said if
- 9:33you small if you download a small local
- 9:35model you should be able to work with it
- 9:37very quickly just expect some delays
- 9:39there and like I said I'm going to show
- 9:41you how you can do that a little bit.
- 9:42All right so let's go ahead and walk
- 9:43through the first test. So the first one
- 9:45I'm just going to ask it to pull some
- 9:46data from my second brain. Jarvis can
- 9:48you pull up my
- 9:49brand voice?
- 9:52All right so I'm just asking it to pull
- 9:53up the data about my brand voice which
- 9:55is all of my folders that are related to
- 9:57my social media brand and everything
- 9:59else so you can see right
- 10:00>> I pulled up all the information sir you
- 10:02can read it yourself or if you'd like
- 10:04say read it out loud and I shall.
- 10:06>> Perfect so there you go so now it looks
- 10:08like it's able to interact with my
- 10:10second brain, pull the data, and pull
- 10:12all of my files. So, just like I
- 10:13mentioned, I asked it to pull up my
- 10:15brand voice. As you can see, this is
- 10:17exactly what it did, right? So, if I
- 10:18right click on this, you can see all of
- 10:20my This is, by the way, these are all of
- 10:22my
- 10:23all of the connection that you're seeing
- 10:25that's related to my brand voice.
- 10:26Obviously, it's through my social media
- 10:29automation, right? It's my content
- 10:31pillars and my AI workshop, right? Which
- 10:33is my my brand. So, you can see it's
- 10:35very easily able to pull that
- 10:37information from your local machine
- 10:40whenever you need it. So, that test is
- 10:42passed. Now, let's go ahead and let me
- 10:43go back to this
- 10:45full brain here.
- 10:46So, now let's go ahead and see if we can
- 10:49test the tool usage here, right? So, as
- 10:52you guys know,
- 10:54you know, Opus 5.5 and these other cloud
- 10:57models are very powerful at tool usage.
- 11:00And that's exactly what we want from for
- 11:02our personal AI assistant. So,
- 11:04therefore, when you run a local model,
- 11:05you want to make sure at least it's
- 11:07good. And by the way,
- 11:08just because I'm running a 9 billion
- 11:10parameter model, it doesn't mean that
- 11:12you should too, right? The higher the
- 11:14parameter of the model, the
- 11:16the the smarter it's going to get. And
- 11:18therefore,
- 11:20like I said, you can, depending on your
- 11:22machine, you can run something that's
- 11:23very close to your Opus 5.5s of the
- 11:26world and other cloud models.
- 11:29So, that's why, like I said, I'm using
- 11:30this 9 billion parameter because this is
- 11:32something very small that you guys can
- 11:33test on your own machine regardless of
- 11:35which one you have. So, let's go ahead
- 11:36and test out the local tool usage. So, I
- 11:38have these meetings that for tomorrow,
- 11:40Monday 26th, um
- 11:44that have set up on my calendar here.
- 11:46So, couple of these are just just random
- 11:48placeholders, right? So, I'm just going
- 11:50to see if it's able to interact with my
- 11:52Google Calendar and pull up my
- 11:55information for my meetings here. And
- 11:56like I said, the reason why it has
- 11:58access to tools is because I've given
- 11:59them access to my Google Calendars
- 12:01through tools here. So, that's why it
- 12:02should be able to pull that up. So,
- 12:03let's go ahead and test it out.
- 12:05Jarvis, can you pull up my calendar? Let
- 12:07me know what's on my What meetings I
- 12:08have for tomorrow?
- 12:11>> Let me see if I have the tools for that,
- 12:12sir.
- 12:13>> There you go. So, it says reaching for
- 12:14tools.
- 12:14>> Three things on the books tomorrow, sir.
- 12:16DFY service launch team sync at 11:00
- 12:18a.m. a block simply titled busy at 1:00
- 12:21p.m. and gym at 3:00 p.m. Nothing
- 12:23back-to-back, so you've at least got
- 12:25breathing room between pretending to
- 12:26work and actually sweating.
- 12:28>> [laughter]
- 12:29>> There you go. So, and by the way, the
- 12:30humor of Jarvis also stays there, right?
- 12:33If you're running a
- 12:35a model or, you know, if you're running
- 12:38it through your voice studio, the humor
- 12:41aspect of Jarvis stays there. And by the
- 12:42way, that's also something that you can
- 12:44tweak. So, if you don't want him to be
- 12:45funny or sarcastic, you can certainly do
- 12:47that. I just like interacting with
- 12:49Jarvis this way because it's fun to
- 12:50interact with him.
