The first step to forging a new path is questioning the one you're taught to discover it was never the only way. Anthropic continues to make history with its artificial intelligence products. Not only did they manage to offer cloud models consistently ranked among the best, but they also improved the application and features at such a pace that they're now among the top platforms.
However, what truly sets them apart isn't their chat mode, but rather their Corde functionality. These advanced options are reserved exclusively for paid plans. But is there a way to unlock them?
The answer is yes. To find out, we'll explore five use cases, including adding any model to the cloud. With a single click, we'll connect various AI providers, including official ones, all without losing cloud functionality, among other use cases.
Furthermore, you can apply each method we'll cover today, completely free of charge. Sounds good, right? So let's see how we're going to get started with a new version that completely changes the Olama interface.
This version also adds different buttons so that with a single click we can change the cloud code and unlock it for any model we want, whether it's a model that runs in the cloud or a model that we can run locally. This means we can now use it not only for free, but also with unlimited and private access to explore how it works. Let's jump to Olama, this platform which has also completely changed its website with this new version, now having a very clear focus on running open-source models to allow for greater freedom.
From here, keep in mind that to install it on our device we have to click on this download button. And note that for both macOS and Windows devices, we'll have a simple installer. And for Linux, we'll also have this option from the terminal.
We simply run the command and we'll have the application. Once it's installed, we'll see an interface like this one. But before we look at this new feature and how to unlock the cloud, the first thing we need to do is install the app, even if we don't have a paid license.
To do this, simply go to the link I'll leave below in the description, along with everything else we'll cover in this video. From there, you'll see that you can download it on virtually all devices. Once installed, you'll see an interface like this one, but unless you have the paid version, as I do, you won't find the Cowork feature, which is what actually lets you automate tasks, nor the cloud code feature shown here.
But with the method we'll see, we'll be able to unlock all those features with any other subscription model. To do this, if we go back to the Dorama app, you'll notice that they've now brought the apps feature to the forefront. From here, we can differentiate the Cloud Desktop app, because if we simply click this button, it will ask if we want to restart the Cloud Desktop app to unlock it and connect it with models other than the Cloud ones.
If we click it, it will restart immediately. And although everything seems exactly the same, with the chat mode, coworking mode, and cloud code mode, just like in the paid version, notice that in the bottom right corner, I don't have a pro account connected, but rather an account linked to Olama that unlocks all these features. This is because.
. . If I go back to the home screen, you'll see that from here, in the models section, we can now connect it with any model we want.
And this is key, since Olama really does work with all open-source models, whether on-premises or in the cloud. But keep in mind that in future cases we'll also see how we could connect it with proprietary models, even with our own. Antropic.
The interesting thing about what we just saw with Olama is the possibility of making Clot Desktop independent of a single model and company. And that philosophy seems key to me because, depending on the task, one model might be better than another. That's why I want to introduce you to Mamuth AI, a platform that unifies many of the top AI models from leading companies in one place.
From a single interface, you can access models like GPT, Clot, Gemini, Grock, Tipic, Lama, and many more, and choose which one to use at any given time. If an answer doesn't convince you, you can even resubmit the same request, but with a completely different model. Furthermore, we'll have models in virtually any format.
You can generate images, create videos, conduct AI research, work with reasoning models, or use lighter options when you need faster answers. But here, taking this idea even further, Mamuth also has its own API platform. This is compatible with OpenI and allows you to connect all the unified models in its ecosystem to any of your own tools so you can continue your own workflows.
This gives you a much more flexible workflow without depending on a single provider. And all this with a single, no-commitment subscription starting at €10 per month. And thanks to the sponsors of this video who left me a link, which I'll leave in the description below, so you can try it for yourself.
Now that we know this, let's look at the next use case. [music] In this second section, let's talk about open-source models, since they have truly advanced in artificial intelligence to such a level that they are even surpassing closed, paid models. In fact, just this week Olama shared this tweet with this graph showing a practically vertical surge due to the expansion of open-source model usage .
