¡PewDiePie acaba de Reventar la IA! Lanzó Odysseus, IA GRATIS y SIN LÍMITES
I paste it here, click submit, and everything will start running.
Neither China nor the United States. Sweden is home to one of the most influential YouTubers in history. His name is Felix.
O que voce vai ver
- With this, we're no longer just talking about interacting with a personalized agent; we'll actually be able to work with an entire team, communicating in parallel both amongst themselves and with us.
- Here, for example, we've made a to-do list application, but I've previously made applications like Snake games, calculators, and others, and everything was functional and I could use it directly.
- Furthermore, we also have a section here for everything related to artificial intelligence skills, where we can see all the skills it has automatically created or that we have created ourselves.
- I pasted the API, specified how many results I wanted (in this case, 10), and clicked " test." Since everything appeared to be working correctly, it gave me the option to select it so quickly .
- If I click Start, I'll start a test so that if we send any query, like "Tell me an interesting fact about what I could do with AI agents," both models will start processing it simultaneously.
He was the biggest YouTuber in the world for years. And what does this have to do with AI?
Today, with over 100 million followers, he has just made history in the world of AI. A few days ago, he launched Odysseus, an artificial intelligence platform that aims to end dependencies on Big Tech, offering privacy and removing limitations. I've been using it for the last three days since its launch, and it truly is the most complete open-source AI platform I've ever seen.
Por que isso importa
To prove it, we'll be exploring what Odysseus is and testing it through seven use cases, from creating our own agents and automating any type of task to conducting in-depth research, using advanced features, working with images and artificial intelligence, and even using it from any device, among other uses. The best part is that this developer has made a bold move by offering this platform, which we can use completely free of charge, with unlimited access, and privately. Sounds good?
Let's learn more. [music] Obsidius is a custom AI workspace.
Its interface allows us to work with any AI model, whether through chats, autonomous agents, tools, emails, or other integrations. It prioritizes privacy, doesn't use telemetry, and everything stays between you and your device.
Na pratica funciona assim
To learn more, let's jump into Obsidius, the platform I'll be providing information about in the description below. Upon landing, you'll find a page that highlights the key points I've mentioned: a workspace that runs on your own devices and operates locally, privately, without telemetry, and with the models you choose. From here, we'll find a breakdown of all its functions, but since we'll be looking at them all, let's move on to the installation.
To do this, we'll scroll to the bottom, and you'll see that we only have one command line here, since it's open source and free, so we can install it easily. To do this, we simply have to copy this command from here.
We open our terminal, and note that we only need to see the terminal to paste this command, as its actual use won't even be here. So, I'm going to paste it, press Enter, and that's how we'll have it installed.
Pontos chave
In my case, it tells me that it already exists, since I had it installed previously. And once we have this, to see the next steps in more detail, we would have to click on "View on GitHub," since from here, by scrolling down a bit, we'll find details on how we can start this application depending that we simply have to copy and paste, but this will depend on the device you're using. I, for example, use Apple Silicon processors, so I 'll run it directly from my Mac.
As recommended here, since Docker doesn't directly utilize the GPU in Apple processors, it's better to use it directly. So, in my case, since you already have it installed with the previous command, I simply go to the second command that specifies the application's directory, paste it, press Enter, and once there, I just need to add this last command.
I paste it here, click submit, and everything will start running. Keep in mind that you could also do this completely automatically with any AI agent, such as Google Antigravity, Cloud Code, or even Codex.
Cuidados necessarios
You could simply share the GitHub URL, tell it what device it has, and have it follow all the installation steps. However, as I mentioned, you only really need one command and then another to run it generally. Once you do this and create an account—it will simply ask for a username and password— you'll see an interface like this one.
This means we can start using it. So far, we've only had an interface with many functions, so now let's move on to the second case, where we'll be connecting artificial intelligence models.
[music] We'll be looking at all the ways we can connect artificial intelligence to this application, which we already have in a private and proprietary way. Keep in mind that we have two main methods.
Ferramentas uteis
First, we can run models that we have installed locally. This way, no information will leave the application we just installed.