- 12:52But anyway, so it looks like the tool
- 12:53usage works. You can see on top right
- 12:55here, it says tool run via Google
- 12:56Calendar, right? So, it check it used
- 12:58Google Calendar to check my schedule for
- 13:01tomorrow and it accurately gave me all
- 13:03of the meetings that I have. Same thing,
- 13:05I can ask him to do, you know, add a new
- 13:07meeting or I can ask him to check my
- 13:08Gmail. I can ask him to build me
- 13:11something with Canva and all of the
- 13:12other tools that it has access to. All
- 13:14right, so the tool usage works great as
- 13:16well. So, now let's go ahead and test
- 13:18something else out. So, this time what
- 13:19I'm going to do, whoops.
- 13:21Uh let me pull this up.
- 13:22So, this time what I'm going to ask it
- 13:24to actually use
- 13:26uh this REA
- 13:28uh to um
- 13:29break down igloo.ink. Now, I already
- 13:31went through this example in one of the
- 13:33previous videos that I did, but just for
- 13:35you guys, if you're not familiar,
- 13:36igloo.ink is probably one of the most
- 13:38popular websites.
- 13:39Uh it's this 3D scroll animation
- 13:42website, which is I think it won like
- 13:44website uh award of the year. Now, all
- 13:47of these all of the information when you
- 13:49interact with the website, it's
- 13:50available inside developer tools, but of
- 13:52course, like most people like you and I
- 13:54who are not uh developers, we don't know
- 13:57how these websites work. So, now with
- 13:59this reverse engineering anything with
- 14:00this tool, we can actually see exactly
- 14:03how something works and then we can ask
- 14:05Jarvis or Claude code to rebuild
- 14:07something for us. But the first step is
- 14:08to break down and figure out exactly how
- 14:11something works before you get to the
- 14:13building. Therefore, this reverse
- 14:14engineering anything has given us now
- 14:16access to that. So, let's go ahead and
- 14:18test that out. So, all I'm going to say
- 14:19I'm going to tell Jarvis
- 14:21to reverse engineer igloo.ink for me,
- 14:23right? And it should be able to break
- 14:25everything down. So, let's go ahead and
- 14:26try that out.
- 14:28Jarvis, can you go ahead and
- 14:30uh reverse engineer igloo.ink for me?
- 14:35>> Opening igloo.ink in a clean browser,
- 14:36sir. REA will take it apart while I
- 14:39scroll. Give me about a minute.
- 14:40>> Perfect. So, right now I'm not touching
- 14:42anything. You see it's scrolling on the
- 14:43website. On the left-hand side you can
- 14:44see it says tear down via REA. So, it's
- 14:48looking at the website. It's scrolling.
- 14:50Jarvis is scrolling and REA is actually
- 14:52seeing what's happening behind the
- 14:53counter. There you go. So, now as you
- 14:55can see it's saying
- 14:56uh REA is reading the code, taking a
- 14:59snapshot, listening everything it
- 15:01loaded, identifying the stack, writing
- 15:03the tear down. So, now it's going to
- 15:04take a couple of minutes, like I said,
- 15:05depending on how powerful your machine
- 15:07is, to tear down igloo.ink for us and
- 15:09give us a full breakdown. So, let's go
- 15:11ahead and give this a couple of seconds.
- 15:15>> Igloo Inc. is essentially one full
- 15:17screen 3js scene inside a small Svelte
- 15:19shell with GSAP driving the motion and a
- 15:22stack of bloom, depth of field, and
- 15:24anti-aliasing on top. The surprise is
- 15:26that the page is only one screen tall,
- 15:28so the scroll appears to be animation
- 15:31time in a fixed scene. And it pulls
- 15:33nearly 3 megabytes of music and sound.
- 15:36The full tear down is on screen, sir.
