← Back to search

ATA 802 Connected Hermes Agent to the Internet and YouTube

Atareao con Linux · 2026-06-04 · 27 min
relevance 79 4240 words Episode page ↗ Audio ↗
Show full episode description
Si en el episodio anterior te hice una pequeña introducción de lo que es capaz de hacer este agente, hoy quiero bajar al barro contigo y contarte cómo le he dado ojos a nuestro modelo para que pueda explorar internet, documentarse y encontrar la mejor información de forma completamente soberana, libre de anuncios y con el máximo respeto por nuestra privacidad. En este episodio nos vamos a centrar en dos "conectores" brutales que he integrado en su arsenal: uno para realizar búsquedas en la web general y otro para realizar búsquedas avanzadas en YouTube. SearXNG: Tu propio Google privado y sin publicidad La primera pieza de este rompecabezas es SearXNG. Es una herramienta maravillosa que actúa como un buscador de buscadores. En lugar de ser un motor de búsqueda que rastrea la web entera por su cuenta, SearXNG lo que hace es consultar de forma simultánea a los grandes del sector: Google, Bing, Brave, DuckDuckGo y los que tú decidas. Recopila las respuestas de todos ellos, limpia los molestos rastreadores que intentan identificarte, elimina los anuncios y te devuelve una lista limpia de resultados. Invidious: Buscando en YouTube con total soberanía La segunda herramienta que le he dado a Hermes es Invidious, y tiene un propósito muy claro: facilitarle el acceso a la inmensa base de conocimientos que es YouTube. La magia de los subagentes paralelos Lo que me tiene entusiasmado de este sistema es la capacidad de Hermes de combinar ambas herramientas de forma autónoma gracias a los subagentes. Imagina que le pido a Hermes que me recomiende cómo aprender a programar en el lenguaje Rust. En lugar de darme una respuesta estática, Hermes decide de forma inteligente lanzar dos subagentes: uno se va a la web a través de SearXNG a buscar documentación oficial y artículos de referencia, mientras que el otro se va a YouTube a través de Invidious para buscar cursos en vídeo. Ambos asistentes virtuales regresan con sus hallazgos y Hermes hace un cruce de datos espectacular. El poder de los contenedores Docker Para montar todo esto sin complicarme la vida ni ensuciar el sistema operativo de mi equipo principal, he recurrido a mis queridos contenedores Docker. Cada herramienta corre en su propio compartimento aislado. Da igual que SearXNG use unas librerías de Python concretas o que Invidious requiera otras dependencias distintas; al estar encapsulados, nunca entran en conflicto. Si quiero actualizar algo, simplemente descargo la nueva versión del contenedor y listo. En la entrada del blog te dejaré las plantillas de Docker Compose preparadas para que tú también puedas desplegarlas en tu servidor y empezar a cacharrear hoy mismo. CAPÍTULOS DEL EPISODIO:
✨ Episode Outline — click any point to jump to it in the episode
Problem solved
Giving Hermes private, ad-free web and YouTube search via self-hosted MCP tools for digital sovereignty.
Benefits
  • Private web search returning JSON, no ads
  • Ad-free, tracking-free YouTube via Invidious
  • Self-hosted on home Slimbook One in Docker
  • Subagents cross-reference web and YouTube results
  • Automated documents and PDFs from live web data
Use cases
  • Searched BASH/Linux scripting articles via SearXNG MCP, returned guides and courses
  • Generated a PDF presentation by auto-searching the web and assembling it
  • Returned 20 YouTube videos with title, channel, views, duration via Invidius
  • Found Ned Gorchak video with 947,000 views and Nate Kirk 1-hour course with 222,000 views
  • Launched two subagents to find best Rust tutorials and gave a reasoned recommendation
KPIs / results
  • 20 videos returned with metadata
  • Top video 947,000 views; Nate Kirk course 222,000 views
  • Two Docker containers: SearXNG plus MCP
Tools / build
0:00 / 0:00
📑 Chapters — tap a time to jump there
00:00:00
Presentación: El asistente de IA definitivo y la soberanía digital
  • Intro: ultimate AI assistant and digital sovereignty
00:01:40
¿Qué es un MCP (Model Context Protocol)?