Second, although its creator doesn't recommend it as much, we can also run it through APIs. This way, all the execution and processing would be done through other servers , and we wouldn't need a powerful computer to use the best and most advanced artificial intelligence models.
We'll look at both methods, and I'll also be discussing other features that I haven't seen in other tools. And getting back to the platform, let's start with a feature I haven't seen in other tools to kick off this first section strong.
For this first example, we'll be looking at how to install models locally. And here's the easiest way I've ever seen: if we click on the "tools" section, we'll find a bunch of functions.
We'll look at them all, but let's start with the one called "cookbook." Clicking on it will automatically scan all the hardware in your computer. In my case, I have an Apple M3 Pro processor with 13-bit RAM and 18 GB of storage.
So, with all this information, even though I haven't provided it, it will already tell me which models I can run perfectly locally. Here, in fact, as you can see, I would have all these here, but as I scroll down, larger models appear that I wouldn't recommend installing, like the Quen model with 14 billion parameters, others like Gemma's with many billions of parameters, and so on. So now I can go directly without missing any of the models, select any of these recommended ones, such as this Deepsek R1 model.
I simply have to click, and it will give me a couple of options: run it or download it. I'm going to click download, and that way, it's automatic. Notice that I'll be doing all the work for me.
With this, we can say goodbye to the difficulty of working with local models. We no longer have to install one and see if it will work on our device; this tool will scan beforehand to see which one will work, and that way, before installing the model, we can be sure it will work perfectly.
However, now imagine you already have models downloaded locally through other tools like M Studio or Lama Yang, or others. You can easily reuse these as well. So, going back to where we left off, notice that we also have the model downloaded here, ready to use, but we can also work with those models we already had.
To do that, I'm going to close this window. And notice that if I go to the settings section in the bottom left, within the AI models section , I'll be able to see all the servers I've connected, both local models and those via API.
I'm seeing information because I've been testing it these past few days, but I'll delete everything so we can see everything from scratch. So, to do this, I'm going to delete the Olha models I had installed locally. I'll also delete a couple more that correspond to APIs, and this is how you'll see it.
Let's start by linking models that we already have installed locally. For example, in my case, I have three models in Olama.
So, for that, all we would have to do would be within the In the local section, we would need to enter the endpoint here using a URL. Keep in mind that all tools like LM Studio, OLAMA, and others have this feature, so we'll be looking at how to do it with OLAMA.
To do this, by doing a quick internet search for "Olama Endpoint," you'll see that the URL appears directly here, which is this one. So I'm going to copy it and paste it into this section.
Now, for this to actually work with the application you have installed locally, in my case Lama, I just went into the settings section and enabled this option, which is normally unchecked. This is called "Expose Lama to a network" so that other platforms can use it. With this done, if I now click on "Test," you'll see how it tells me that it has found three models that I have installed locally: YMA 4, Quen 3.5, and another YMA 4 variant.
So now I simply had to click on this "Add" button. By doing that, I would already have those three models available to use from this platform.
If we go to the Do Lama tool I had before, those are indeed the three models I had available. But now, going back to Odisseus, if I close this section, you'll see that in the models section I'll have precisely those three available and ready to use.
Another way we can connect AI models, especially if we have a computer that can't run a model with the capabilities we'd like, is through APIs. To do this, again, if we go to the settings and add models, you'll see that beyond being able to use models locally, we also have models via API. Here, from the selector, we can choose between the main providers, although I personally recommend both Open Router and Grock models, since they unify many AI models.
You could also add credits for them, but if you don't want to do that, you'll also have many other models that you can use for free running in the cloud. So, for this example, I'm going to start with Open Router, and it simply asks for an API key.
Therefore, in my Open Router account, I'll create a new API key and give it a name. I'll set a limit of one dollar, since the platform provides it for free.
I click "create," and then I simply copy and paste it into the tool. When I test it, I see that we have more than 300 models available.
So now I'm going to click "add." If I close that and click again on the models section, I'll find a ton of available models. But the most interesting thing is that in Open Router, if we type the word "free," it will show us all the models we can use for free without needing to add any funds.