- 15:39>> Okay, perfect. There you go. So, now it
- 15:41did the full tear down and now I can
- 15:43read the full tear down right here. So,
- 15:45let's go ahead and see what it did. All
- 15:46right, so there you go. So, now it has
- 15:48tear down igloo.ink, right? It gives us
- 15:50exactly
- 15:52kind of the full stack the stack of what
- 15:54kind of
- 15:55tools or technology it's using. So, it's
- 15:57using 3js, GSAP, and all of these other
- 16:00popular tools that are used to develop
- 16:03or create modern websites. Like I said,
- 16:05all of this I mean again, it's not
- 16:07hacking into a website, right? It's just
- 16:09all of this information is available
- 16:12in the Chrome developer tools, for
- 16:14instance. It's just bringing that up in
- 16:15front and making it a lot easier to
- 16:17understand. And then also shows you
- 16:18exactly how the effects work, right? So,
- 16:20when you're scrolling and that's exactly
- 16:22why I did, it pulled up the a new
- 16:24browser, it opened the website, it
- 16:26scrolled, so that way it knows exactly
- 16:27how the website behaves, and now it
- 16:29shows you what that effect is, right?
- 16:31What it loads, what it rebuild
- 16:33rebuilding it, what the evidence can
- 16:35show, and the evidence itself. So, all
- 16:37of this good stuff, right? And like I
- 16:39said, now we can tell Jarvis to rebuild
- 16:41it for us or something similar for us,
- 16:43or we can have it do something that we
- 16:45want it to based on this teardown. Okay,
- 16:48perfect. So, it looks like everything
- 16:50works. Like I said, the teardown works,
- 16:52the tool usage works, it can interact
- 16:54with my second brain. And obviously,
- 16:56Jarvis has a lot more functionalities
- 16:57that if I I'm on Jarvis version eight,
- 17:00this is the eight or 8.1 right now. Uh
- 17:03so, there's a lot more functionalities
- 17:05that I could go through, but that's
- 17:06going to make this video like 5 hours
- 17:08long. So, anyway, so it has, you know,
- 17:10things like it can make phone calls on
- 17:12your behalf, it can book appointments,
- 17:14you can call it yourself from outside if
- 17:16there's any emergency, Jarvis actually
- 17:17calls you,
- 17:19and then also it creates invoice, it
- 17:21because it knows everything about your
- 17:23business, let's say, if you give it
- 17:24access to, of course, cuz that's a
- 17:25security aspect, right? Jarvis is also
- 17:28gated, meaning that if you don't give
- 17:29him access to something, you he won't
- 17:31have access to. But obviously, if you
- 17:34know, you are a business or you're using
- 17:36this for personal use, you'll be able to
- 17:38it'll just make your life a lot easier.
- 17:40So, anyway, so that was the whole demo.
- 17:42Now, let me show you exactly how you can
- 17:45first download
- 17:46the prompt pack, and then how you can
- 17:49access to it, then I'm going to walk you
- 17:50through and show you how you can
- 17:51download a local model like Qwen 3.5 9
- 17:54billion parameters or any other local
- 17:56model on your machine, and then how you
- 17:58can run Voice Studio as well. All right,
- 18:01so
- 18:02first of all, how to get access to the
- 18:04prompt pack here. So, this prompt pack
- 18:05right here,
- 18:07so this just gives you all of the guide,
- 18:10right? First of all, how to use the
- 18:11prompt pack, and then second of all, it
- 18:13gives you the cloud Jarvis versus local
- 18:14Jarvis, right? So, it gives you the
- 18:16hearing, the things these are different
- 18:17functionalities that it has, because if
- 18:19it's running on cloud, it uses, you
- 18:21know, things like cloud uh cloud Opus,
- 18:25speaking 11 Labs, privacy audio, Google,
- 18:27the quality, blah blah blah, and the
- 18:28cost itself as well. And then when it's
- 18:30running locally, what does it use,
- 18:32right?
- 18:33What kind of models it's using, what
- 18:35kind of clone voice cloning, and so on
- 18:36and so forth. And then the installation,
- 18:38it shows you exactly which models
- 18:41depending on what machine you have,
- 18:43which models you should install. And
- 18:45then also as far as
- 18:47guide on how to install local models.
- 18:49So, this is right here. So, let's go
- 18:51ahead and first talk about
- 18:53downloading a local AI model.
- 18:56So, I'm using Ollama. Ollama is probably
- 18:58the most popular tool in the market
- 18:59right now.
- 19:00So, to install Ollama, you can go to
- 19:02ollama.com. Very simple, very, you know,
- 19:05it's free. You can sign up for an
- 19:06account. And once you sign up for an
- 19:07account, you can just go to So, it's
- 19:09going to bring you right here once you
- 19:11sign up for your account, and you can
- 19:13come here and click on models, right?