  • What an MCP (Model Context Protocol) is
00:03:26
SearXNG: Tu propio motor de búsqueda privado y sin anuncios
  • SearXNG: private, ad-free search engine returning JSON
00:05:53
Poniendo a prueba las búsquedas web con Hermes
  • Testing web searches with Hermes via MCP
00:07:32
Cómo está montada la infraestructura de búsqueda con Docker
  • Search infrastructure built with two Docker containers
00:08:48
Automatización real: Creación de documentos con información web
  • Real automation: generating documents from web info
00:09:38
Invidious: YouTube sin publicidad y sin rastreo
  • Invidius: YouTube without ads or tracking
00:11:43
Buscando y analizando vídeos de YouTube con Hermes
00:16:20
Superpoderes combinados: Búsqueda web + YouTube
  • Combined powers: web search plus YouTube
00:20:40
Por qué autoalojar todo en contenedores Docker independientes
  • Why self-host everything in separate Docker containers
00:22:04
Analizando la recomendación razonada de la IA para aprender Rust
  • Analyzing the AI's reasoned recommendation to learn Rust
00:24:09
Próximo episodio: Planificación de menús y bases de datos vectoriales
  • Next episode: menu planning and vector databases
00:25:09
Taller presencial en Linux Center de Slimbook (¡ven a cacharrear!)
  • In-person workshop at Slimbook Linux Center
🌐 This transcript was automatically translated to English from the original.
Hello, I'm Lorenzo and this is Atareao con Linux, episode number 802. Last Monday, in episode 801, I introduced you to Hermes for the first time. I simply gave an introduction to what Hermes was, what Hermes Agent is, what options it has and I was also showing you different characteristics, different abilities, different properties that could be done with Hermes. That is, to end up having an artificial intelligence assistant that helps you do almost any task you can imagine. In this sense, I was providing him with a series of tools, many of which I showed you in that episode, but I stayed only on the surface, what tools I had given him, what tools I had made him and what possibilities he had. In this episode I want to go into two of the characteristics that I found most interesting or two of the characteristics that are really very interesting and that, let's say, go beyond simply a model of intelligence. And basically they are two tools that will allow us to carry out two very interesting operations. The first is to search the internet and the second is to search on YouTube. These two operations are actually very powerful. The first because it will allow us to document and the second because it will allow us to complete the circle of everything we are looking for. And for this, one of the issues I talked about in the previous episode was MCP, which is neither more nor less than a Model Context Protocol or, rather, in the end it is a tool. A tool that we give to our artificial intelligence agent so that it can do things. Well, it's really a tool that we give to our intelligence model, to our language model. A language model that obviously urgently needs to have the ability to use these types of tools. Specifically, the two tools are SerksNG and Invidius. And they are two tools that go completely beyond or let's say they go beyond simply a language model. Because on the one hand it is a tool for searching, as I told you before, and on the other hand on YouTube. And it's not that it is a tool, but in the end, for example, Invidius or, rather, SerksNG is a search engine. A search engine search engine because it will allow you to perform searches far beyond simply having a tool that we have given to our people. And the same thing happens with Invidius. And this is precisely what I want to talk to you about in this episode. Which is not only about intelligence models, but also about digital sovereignty. Meanwhile, I am going to give you two tools with which you can perform searches in a slightly more private way. So let's go straight to the turn where I'm going to talk to you about Invidius, SerksNG and those two tools that I'm going to give to Hermes. And the first of the two that I am going to talk to you about is SerksNG. SerksNG is neither more nor less than a tool that will allow you to perform searches. The great advantage that SerksNG has for me and why I have searched for it is because it is a tool that allows me to perform searches directly and returns the results in JSON format. And this is very powerful because it is very easy to integrate it directly with our intelligence model. In this sense, you can choose or you have preferences of all kinds of options regarding options that amount to general options. Others that are related to the user interface. Others related to privacy. The motors that can be used, you can select the motors you want. For example, you can select Bean or Brave or DuckDuckGo or Google. Either all of them or the ones you want. You can define a series of special queries and then you can do a treatment with cookies. This is basically the options that SerksNG offers you. So in the end it is a search engine just like any other. If I now search for Hermes Allen in this search engine, what it will return directly to me is a search where it has found different pages that have Hermes Allen. With a fundamental advantage, that you will not find advertising. At least you won't find it here. This does not mean that Google, for example, or DuckDuckGo or whoever, is not able to identify you. Because what I have done is host both SerksNG and Inviduos inside a Slimbook One that I have at home. So, of course, in the end it will recognize the IP it comes from and it is very likely that it will know exactly who I am. The reason for having it directly at home, I will tell you this a little later so that you have an idea of ​​the reasons. Although it can also be done in other ways. But