Something similar happens with Grock, so you should know that, going back to the API section, you'll be able to interconnect as many platforms as the official ones from Antropic, Open, Gemini, and others. In this case, I'm going to connect a second platform, Grock, and copy the API key from the other platform, pasting it back here and testing it.
It tells us that in this case we would have 16 models, not only text models, but also transcription models, images, and more. And by clicking "add," as you can see, we'll be able to have several providers, both for local models and even for tools with APIs. With Odysseus already installed and with AI models, now comes the interesting part, as we'll be able to create real projects and automations.
Because it's increasingly easy to develop an idea, but we also need it to look professional. Therefore, I want to introduce you to Kittel, an AI-powered design platform designed to create professional brands, images, and assets in minutes.
Here you can generate images with models, edit the result, and take it to a final design without switching between other tools. It also has features like flows that allow you to create variations of an idea, test styles, and Colors and compositions with a single click.
And if you're creating a brand, you can work with logos, fonts, graphics, templates, and keep everything consistent with your brand image. Crucially, you can create mockups with a single click to see how your design looks in real life, like on t-shirts or other products.
They also offer templates where you not only have a design, but you can bring it to life with video formats. And hey, depending on the plan and license, you'll be able to use your designs in commercial projects. Thanks to Kittel sponsoring this video, they've given me a special link, which I'll leave in the description below, so you can try it out for yourself.
Now that we know this, let's look at the next use case. [music] In this second use case, having already covered the groundwork, including installing the platform and interconnecting any AI model we want, we'll now see how we can interact with this tool through chats and AI agents.
To do this, notice that the platform has an interface very similar to a GPT or cloud chat, only here we have more advanced features and can choose from any model. I 'm going to start with the free Open Routro model, and once I've selected it, you'll see that since it's a locally installed agent, it can really help us with anything , as it could even manage our computer. To show you the information it has, I'm going to ask it, "Hey, could you list the files I currently have in my downloads folder?" I'll click send, and you'll see how the agent, after processing this, will enter another sub-entity where it will access the downloads folder, and indeed, it will have all these files there—all the documents I actually have in my folder.
And with this, beyond just being able to have normal interaction, it will be able to control our computer and any tool we want to automate. From here, keep in mind that we'll have both agent mode and chat mode.
The chat mode is simply for asking specific questions and getting answers. If we click on "New Chat," we'll go directly to a new interface. If we ask it any question, such as, "Tell me three ways I could make money using AI agents," it will have already processed it.
And regarding the chat itself, you'll see the answer here, which it has also gathered from various information sources. Beyond interacting like any AI chatbot, we also have different tools down here, such as the ability to enable or disable internet searches, and the ability to execute terminal commands to automate any kind of task. But before we look at more features, I want to show you how, directly from the agent section, it can execute different tasks and even preview applications we create here.
For a quick example, I'm going to ask it to create a simple HTML application. I click send, and here, notice how a preview opens here where all the code should be displayed.
In a few seconds, sure enough, a simple code snippet for that application appears here, which we can also preview. To do this, simply by clicking on the eye icon, I can see the application it just created.
I can enlarge it and even use it from here. I can also return to this preview and zoom out even further.
We could also tell it to work on projects we have locally, but one of the most important things to keep in mind is the model selection. And here we also have a very interesting feature... If we close the Tools section here and go to the Compare section, we'll find an interface where we can compare all the available models.
For example, I could use the free Yma model with 31 billion parameters and compare it to the more powerful open-source Openi model with 120 million parameters. Now that we have these models, I can compare them in chat, agent, search, or deep research modes.
However, I'm just going to run a chat. If I click Start, I'll start a test so that if we send any query, like "Tell me an interesting fact about what I could do with AI agents," both models will start processing it simultaneously. Here we can see that the second model is a bit faster in its response, and the first one hasn't even given us any answers yet.
So here I'm going to vote for the one I liked best, and that way we can see which model really works best for us without bias, knowing first which one we selected. Also, to finish up in the chat section, notice that if I start a new session today, we'll have advanced features by clicking on this section here.