And this isn't really due so much to the Dama platform itself, but rather to the fact that the models being released as open source are truly excellent. Just a few weeks ago, Alibaba launched Quen 3. 8 Max, a model that even ranked among the best AI models.
Chemistry 3 also excelled in the frontend category, even surpassing Fable 5. Prior to this, GLM 5. 3 was released, achieving the number one spot in almost every ranking.
These three models share a common characteristic : they are all open source. But this doesn't stop there. Just a week ago, these updates, which we've discussed on this channel, were released, and this same week, just two or three days ago, these open-source models continued to receive updates.
GLM 5. 3, the Flash version, was recently released. It went viral under the name OX Alfa, and everyone speculated it might be Yine 4 Pro, but Zi finally revealed it was their own model.
We also found new versions that were released just a few days ago, such as Quen 3. 8 Flash version, which not only has a very good position but also ridiculously low costs, allowing one million inputs for only about 15 cents, and the token output doesn't even reach 50 cents per million tokens. This is key, because beyond being able to use very cheap and efficient models, it also helps other providers be very aggressive with their offers and provide these powerful models for free.
And we'll be able to connect this to our unlocked cloud so we don't have to worry about limits. To see this, let's go back to where we left off with the unlocked cloud from Olama and start by talking about all the free cloud models we can use from this provider. By default, we'll only look at one model.
In this case, we have the YMA 4 with 31 billion parameters, a truly excellent model, but keep in mind that we could also interconnect more models from the same provider. To do this, let's first test it out, starting with the chat by asking, "Hello, how are you? " We'll press send, and from here, as we can see, it responds.
. . This is done almost instantly with all the cloud animations, but using an open-source model.
We'll be able to use this same approach for other cloud functionalities. But before we get into that, let's go back to the Dolama platform, as it's crucial that we click on the settings section. Beyond configuring certain parameters here, the most important thing is that we have a dedicated section for the cloud desktop application.
From here, we can replace the models we had in Cloud with our own choices. Notice that by default, in Sonet 5, we have the Yma 4 model, which is precisely the one we've been using, but I can also specify powerful models that I want to replace with F, such as the Minimx M3 model. In OPU 5, we could replace it with another model, such as the Deepsic V4 Pro.
And ultimately, I would have configured the rest of the models this way. Notice that when you click on this selector, some options will appear in gray, which means it requires a plan. However, we'll later see how we can connect it to any other provider so we don't encounter these limitations or get tied to any one platform, since we won't be using Clot, nor will we be using Olama.
Even so, we still have a lot of options, both in the cloud and some local ones, which we'll look at later. From here, once we've selected the models we want to use directly in Cloud Desktop, we have to click on "restart" for these changes to take effect. So I'm going to click it, and once that's done, OAM will ask us to restart it again.
So I'll confirm, and a little later, look what we have here. We'll find this selector with all the models we've configured, including Minx M3, for which we can also change the effort from the lowest to the maximum. We're going to have the Deepsic V4 model, the Yma 4 model with 31 billion parameters, Deepsic Flash, and Nvidia's Nemotron.
And with this, we're no longer just talking about unlocking the cloud with small, locally installed open-source models; we 'll actually be able to connect very powerful models, but with the entire ecosystem and advanced features that Tropy provides. In fact, if I click on any other model now, like Deepsic V4, I'll be able to use it for features like Coworking. From here, I'll also have options such as how I want approvals to work.
In this case, I'll set it to automatic. I can also select a folder. In this case, I'll select the downloads folder and click "open," and now I'll have access to it.
Now I can ask it anything, like, " List all the items in my folder. " When I click send, it starts processing, and as we can see, it tells us it's going to list everything in the download folder. Notice that in practically a matter of seconds, it listed everything we have, both the individual files and the folders I've created.