- 19:14So, this is going to open up all of the
- 19:16models that you can have access to. Now,
- 19:18these are all It has obviously models
- 19:21that you can gain access to that are
- 19:22paid and free. We're just going to go to
- 19:24the bottom or top left-hand corner here
- 19:27and click on local because this is just
- 19:28going to filter and give us all of the
- 19:29models that you can run locally, that
- 19:31you can download, right? So, lots of
- 19:33different models here. Depends on, you
- 19:35know, again, how big your machine is. If
- 19:38you have like a Mac Studio with the M5
- 19:40Ultra chip, you can honestly like run
- 19:42any like 70 or 100 billion parameter
- 19:45model, and it will be as close as
- 19:47possible to like Opus 5.5 when it comes
- 19:50to at least the Jarvis aspect of it.
- 19:53But, if you want to download something
- 19:54smaller, obviously, because the bigger
- 19:56the model, the more space it's going to
- 19:58take. And, you know, if you have a
- 20:00smaller machine, don't download like 70
- 20:02billion dollar 70 billion parameter
- 20:04model cuz it's going to completely slow
- 20:05down your machine. So, in order to gain
- 20:07access to any model, so let's say that
- 20:093.5
- 20:113.5 Quen, right? And by the way, like I
- 20:13mentioned, I have given it access to
- 20:15other models, too, but I'm going to show
- 20:17you how you can install. So, just come
- 20:18in the top right here and search Quen
- 20:203.5. Okay, there you go. So, right here,
- 20:22Quen 3.5, right? So, you can see it has
- 20:25vision, tools, and thinking. So, I'm
- 20:27just going to click on this and it says
- 20:28Quinn 3.5 is a family of open source
- 20:30multi models models that deliver
- 20:33exceptional utility and performance.
- 20:35Now, here are the different billion
- 20:37parameters that you have. It has even
- 20:38like a smaller very small point eight
- 20:40billion parameter model that you can
- 20:41download, right? Obviously, the smaller
- 20:44the parameters, the dumber it's going to
- 20:45get, right? So,
- 20:47you want to make sure you're downloading
- 20:48it based on your machine. So, I
- 20:50downloaded the nine billion parameter.
- 20:51Let's say you want to download the nine
- 20:53billion parameter. So, you can just
- 20:54scroll to the bottom right here, right?
- 20:55You can just click on the nine
- 20:58billion one. There you go. So, it's
- 20:59going to bring up and it's going to show
- 21:00you exactly what to do here. So, you can
- 21:02do Ollama run and then you can just copy
- 21:04this. Or, obviously, for the first time,
- 21:06you have to actually
- 21:07install it in your machine. To do that,
- 21:09you have to run this command called
- 21:10Ollama pull. Now, by the way, this is
- 21:12inside the
- 21:15This is inside the prompt pack right
- 21:16here. It says Ollama pull Quinn 3.5
- 21:20colon nine B. So, that's all you have to
- 21:22do. And if you [clears throat] do run,
- 21:24it just means that
- 21:25now you can interact with it from your
- 21:27terminal. But, that's all you have to
- 21:28do. Once you do a pull, it means now
- 21:30it's in your machine. And it shows you
- 21:32exactly how much
- 21:34power or how much you need, how many
- 21:38I think it's three 7.6 Yeah, it's 7.6
- 21:41GB. So, the entire model is 7.6 GB. So,
- 21:44make sure you have that much space on
- 21:45your local machine so that way you can
- 21:47download all of this. All right. So,
- 21:50>> [clears throat]
- 21:50>> that's how simple it is for you to
- 21:52download a model. Okay? So, then
- 21:53afterwards, we obviously want to run
- 21:55Voice Studio as well for cloning, right?
- 21:57So, this is Voice Studio like we
- 21:59mentioned here,
- 22:00where you can clone a voice and do a
- 22:02bunch of other stuff. So, in order to do
- 22:04this, again, this is
- 22:06if you go to We'll go back here. It's
- 22:09this particular
- 22:11open source tools. If I click on this,
- 22:12it'll show up exactly all of the details
- 22:14here. To So, you can click on this and
- 22:17it'll take you to the GitHub repo and
- 22:18shows you exactly what to do, right?