I preferred to do it this way. So, if for whatever reason I search for, for example, Task or to do the typical thing that I always do, what I will do is perform the search and we will find different options, different entries related to Task. That said, and once you know this, the first thing is, for example, to ask him or her or do a search, a simple search, right? Search articles about BAS. I mean the Shell, right? To the Shell BAS. So, I have simply told him this and now what he is going to do is search directly, using this MCP, using this tool, what he is going to do is search for articles related to BAS. That is, what you are going to do is basically connect to SerksNG through the MCP, which is what MCP appears directly, MCP SerksNG. And what you are going to do there is do different searches. Searches for articles about BAS in Linux, BAS Scripting Tutorial 2026, BAS Terminal, Linux, Tricks, Scripts. There are different options that he is looking for to give me solutions. And here you go for different things he has found. Commands, scripts and tips, complete BAS guide, courses. Well, there are all kinds of things in both Spanish and English to do this. In the same way, just as I told you this, the great advantage is that you can search for anything you want. You can perform searches, for example, looking for articles about Hermes Allen. And perform exactly the same operation. You're going to do a search on Hermes Allen doing all this. How is this set up? Well, this is basically set up using two Docker containers that I will leave you in the episode notes. A first Docker container that is the one that builds SerksNG and another second container that is the one that builds the MCP. With these two containers I have given you the tools. Well, with a container, which is what I use to do the searches. And the second container, which is the one that the MCP gives me, is where I carry out all these types of searches. In the same way, I can tell him anything you can imagine. For example, launch or raise a subagent, a subagent to search for news about Busy with Linux. Let's see what we find out there. And it is simply this, a gateway to the internet. Depending on what you need, this is a very powerful solution because you simply have this type of search at your fingertips. But not exactly at your fingertips, but at people's fingertips. So when you need to do anything, it will be the agent who will be in charge of searching or selecting the tools he needs to do things. For example, the other day I needed to make a PowerPoint presentation. Well, in PowerPoint. A PDF presentation, let's leave it at that. A PDF presentation and what I told him is, I want to do a PDF presentation on this topic. And he directly did the basic operations, that is, he connected to the internet using this MCP, searched for the information and generated the PDF for me. I didn't have to do anything else. Well, maybe I'll retouch something, say, hey, on page 32 where you put I don't know what, you have to put I don't know how much. Or things like that. And this is basically the first one, the searches. The next one, the one about YouTube searches, is a little more special. In the sense that YouTube is very careful about us directly attacking its API. If you have a key, an API key, you will be able to access it. But if not, it always makes things a little more complex for you. In this sense, what I am using is Invidius. That Invidius is neither more nor less than another tool with which to search on YouTube. And it is very powerful because the first one, just like with SetsNG, you are going to completely forget about the... What is it called? Now it's not going to work out for me. For God's sake and for... You're going to forget about advertising. My goodness, it didn't work out for me. I was completely stuck. So, as I say, it's going to show you... I mean, you can do any search, like for example, something we talked about before. Look for Hermes Allen, which is so in vogue. Now I look for him and he is going to find me a whole series of videos directly related to Hermes Allen. Well, it's that simple, without having to do anything. This using Invidius. As I told you before, why do I have it at home? Well, the problem with having Invidius on a VPS is that basically what is going to happen is that since YouTube is going to realize that they are making API calls from there, it will immediately block the account. In other words, Daniel Primo said that there are tools, there are options, there are possibilities through Cloufer to avoid this, to deceive YouTube for the second time to do this. But it seems silly to me. In the end, when I work, I normally search YouTube at home, so if I can do this from home, then perfect. What's more, if I have to search for anything, I do it directly from Invidius and I have that advantage. In the same way, in the same way that I told you before, you can do the same thing that we have done with SerksNG, you can do it with Invidius. That is, now with Invidius I can tell it to look for me, look for videos, look for videos of Atariado con Linux. And what you are going to do is exactly the same operation. It will connect to Invidius, it will ask for this and there in Invidius it can directly attack the API, which will return the results. The great advantage, as I tell you, is that it is a frontend where you can do all these types of operations and where you will not have cookies, ads, or anything at all. It is true that, since it is coming out through the home IP, really what you are looking for is going to come out. But it's that simple. What do you find? Well, for example, I found, when I searched for videos related to Attached with Linux, I found the most viewed ones, for whatever reason, one of PDF in Linux, your own private AI in Linux, in short, a whole