And indeed, it's exactly what I have in this folder. One point I want to share is that I just got an error when I used the Deepsek model, as it said it required a paid license from Lama. So keep in mind that they're currently in a transition period with this new update, so you might encounter this.
However, you'll be able to change this to another model. For example, I'm going to remove Deepsek Pro from here, and that way I'll be able to continue using it. Another section I want to show you is that if we expand these options, notice that we'll keep all the options, such as cloud projects , the entire artifacts section, the entire scheduled tasks section, and the entire customization section that we can enable from here by creating skills and connectors, but without We're limited by the small constraints we have in the cloud.
To finish this section, notice that if I go back to the code section, we can also use it from here, and it will show us the history we've been working with the official templates. For example, this little dot you see here is because it was used with the Opus templates, but we can actually modify it here with any other template that appears here, and we can ask it to do anything we want. For example, if I say here, "Create a simple Notion-style application for task management where I can see the status of tasks with a Canban," when I click "send," it will ask for access to a folder, so I'll create it, open it, and by clicking "send," this template will start processing with Cloud Code.
Furthermore, notice that we'll again have all the options we have in Cloud Code, including manual mode, accepting edits, plan mode, and even auto mode so it can manage permissions for us. This is a feature that was recently enabled in Cloud Code, and we'll be able to use it here from our models. And as you can see, we'll also have all those modal windows characteristic of Clot.
Here, for example, it's asking me three questions about the application I'm going to create. So here I'm going to tell it to do it in HTML. I'm also going to select here that I want the Canban option with tags to keep it simple.
And here I'm going to tell it to start with local storage. With that, it would continue working, just as Cloud Code could have done, asking us these intermediate questions to give us the result we're looking for. A little later, on the first try, it would tell me that I've already created this mini-application here with the Cam style and some options.
Keep in mind that alongside this visualization we can do directly in the cloud, we'll have all the same functionalities, such as the ability to annotate changes simply by surrounding them, or even the ability to export everything. Here, for example, we could even open the file and find all the code and programming I've done in this application. In this third section, we'll talk about how we can use all the technology of Unrestricted and Private Cloud.
For this, we'll have to use local models, and here's some good news: we can now say goodbye to downloading and installing a model that slows down our devices and leaves us unsure if it will actually work before downloading it. A platform called Can Run AI has just been launched that simplifies all of this . This platform is completely free, and you won't even need to register.
I'll also include the description below, and as soon as you log in, it will scan your device. In my case, I have an Apple M3 Pro processor, which is why it selected it this way. It also showed me my V-rank, which in this case is 18, and some other aspects like speeds, cores, and so on.
With all this information, it can tell us which local models would work perfectly on our computer before we download them. From here, I can see some top choices, such as the GPT OS with 20 billion parameters, which would work very well, as well as other models, including those for images and other formats. Here, for example, we can find models like Yema or Mistral .
But if you want to see all this in more detail, I recommend clicking on the models section, where you can see all the models that will run smoothly on your device. In this case, out of about 104 models, about 41 would fit well on my device. And from here we'll also be able to see a scoring of all these models There are models available, such as video cards like the Junyuan, and models from Quen, like this one here with 9 billion parameters, and other models as well.
Notice that some of them appear in red. This means that these wouldn't run well on my computer. So now I can say goodbye to downloading a model and crossing my fingers to see if it actually works or slows down my computer, since I can check from here if it will run well with my device.
And if it doesn't, I won't even have to download it. Also, note that if it doesn't recognize your device properly, you can go to the devices section and select your graphics card from there . If you have an Apple Mac, you'll be able to see all the Apple Silicon devices, but you'll also find Nvidia graphics cards, and further down, you'll find all the cards from AMD, Intel, ARM, and other companies.
So now let's see how we can easily download models locally. To do this, we'll go back to Olama and, on its official website, look for a section called "Models. " By doing this, we'll be able to see all the latest popular models, including both cloud-based models and those that can be installed locally.
