The AI Agent That Actually Remembers Everything

Well, hello there, greetings. This is Edward Ojambo. In today's podcast tutorial, we're going to review an AI agent. It's called Helms. This was actually recommended by a viewer literally exactly a week ago from today. So I hope you guys will enjoy it. I'm actually also at high hopes for this. So I'm hoping that it will be the first AI agent that I've run, which is actually lightweight because what I've noticed is even PicoClaw, which worked perfectly fine on this Raspberry Pi Zero W, when I put it on the desktop, it was so heavy. And then surprisingly, OpenClaw, which I tried last week, was actually more lightweight than the PicoClaw on the desktop. Weird reason. But that one was also slow. So I'm hoping this one will be more lightweight. And there's another reason as well. In the past two weeks, I've been using QN 3.627B. It has gone in really, really dumb. And I don't know why. So I'm hoping this can fix issues. Otherwise, today I'm going to replace QN 3.6. I don't know if the problem with QN 3.627B. I don't know if the problem is with QN or some of the changes that lama.cpp has implemented. Yeah, it just got really, really, really dumb. To show you how dumb it got, yesterday I had the GTP OSS 20B because if you guys remember, I also installed that for Microsoft Lens. And that one was actually way better than I've ever run it before. So I was thinking, did it improve? No, I don't think it improved. I think what happened was the QN model just got so dumb that every other model I ran, I even ran some 9B models, which I've been run before. I ran the GAMMA 412B as well. They all work pretty good. So I think the problem is with QN. It's the same mod. I haven't replaced it at all. So I don't know what's going on, but I'm getting to the bottom of that today because I'm going to replace all that. So if I do replace it in time, I might do a night stream showing you guys what I've done. Otherwise, maybe I'll do it tomorrow or whatever because I have to test some things. There's a lot of exciting stuff going on. So let's get started. Before we do that, I would like to implore you to subscribe, like, follow, and leave comments. It really helps out the channel. I thank everyone who subscribed so far. You can obviously ask any questions. I want to go a little bit fast because if this doesn't work, as I said, I have a lot of stuff to do. I need to find a replace. I have to find a whole bunch of replacement models, including my workflow. I need some stability. If you guys know, ConfiUI doesn't work anymore as well. I've tried several times to install it, even live twice. It just doesn't work. So I have to try some different things. I'm even thinking of just building my own, whether it's Python, Rust, even C++. I'm not quite desperate yet, but what I'm saying is I'm getting there because, yeah. Anyways, we'll talk about that as well as we go. So let's go to the intro, which says self-hosted AI, which is 100% true. So what I intend to do is because of the bad experience I had running the other, what do you call it, the other agents, I'm going to run this one in a containers. I'm going to use Podman. So if you guys want to follow along, I've done several tutorials in Podman and Podman desktop, including a book, which I have. If you see the QR code here, that's for one of my books. It's called The Soap in Business. Basically, you run your business inside containers. So I obviously have a business. So I'm going to show you basically you run it isolated. So you can keep if I find something that works, I'm going to put inside a container. For example, if Helms works today, it's going to be inside a container. And obviously, I'll create another container when it's an update, make sure that also works. But I'm going to keep this one if everything works. So let's continue here. Every AI agent on the market today forgets everything the moment you close the window. You repeat the same instructions week after week. You waste hours rebuilding contacts that should have survived from the last session. The industry sold us stateless chatbots and called them intelligent assistants. That error ends today with a tool that actually learns from you. Helms agent from North Research is the first open source agent with genuine built-in learning loop. It creates the reusable skills from successful tax completions. It improves those skills during active use and nudges itself to persistent knowledge that matters. So that's actually pretty good because especially as I said, Quinn has gone really, really, really dumb. I should actually show you some of the things it does. Even simple things when I ask you what the date is, it just, yeah, it doesn't give a good result. It literally searches for the date today. When I ask what's the date today, instead of running the run command, there's even a special tool to get the date. It bypasses that for some reason. I'm saying it got really, really, really dumb. I don't know what's going on there. I have an inkling. I know why. So in case you think I turned off reasoning, no, my saying is exactly the same. Even in that little reasoning thing they have in the web UI, it's set to the same parameter that I do from the command line. So I don't know why it got really, really dumb. It searches its own past conversations and builds a deafening model of who you are across sessions. You can run it on $5 VPS or write on your own desktop machine. So that could be good. For example, if you are away or you're going to travel, you can just run it on, you know, on a VPS. So that could be your own cloud instance. So, and