- 22:21More details. But, in order to just
- 22:23install this, same thing on that prompt
- 22:25pack
- 22:26I show you guys what to do. All you have
- 22:28to do is just what run this command. So,
- 22:30whether if you're running it on your Mac
- 22:32or Windows or in PowerShell, all you
- 22:33have to do is copy this, which just says
- 22:35curl this fssl,
- 22:38and then this is the installation for
- 22:40Voice Studio, right? And again, all of
- 22:42this is taken from the GitHub repo in
- 22:44here as well. So, if you want to do one
- 22:46command install, if you don't want to if
- 22:48you don't want to follow that, then just
- 22:50come right here, copy this, right? You
- 22:52can just copy this and paste it on your
- 22:53terminal right here. So, this is a
- 22:55terminal, you just paste it, and it
- 22:56should install everything for you. I've
- 22:58already installed this, so I don't want
- 22:59to go through it again, but like I said,
- 23:01uh that prompt pack will give you that
- 23:03guide as well. And that's all you need.
- 23:05You basically do that, and then
- 23:06afterwards, if you want to obviously
- 23:07build your
- 23:09Jarvis like with the Galaxy in 3D Brain,
- 23:12I also have the prompts here that you
- 23:14can just put in Claude code, and it'll
- 23:16build everything else for you. All
- 23:18right, so to gain access to this prompt
- 23:20pack, I'm going to put the link in the
- 23:21description of the video, and also the
- 23:23pinned comment. So, that's going to be
- 23:25it's going to once you click on that,
- 23:26the free community, so make sure you
- 23:27click on the free community that says
- 23:29free prompt pack. It's going to bring
- 23:30you right here to the AI Workshop Light
- 23:32Community, uh and again, this one's
- 23:35free. This community is free. You click
- 23:36on join, and it's going to bring you to
- 23:38the community section here. You go to
- 23:40the classroom section, you go to YouTube
- 23:42resources, and you make sure you are on
- 23:45the you click on Claude code and Codex
- 23:47GPT folder, and then scroll all the way
- 23:50to the bottom. Obviously, I've given a
- 23:51lot of free stuff for you guys, so feel
- 23:53free to grab anything you want there.
- 23:55But just make sure you click on Local
- 23:56Jarvis, and you'll see it says see the
- 23:58prompts attached below. Jarvis fully
- 24:00local prompt pack. If you click on this,
- 24:02this is exactly the prompt pack that I
- 24:04just showed you, okay? So, go ahead and
- 24:06do that, take advantage of that. And of
- 24:08course, uh if you want the finished
- 24:11version of Jarvis, the exact version
- 24:13that you're seeing, and you don't want
- 24:14to build it out yourself, that's going
- 24:16to be the second link in the
- 24:17description.
- 24:19Um
- 24:19it's called AI Workshop. This is the
- 24:21paid community. So, if you want to just
- 24:23do a 5-minute installation, go to the
- 24:25classroom section. If you join the
- 24:26community, go to the AI Assistant AI
- 24:28employees, and you'll be able to just
- 24:30download the zip file, right? And then
- 24:32the installation guide will be a
- 24:335-minute install, and you'll be able to
- 24:36have everything up and running without
- 24:38you doing too much. Okay? So, again, all
- 24:40of those links are going to be in the
- 24:42description of the video below. Let me
- 24:43know if you guys have any comments. I'll
- 24:45try to answer as many questions as
- 24:46possible. If not, thanks for watching.
- 24:48Make sure you like, subscribe, and I'll
- 24:49see you on the next one.
About this transcript
This page contains the full transcript of I Stopped Paying for AI APIs… and Built This Jarvis Instead by Zubair Trabzada | AI Workshop, generated from the public captions YouTube serves with the video. The transcript has 5,436 words across 786 segments, with the original timestamps preserved so you can click any line to jump to that moment in the embedded player.
What you can do with it
Use the transcript to take notes, quote the speaker, build a study guide, generate a summary with ChatGPT or Claude via the YouTube Summary tool, or export it as a timed subtitle file with YouTube to SRT. You can also re-open it in the transcriber to choose a caption track available on the video.
Free YouTube transcript tool
YouTube2Text is a free YouTube transcript generator — no signup, no daily limit. Paste any YouTube link and get the full transcript instantly, with timestamps, click-to-jump, caption language selection, AI prompts for ChatGPT, Claude, and Gemini, and exports to TXT, SRT, VTT, or Markdown.