Beyond being able to attach files and documents, we'll also be able to play around with the whole prompts section. I'm going to click on it, and you'll see that from here I can add a prefix or suffix to all prompts to tell them, for example, to be direct.
If I put "be direct and concise" in all my questions, it will automatically add this, even if I don't see it, so that it always follows this rule. In addition to this, I'll be able to adjust the token creativity index, but what caught my attention the most is that we also have a persona option here, which allows us to create our own conversational agents.
Here, in fact, to test it out, we already have some like Socrates, with whom I can have a conversation if I click on "Start Persona." For example, if I say to him, "Hi, how are you?" and hit send, he'll be here replying. But the most interesting thing is that we can create our own personas, that is, like custom GPTs, however we want. And we can add all of these as a kind of WhatsApp group so they can interact with each other.
To see this, I'm going back to the "Person" section, and within the "Group" section, I can specify how I want them to interact: sequentially or in parallel. I'm going to add sequential. From here, I'm going to add Socrates with the Dolphin model from Mistral, and I'm going to add Spark with the Yenma 4 model.
I click "Start a New Group," and that way, besides me interacting, they'll be interacting directly with me and with each other. With this, we're no longer just talking about interacting with a personalized agent; we'll actually be able to work with an entire team, communicating in parallel both amongst themselves and with us. These teams could have skills such as application development, document organization, and we'll be able to access all of this in a unified way.
To wrap up the chat section, I'm going to start a new chat so we can see find various shortcuts to make the user experience even faster. Starting with the third use case, we'll move a bit faster so you can get a comprehensive overview of everything you can do with the tool, as it offers a multitude of features. And in this third use case, we'll look at another very interesting tool: Deep Research.
It's helpful to know that we could do an internet search. Notice that among the tools we also have this in-depth research section, so I'm going to click on it and from here we'll be able to do our research. An automated method.
Notice that I encountered some errors here, and this is because both the default search engine, which is completely free, and the one that appears as the default (meaning if the default one fails, this other one pops up, which is also free and doesn't require an API), were failing. So I went to this tool called Tavili, a platform which is this one here, and I'll also leave the link below in the description.
I realized it has an API that gives us 1000 credits for free. So I created one, copied it, and from this tool, going back to the settings section, I went to the search section and selected Tavili. I pasted the API, specified how many results I wanted (in this case, 10), and clicked " test." Since everything appeared to be working correctly, it gave me the option to select it so quickly .
So now, if I ask it anything from here to do a research project, like, for example, "What do you think the future of artificial intelligence will be like? Will it be very expensive and inaccessible, or on the contrary, will it become democratized and we will all have free access to it?" I'm going to ask it this question here. From here, I'm going to tell it to run the rounds automatically.
Use that search model. The point, that is, the tool on which the artificial intelligence is based.
In this case, I'm going to use Grock and I'm going to stick with the GPT model of 120 billion parameters. I click " start," and right here, as you can see, it would begin that in-depth research planning, where, as it shows us with these animations here, it would be branching out all the information sources automatically.
In fact, as I've already mentioned, a short 15-second research report appears here . It actually took quite little time, so let's see if it did it right. And to do that, I'm going to click on "view report." From here, all this research would indeed appear to us in a very clear and visual way.
With the question we asked , it generated an image using artificial intelligence, and here it tells us that in about 13 seconds, it performed two rounds, four searches, and acquired 14 URLs. We now have this kind of report here, which is actually quite comprehensive.
But here's another report from a previous test I was doing; as you can see, it really depends on the type of question we ask, and the extension adapts accordingly. [music] In this fourth use case, we'll be talking about the entire organizational aspect, both for us as humans and even for the artificial intelligence agents that are performing tasks not only when we ask them to, but also autonomously. To do this, notice that the tool has a section called "Brain," and if I click on it, we can add or even modify the memory of the AI agents we have installed locally.
Here, in addition to being able to modify it, keep in mind that everything is completely autonomous, so the more you use it and interact through this tool, the more it will save everything in its memory. But note that unlike other major technology companies, you will only be able to access this memory. Furthermore, we also have a section here for everything related to artificial intelligence skills, where we can see all the skills it has automatically created or that we have created ourselves.