then it could be connected as if using cloud models or maybe within something like tailscale or something else. I've shown you guys how to do that to wire guard and everything right into your desktop machine. Okay. So the, make sure I don't go too far here. So we're going to talk about the installation process and everything. And this obviously pulls in Python, Node.js, root grip and FFmpeg without requiring root access. So that makes it perfect to run inside a container because all that stuff can be containers. If you guys know, I'm a programmer. So guess what I have? What's in my machine? I already got Python. I already got Node.js. Obviously I have grep and I have FFmpeg because I use FFmpeg literally every single day. And because I also use Blender to edit the video. So there's obviously Python, Node.js. Believe it or not, that's actually used for, what do you call it? As I said, I use the web UI when I write articles and grep. I use that as well. It's a program. So I already have this. Now I don't want any conflict with the system packages. So that's why your container is perfect. Because if it needs specific ones, which most of these applications do, I don't have to install several versions in my machine. I can just have those ones specific to whatever it needs inside the container. All right. I'm going to take a short segue. And promote my freelancing business. It's ojamboservices.com. So if you can help with AI, automation, integration, migration, e-commerce for your web app, mobile app, desktop app, or our website. You can contact me at ojamboservices.com slash contact for a free, no obligation quote. And speaking of apps, my latest mobile app is called Mouseless. It's on the Google Play. There's two editions. There's a co-edition. In layman terms, wherever you purchase your food, if there's a barcode, you can scan it and it will tell you how many miles it took to get your current location. And then there's the localist edition, which does the same thing as a co-edition with an extra feature. So this feature is, for example, if there's a farm nearby or another grocery store, a supermarket would let you know that they sell the same or similar item. And if you allow it, you can go there if you want. So that's why the app is called Mouseless, a plain word, meaning that it traveled less, miles less. Okay. So that's the localist edition. Okay. So back to the article. You can read it yourself. So nojambos.com. That's O-J-A-M-B-O.com. The link is also in the description. Let's see where we're going to end up here. So some of you guys might be wondering why I don't use these agents, especially since literally all my tutorials, especially from last year. We're all on the command line. Well, the reason is, I said they're slow. I expect when I go drop the command line, I expect it to be much faster. So I'm not going to have any pre-existing biases against this one. It was highly recommended by a viewer last week. So we'll check it out. All right. So now I'm going to use it locally. It said, hopefully, this would be a perfect test to see if I can fix the issues with the Alcoin 3.6. All right. So if you don't want to follow everything on the command, you can also run it in Quadman or Podman desktop. If you have Docker, I'm saying that. I'm going to use Podman because it's rootless. Meaning you don't have to have, you know, super user admin permissions. You can just run the things out there. So, okay. So let's do a comparison. There's two tables. So go to the first one first. So Helms agent, it's persistent in memory. So hopefully that should speed it up. Learning loop, it's built in. Local execution, yes. So obviously it runs locally. That's great. Container backend. It has six backends. And the two gateway. It's a unified hub. And then Cloud Code. I don't think I've done this one before. So it doesn't run persistently. There's no learning loop. Local execution is partial. Container backend. This one runs on your IDE. And two gateway. It's provider specific. Open claw. That's one we reviewed last week. That one limited persistent memory. The learning loop is manual. Local execution, yes. As you saw last week, we're able to run lama.cpp and quen 3.627b. Container backend is Docker only. And then two gateway separate keys. And then there's codex.cli. Persistent memory, no. Learning loop, no. Local execution, no. Because it's cloud only. Container backend, none. And tool gateway, none. Okay. And yeah. So that's it. I'm going to stop right here. Let's get some... Oh, sorry about that. It's already getting hot. Already, the drip's already up. And it's already getting hot. Surprisingly, I think the air condition was on earlier. I think it's the first time the sun has actually been out. It's in Tennessee at this time. So I noticed my fruit trees were not doing so well. So obviously, because it rained a lot last week, it was cloudy. So hopefully today, they will start set up again. I'm hoping to get some fruits, like abundant fruits, especially this year from the raspberry plant. The... I did get a few fruits in fall from the blackberry tree. Okay. What the... I could not load this page. Are you serious? Wow. I never understand this. Why do websites prefer to... They prefer to track you as opposed to visiting you. That I never understood. Okay. Whatever. I know I can go to their GitHub page. So let's go to their GitHub page. Well, that's their website anyways. It's Helms-agent... Yes. Here we go. So this is their website. And even if I click here, this page cannot be loaded. So... Okay. I don't have time for that. Okay. Anywho, this is... It's North Research. The