series of things. We can tell you now to launch a subagent to analyze them. Launch a subagent to analyze them. And, well, it is going to do the same operation that we have seen or that we have done previously with SexNG. What you are going to do is launch a subagent and that subagent is going to do its own operations. Here we have only made a list related to the videos, right? For example, a list where the most viewed or the most recent are. But just as I tell you that, you can do things that are a little more complex, right? Tell it to give you a list of title, channel, visits and duration. From, for example, InVidius. We can directly tell you that. Give me a list of InVidius videos with title, channel, views and duration. And with this we will generate exactly that information. Just as I tell you this, of course, you can already get an idea that this, precisely combined with the power of SexNG, you have two tools with which you can perfectly prepare any presentation, any work, any documentation without touching anything at all. It really is very simple. Look, for example, the subagent that we had sent previously has finished and has produced a list, let's see, a list of the three most viewed. And he has also made me, well, here he has made me a list relating, solely and exclusively to Task with Linux, regarding visits, durations. But now I'm going to tell you what you're looking for. Give me that same information from different channels and related to Hermes Ayer. And so he is going to do exactly that same search for us. Everything I'm telling you, obviously, you can do exactly the same through InVidius. With InVidius, well, you are not going to have this information, let's say, extracted in the same way, but you can do similar things. For example, launch a subagent to search for Rust tutorials. Well, now it has returned exactly 20 videos about Hermes Yesterday, where it tells us the title, the visits for each of the videos, the duration of each of the videos. Anyway, you have a whole series of options here so that, very quickly, you can see it. For example, as usual, he always gives you his point of view. He tells you curious facts and tells you, Ned Gorchak takes the cake with 947,000 views in a single video. Then you have Jack Robert with three videos. Sarbel ready. And then the Nate Kirk video. It's a one-hour course with 222,000 views. Well, anyway, you've already seen all the power. And we have one more thing left, which is to squeeze both together. That is, now it would be launching two agents in a way. An agent to give you YouTube results. Another one to give you normal search results. Let him cross both and give you a recommendation. Something as simple as telling him, launches two subagents in search, looking for information. No, looking for the three best Rust tutorials. A sub-agent who searches the internet and another on YouTube. When you have all that information, cross-check the results, for example, cross-check the results and give me a recommendation. reasoned. Now for reasoned, reasoned. Okay, that's it. In the other one, in the previous one, he had already given us a list of tutorials for, okay. The sub-agent that we had left running there says, he has returned and brings us three courses or three, look, three Rust tutorials. A tutorial that is videos, another tutorial. Well, it's three video tutorials and then three article tutorials. Now, since I have told you to make the combination, the most important thing is that you give us the recommendation. But if you realize, when I told him to do the operation, because I told him before, look for the best Rust tutorials. I've only told him that. Launch a subagent to search for Rust tutorials. And he has already directly decided to do a combined search of SexNG and Envidius, where he has selected the tutorials, on the one hand, of Envidio and the article tutorials. And he has given us the answer. We do not have the answer, the reasoned recommendation, which we will have now. But if you notice, he has already been able to deduce that the best solution is to make a combination of both tools. Here it already tells us, it returns us the results of each of the subagents that we have sent. With these two subagents he tells us, for example, the titles of each one. For example, here on YouTube, through Envidio, he tells us that MauroDev's Rust course from scratch. And as far as articles are concerned, it gives us, well, it gives us different options. And the reasoned recommendation says, start with the first one, which is Rust for the Impatient. Rust to delve deeper into the Rust Book. And then, if you are Hispanic or non-Hispanic speaking, MauroDev's. Well, it gives a roadmap, Rust Lang in Spanish, and the combination with MauroDev. The definitive shortcut, if you only have an hour, the MauroDev course. Definitely so. So, this is a brutal solution, because with this you can already do it. We could have even told him that, give me only results, only, or search only for results in Spanish. Only search for results in Spanish, so that it tells us exactly the same operation. But, notice that the most important thing is that he is now capable of making the combinations on his own. Using this. Well, to tell you a little about the infrastructure, which I think I have already told you a little about, which I have told you about before. The reason or explanation for mounting everything in containers is simply the fact that we are not going to directly dirty our equipment. Each of the containers will have all the tools inside it. And with the advantage that, regardless of what you do, regardless of, of, for example, SearchNG that is programmed in Python, uses a specific library, and Invidius that is programmed in Node, in JavaScript or in TypeScript, and uses a specific library, by having it in separate