And each of these, again, we can edit, test, audit, or even delete. If we want to add new things, either to the memory or to the skills, we simply have to click on this "add" button.
And from here, notice that we can add something to our memory. I, for example, can say, "Remember that my name is Aleji.
It's already here." I press Enter, and it tells me that it has been added from memory. We could do the same with all the agents' skills. And to put this to the test, notice that if I go back to the part From memory, it displays this new phrase here, which we had previously set up.
And if I now go back to a chat and ask, "Hey, could you tell me my name?" And then, if I switch to agent mode and hit send, this free OpenRouter model will start processing it right away. And notice how, in its reasoning, it would be searching its memory.
And seconds later, as you can see, it has indeed known that my name is Alex Javi, regardless of the model we use, whether it's OpenEI apps, cloud apps, OpenRouter apps, or even local models. With this, we can say goodbye to having to train each platform or worrying about what happens when we switch to another, since it will be able to know everything we want in a centralized way. Keep in mind that the whole memory aspect is important, but the skills will enable people to do anything professionally.
But if you'd like to learn more about this, I already have a video here where I talk about it in detail. In this fifth section, we'll see how agents can perform tasks completely autonomously using three key tools. The first tool I want to show you in this section is called "Task." If you click on it, you'll see a multitude of tasks that can occur without your knowledge.
Many of these are currently inactive, such as classifying calendar events from emails, summarizing all your emails, and other actions. Here, I have several tasks that these agents are currently running, but keep in mind that you can pause them all, or even start or stop them individually . You'll also have complete control, being able to see everything the agents are doing and when they did it.
From here, you can even add specific tasks, such as scheduling a prompt to run or writing a specific prompt that you configure. For example, when someone rings the doorbell, we could automate this and all these other actions. For a quick example, imagine I wanted to see something on a recurring basis.
I could click on "pronouncer" and from there I could say, for example, "AI news," and the prompt would say, "Tell me the three most important news stories in the world of generative AI that happened yesterday." Once I have this, I would simply select when I want the prompt to be written, for example, daily, and the time I want it to occur. In this case, since it's almost 9:30 PM, I'll set it to 9:35. I'll tell it to output a session, although it could also notify me or even send me an email.
It's that easy: simply select it here and click "create." The task would then be active; I would just have to wait, and when that time arrives, it will notify me here that it 's active. If I go to the activity section, you can see that it has indeed been executed here, and it's telling us all these most relevant news items. This is related to the rise of AI and other results.
With this, we're no longer just talking about using private and free AI, but we're actually going to be able to automate things that were previously very complex to do and that we'll now be able to do with a single click. Regarding the platform, keep in mind that beyond the tasks section, which would also cover many use cases, we also have another section called notes. These allow us to add reminders.
Here, for example, I set a reminder to create an application this Friday at 9 a.m. Here, I can also add more notes, a tutorial, or even draw. These tasks that you see here, to which we can also add reminders, can be useful both for us and for the AI agent itself, so that it can take all this organization into account.
We'll keep all these notes here. Creating one is as easy as writing a title.
I'll put "test" here. I'll add "optimizes functions" here, click save, and that way I'll have these two tasks that I can highlight as a higher priority, so that neither I nor others forget anything. In addition to this, we also have a calendar view where we can see all these notes, like the one I marked earlier: "Good at 5, create an app," which will also allow me to see all this in a much more user-friendly way.
Again, to create one, I can click this button, enter something as an example, like "check email," give it another date, in this case Saturday, click create, and that way I'll have this new event set up here. Also, to wrap up this section related to automated or organized tasks, notice that if we go to the chat section, we'll also have all the sessions we've had saved.
Here, a logo will appear based on the model we've chosen, even though it all comes from the same provider. From here, you can also organize everything by folders or even by artificial intelligence, making it reorganize everything. I just clicked on it, and in just a few seconds, it created different folders for all my chats.
For example, folders for my downloads folder, calculator apps, general chats, AI model chats, personal identity chats, and other things that can also be reorganized automatically. [music] Moving on to the sixth and penultimate use case, we'll see how this platform can also work with the visual aspects.