If you want to find a specific one, such as the latest Quen 3. 8, it's as easy as clicking here, and from there you'll find all the available versions. For now, we only have the 27 billion parameter version, which I think is a bit too large for my computer, so I'm not going to download it.
But keep in mind that if you have a Mac, it's recommended that you download the MLX version, as it's optimized for Apple Silicon processors. Otherwise, download the non-MLX version. To see a small example of this, let's look at a Lama model and install, for example, this latest 8 billion parameter version.
To do this, I'll click on "copy the identifier of this model," and then, opening the terminal, if I type "olama full" and paste the copied name here, and press Enter, the download will begin, as you can see here. The download speed will depend a lot on your device, although in my case, these 5 GB will download in about 3 minutes. After those three minutes, it will tell me here that it has downloaded successfully.
And now, if we go back to the Olama application we installed, when we click, for example, on the chat section, we'll be able to see the model we installed. For example, in this case, I would have it here, the Llama 3. 1 model.
And if I click on it and then give it something, like asking it to tell me an interesting fact about the AI when I click send, this model will respond, and at the exact speed you're seeing right now. Here, in fact, it gave us this interesting fact, which you can pause in the video if you want to see it, but ultimately, the interesting thing here isn't just that we can use this model privately and in a limited way within this application, but that we can actually do it with any application we have directly in this applications section. Right now, we're talking about the cloud desktop application, but we can also do it with all of them, like Codex, Hermes, and others.
So now, if I go back to the settings section for models, I'll be able to modify all of them with models I have installed locally. To do this, I simply have to click here, and anything that doesn't appear as "cloud" will be a model we have on our device. For example, for a more powerful model, I could use this Yma model with 12 billion parameters that I had installed before.
And ultimately, I will have replaced all these models that I already had installed locally. Some models from Google's Yemma, others from Lama, and others from Alibaba like Quen. And now, if I click on "restart cloud" and confirm again, we'll see the cloud interface again, including the Code, Coworking, and Chat interfaces.
However, in the Models section, we'll find all the models we have installed locally. So now, with the model we just downloaded selected , if we ask it anything from here, like "the easiest way to make money with Igenativa," clicking "send" will start processing it. And from here, as you can see, it's already writing this entire response, at this pace you're seeing right now.
In fact, it's only taking these few seconds. Ultimately, it's showing us all these ways we can make money with artificial intelligence, culminating in this response directly in the Cloud Desktop tool. The truth is, many changes are appearing daily with generative intelligence, and this can be overwhelming unless we have a very solid and strategic foundation in its use.
Therefore , keep in mind that I have my intensive artificial intelligence course program, where you can learn with me for two months via video call, not live, so we can have a truly personal and close human interaction. You'll be in a class with a maximum of 25 people to ensure you can closely monitor your learning progress on this generative intelligence journey. Take a look at the course, which I'll link below in the description, as we're about to start, and I hope to see you in class.
In this fourth module, we're going to look at something very interesting that I discovered myself after seeing Lama's ability to unlock the Clop Testo tool. As you can see, if we want to use cloud-based models, Lama does have paid plans, which are quite good, but we might want to get more out of the free version. So, keep in mind that we'll also be able to interconnect it with any provider, and for that, I've found the following trick: if we go back to the cloud desktop application we just unlocked with Olama, and keep it unlocked, if we go to the developer section (which on Mac would be right here at the top, and on Windows, you'll also be able to access it), and click on the third-party inference configuration option, we'll find these settings here that actually open a door for us to connect with any other provider.
Notice that what Olama is really doing here is simply using their static API key with a base URL to their gateway, which in this case would be this one here, with an API key that, by the way, is nothing more than the name of Olama. But keep in mind that if we modify these two variables with any other provider's API key, we'll also be able to interconnect their models here. To see this, first, welcome to Open Router, an AI model provider we've talked about a lot on this channel, as it truly unifies all AI models.