containers, each of them, you simply have to update those containers. Regardless of whether you have Python 13 on your team, and in those containers you have Python 10. That is the same. That is one of the great advantages. In the case of the MCPs, as I have set it up, it doesn't happen that much either. You could have them directly, because in the end they are MCPs that are mounted as binaries. But simply for convenience, I have done it this way. Thus, all you have to do is download the Compose that I have created for each of the pieces and assemble them. That easy. Well, now he's back with all the subagents and he's telling us how we can learn RAS easily. And it gives us the following combination. The first is a contact using Mourodef. The following explains each of the things to us. He tells us, it's an O, for example, Mourodef's says, it's an hour, in one hour you have the fundamentals without getting overwhelmed. Brace explains very clearly. Then the next would be theory with practice. I made you the RAS book plus RAS 101. And it says, the Bible translated into Spanish and a tutorial from a reference in Spanish. Which is the perfect trend. The third, which is a total immersion, which is almost two hours, where he tells us about Mourodef and learns RAS from scratch. And it is a widely viewed video in Spanish. And the last one, to reinforce concepts, are the fundamentals of RAS, by Carlos Azaustre, where he talks about ownership, borrowing and all this type of things, but well explained. So, as you see, you have something really powerful, where you can combine everything, you have it in a second, well, in a second, in a few seconds, and the truth is that it has brutal potential. As I tell you, for me this solution is spectacular, because it allows you to do anything without having to deal with ads or internet searches. You have it like this, in a very simple way. And you have provided the tools for this to work alone. And this is a little what I wanted to tell you in this episode. As you can see, I have talked to you about search MCPs, both SearchNG and Invidius, whether for privacy, because you will not have ads, because you do not have all this type of things, although I remind you that if you do it from home, in the end they will know where you come from. Then you have the advantage of Invidius, because you will be able to do searches through the API. And as you see, you can do parallel searches, using subagents and to create all types of content. That said, next Monday, what are we going to see? Well, next Monday we are going to see the first RAG, but also combined, because as I showed you in the previous episode, we are going to see the topic of recipes, menus and how to do all this shopping list stuff. But in a much more concrete way. I'm going to show you how to make a RAG, how to extract recipes, how to prepare everything, how to combine everything and put it all in a vector database with some supplements, because it's going to be a mixed database. It's going to have a RAG part and a normal part. And how to have everything set up so that once a week it is able to generate everything you need. Apart from that, what I'm going to do is tell it to directly generate a sheet for us, a calendar for the following week, which is the lunch and dinner menu. And finally, before I forget, I remind you that on Saturday, Saturday the 6th of the 6th of 2026, we have at Slimbook, well, at Slimbook's Linux Center, we are going to give a couple of talks, one referring precisely to Hermes and another referring to an agentic solution with OpenClaw. So you can see all this and all the possibilities. Remember that it is also a workshop where you will be able to dedicate a few minutes, well, not a few minutes, a good while, I think a couple of hours or three hours to test all this directly with your own hands, to use an agent directly, all thanks to the fact that Slimbook is going to give us 6 or 7 computers to do all these types of tests. Remember, I'll leave a link in the episode notes if you want to attend. And I think that with this we already have everything. I have already explained two fundamental tools to you. If you want more details, if you want me to make a much more technical video where I show you how all this has to be put together, tell me, leave it in the episode notes and I will be happy to do it. I know that these types of episodes, the more technical episodes, do not attract as many people, but I understand that since it is something a little more complex, it may be much more comfortable for me to explain it in a video and not in a podcast. So this is a little what I wanted to tell you. If you want, you know, a rating on Evox, on Apple Podcasts, on Spotify, it's fantastic for me to publicize this project. Remind you that this is a podcast from the usual suspects podcast network, the list of the best podcasts in Spanish, on all kinds of topics, all thanks to the reference lawyer who never remembers the title of the podcast. I'm damned, let's see if I can write it down for next time. And remind you that if you want to participate in the Telegram group, the usual suspects, Wintablet Info, you simply have to search for that, in t.me Wintablet Info bar. And if you want to participate in the Busy with Linux group, you simply have to search for that, Busy with Linux, and there you will find us talking about Hermes, artificial intelligence, containers, Linux, and anything related to Linux, excuse the redundancy, and Open Source. And little more to tell you, remind you that the day is two days, and one has already passed, so enjoy as if there were no tomorrow, and if possible, with Linux. And in this case, with Invidius and SearchNG, even better. Greetings and we'll hear from you next Monday, talking about cooking recipes. Hey, where are we going? See you later!