To do this, notice that in the left sidebar, we have a function called "Gallery," and from here we can upload our photos. In this case, for example, I have the latest thumbnail from this video I just uploaded where I showed how to create Android applications. So I'm going to select it.
And notice that we also have an editing option here. If I click on it, I select that photo, and you'll see that I have a kind of Adobe-like tool here, since this YouTuber is also trying to compete with Adobe Photoshop. However, he himself says that this tool hasn't quite reached that level yet, although he's working on it.
The really great thing here is that we have much easier-to-use options, thanks to a lot of artificial intelligence. We can click on tools like, for example, crop, and that way we can select how far we want to crop.
For example, I'm going to leave out that bottom part that says "free." I click "apply," and that way it's cropped. We can also transform the image.
I, for example, could expand it and widen it a bit more. I apply it, and there it is. But we also have a lot of artificial intelligence, since we can use this kind of brush to mark everything we want to modify, and we'll be able to work with tools that have in-paint capabilities to add things using artificial intelligence, erase them also using artificial intelligence, remove, for example, the background using AI, but these might require installing some tools, such as this AI background remover that we can install with a single click.
[music] Finishing with the seventh use case, we'll be looking at advanced settings so you can see the total customization and control we'll have in this tool. To do this, we'll first finish with the last two tools, one of which is the Library, from where we'll be able to see everything we've done in a unified way, from the chats we've sent to the documents we've worked on.
Here, for example, we've made a to-do list application, but I've previously made applications like Snake games, calculators, and others, and everything was functional and I could use it directly. From here, we can also see all the in-depth research we've done and everything we've previously archived.
Besides having complete control over everything can also click on this incognito mode option. This prevents people from saving anything we're about to discuss to our device's local memory. That way, everything we talk about from this point on will be temporary.
In addition, if we go to the themes section, we can completely customize all the colors that appear in this tool. I have the default, but we can also change everything with a single click, as you can see: a lighter theme, a darker theme with a kind of background rain effect, like a terminal with black and red and those background animations. And ultimately, we can find a lot more options, such as a GPT theme or even the cloud branding.
The most interesting thing is that whichever one we select—in this case, I'm going to stick with the cloud one—you'll see that if we click on "customize," we can not only change all the colors, but we can also add different animations. For example, instead of a solid background, I can add the constellations you've seen since the beginning of the video, as this is my favorite .
But there are also different animations, like this one here called "flow," which has more complex animations. We could also add rain, and just to look at one last option, we could also add these moving squares. However, I'm going to stick with the constellations again.
I'll close this here, and that should apply it. Furthermore, if we return to the settings section, you'll see that we also have an integrations section where we can add all sorts of things. We can add things via API, import contacts, or add our own email server.
By adding our own email server, the artificial intelligence will be able to read all the emails we want. Keep in mind that since it can run models locally, everything remains confidential on your device.
Beyond simply reading them, it will be able to automatically reply to all incoming emails . It can even summarize the email to highlight key points and other features we might need with this integration.
Before continuing with the topic of emails, note that in the integrations section, if I clicked the "+" button, in addition to adding those features, you can also add MCPs, which allows you to interconnect with any tool you want. In addition to this, and now returning to the email section, you'll see that from here you can manage all your emails, allowing you to add, edit, delete, and perform any other actions you want through email.
You can also add tasks and even customize your writing style to make it seem like you're the one responding. Furthermore, you can configure the reminder settings so the agent will contact you, either via notification or email.
Along with this, you also have keyboard shortcut options, the ability to choose which tools the AI agents use, and even the option to set up a user system so that, in addition to using these tools yourself, you can grant access to members of your team. To do this, you simply need to enable this option, and anyone who wants to access the system will have their own email and password to access this work environment.
Moreover, you can even use all of this from your mobile device, whether it's Android or iPhone. And the truth is, all of this offers a lot of freedom with few barriers. So now the question is no longer so much how we can access it, but what you will be able to do.
Gostou do conteudo?
Deixe seu comentario, compartilhe com seus amigos e inscreva-se para receber novos artigos.
Dê sua opnião deixe um comentário