But not only that, if we search for "free" here, we'll find a bunch of free models that we can use through their apps. So, the first thing we're going to do is create a free account. Once we have the account created, we have to click on "Get App Key.
" After doing this, we have to click on "New API Key" and give it any name. In this case, I'm going to call it "Tutorial. " We click "Create," copy this API key, and paste it into the "API Gateway Key" field.
Now we simply have to change the Open Router gateway URL , which would be this one here. Keep in mind that this provider, like some others, Others we'll be looking at throughout this video, I'll also be leaving an ebook in the description below that I created with all the details of the Gateway URLs and a step-by-step guide to connecting them, with more detailed explanations. But really, with this information, if we now click on "Test Connection," you'll see a green dot and a message below indicating that we 've connected to approximately 82 available models.
Now, if we click "Apply Changes" and restart, you'll see that after the countdown, the application will reopen with all the models from this provider. Notice that if I now click on the "Models" section, we'll find countless more models, including some free ones. For example, I just found one that says "free" in parentheses, but I also wanted to show you how we can even select cloud models to use through the APIs.
This one would require some credit, as these models are quite expensive, but the good thing is that we'll have the flexibility to choose whichever one we want. In this case, I'm going to select this one here, which I wasn't familiar with. And again, if I ask it anything, like why open source is important, when I click send, it will start processing.
And here, it will be writing the answer, only unlike what we saw in the previous section, here it will be running directly in the cloud with a provider other than Lama. Here, in fact, we'll be able to see all the answers it has given us, and if you're interested, you can also pause the video to see it in detail. The most interesting thing about this is that new providers are emerging with very aggressive offers thanks to the availability of new open-source artificial intelligence models with very low costs or that can even run on their own servers.
Therefore, we can find tools like Orca Router, where if we click on models and filter by the lowest price, we can see that they include free models, like these two here, but we can also find more well-known models like those from Tensen Hunjuan, Quen 3. 8, or even Deepsic's own models. So we can also connect it with any other provider, like this one, for example.
To see this, I just clicked on my account to create an API key. Notice that I'm also going to give it a name again here, in this case, "tutorial". I click "create" and copy this API key, pasting it where it says "API" in this developer mode , and changing the URL to the correct one.
And with this, we could repeat the process to move between providers. Keep in mind that for all of this, the ideal would be to have a platform that unifies all the providers and free models so that it switches between them once we reach the limit. But to see all this, keep in mind that I already mentioned a tool called OVN route that actually makes this possible.
[music] In this fifth and final use case, we'll quickly see how we can also interconnect with any official model. We're no longer just talking about connecting with providers like Open Router, Orc router, or others, but we can actually go to the API of a model we really like, such as Deeps Seek, Kimik's models, or even Zai's GLM models, and connect it directly to the Cloud desktop version, since it also has very good capabilities, but at a much lower cost. To do this, I would just come to the official Deeps platform, which I'll also link below in the description.
And if we click on VIP access, we can add a credit from there . I, for example, can see here that I added $2, of which I've only spent $0. 40 on various applications I've been working on.
This video here. I also mentioned to you what was a kind of cloud code, but Depsc just released it, and I'm going to go out on a limb and say I actually like it better because it's much more flexible. But continuing from where we left off, if I go to the API key section and create a secret key and copy it from here, we could go back to the API key section, replace it in the base URL, and change it to the correct one.
With this, we could actually test the connection, but it's very important that we have the correct Gateway base URL and that it's compatible with this method. I'll explain all this in detail in the PDF I'm going to share with you, because, as you can see, this particular instance is giving me this error. If you want to avoid it, that document I'll link in the description will show you which URLs you should use and which ones can be used this way to prevent these kinds of errors.
Now you can take advantage of the most powerful tools available without being limited by a license that dictates how far you should go, because sometimes barriers don't disappear until you find a different way that allows you to move forward.