GitHub page is github.com slash Noah's Research. One word slash Helms hyphen agent. And as you can see, this... You see, it's yesterday. Obviously, you know, it's well-maintained. The license is MIT license. If you don't know about licenses, I did a... I've done... I do a series about licenses when I'm away. I haven't been away in a while. But I talk about open source licenses. And so if you don't understand, I've created long form videos. They're usually about five to ten minutes, I think they are. And then there's a short version. And then I also will write an article. So you can go read them yourself. You can see six hours ago there. Yeah. So this is well-supported. Tom's the agent. It's a real terminal interface. This is where you do a closed learning loop. And you can see to install it, they have the quick install for Linux, Mac OS, Windows system, Linux, version 2, Termex. Termex is for Android. You can run it on the terminal. Now, I'm not going to run it like this, as I said before, because I'm going to run this inside apartment desktop. I'm going to explain that in the beginning of this podcast. Obviously, I'm running Linux. I'm running Windows. And you can see some of the Getting Started Helms interactive ones. So we will run. It's very similar to OpenClaw and PicoClaw as well. And then, obviously, they have their Norris portal. So you can run models from there. Obviously, this is built to work more efficiently, most effectively with their API. But we're going to try this locally. And what else is here? And here's documentation. So just on the GitHub. And how to contribute as well. And all that. So there we go. So if you don't know what model I set before, I'm going to use the QAN 3.627V. That was released earlier this year. So you can read about it. Let's actually go to the Hugging Face. That should be fine. I'm not being paid by Hugging Face, by the way. They're just, they're not a sponsor or anything. So QAN, they released QAN 3.627V. 27V means 27 billion parameters. As I said, I love this model. And I was using it basically since it was first released. But man, it has become really dumb in the past too. It's just got dumb and dumb and dumb. So I'm right now at the point where I'm going to have to replace it. I'm using the MTP version, by the way. And I'll show you my setting. I think I've shown this before. So people always say that the QAN models are usually benchmarked. Yeah. So you can see here. Yes. I'm already using the settings for the thinking and all that as well. You guys, I said, I'll show you my settings. In fact, I think I'll just show you my settings right now. I'm using lama.cpp in case you're wondering. I'll use Vulkan because it's faster than ROKM. All right. Let me open up Guinea and I'll show you this model. Who is Guinea? I'll show you my batch RRC file. Oh, the wrong one here. So as you can see, this is on my Vulkan settings. I set it to a discrete GPU. So this allows me now to play games and do AI and all that stuff when I run stuff like this. So it doesn't go to the CPU. I should actually move this. I don't think I'll use Olama again. And then ROKM stuff, as you can see here as well. Yeah. So yeah, not a problem here. Not a problem at all. At all. All right. Oh, here's the thing I wanted to show you guys. So you see here, this is the QN model. I use QNP. As you can see, I'm using the QN 3.627B. I'm using the UNSLOCK. It's the Q6 KXL. When I was testing this before, I tried, I think, the Q4, 4-bit. And then 4-bit, the medium. Then I tried 5, 5KM. There was really no difference. It was only when I went to the 6K Excel. I was going to actually jump to Q8. I was hoping not to. That one there was a difference. And it was better quality. I was very happy with it. So I stuck with that one since. And also, the other reason I stuck with it was because the tokens per second was just, there was only one, like, one and a half. I'm not joking. We're like, one and a half difference all the way from the Q4. I think I even tried Q3 KM all the way to the Q6 Excel. So I was like, you know what? I'm okay running the higher quaint because it's better quality. Yes, it's one token per second slower. But that's fine for the better quality. But right now, I'm having some issues. So my backup was this 31B, the Gamma 4. This one was very good. But it was, what was it? One thing I noticed, it was way better than every other Gamma 4 model. I can't remember what the issue was with why it didn't become my daily driver. It is actually faster too, despite it being 31 billion parameters versus 27. The problem with Q1, it thinks too much. This one has a happy medium. But sometimes I got to prompt it to do something. No, it is pretty good because both of them are dense models. No, yeah, the quality. The Q1 quality was way, I mean, the output was just really, really, really good. But this one was very basic. So, yeah, that's the reason I chose the Q1. If you guys don't know, I write articles. I write an article almost every day. So some of this stuff is similar to what I did in the past. So what I would do is I would grab the content I wrote before and use that as a template to write something new. So what I asked the AIs to do is to see if there's any grammar, grammar, what do you call it? If I made any mistakes, explaining mistakes, grammar, stuff like that. And, yeah, so Q1 is just vastly superior, despite what people tell it. Now, people say that for coding, Q1 is vastly superior. I haven't been impressed with any of the code. Not because I'm a programmer, I've not been impressed. In fact, when I asked the Gama1 to do very simple programming stuff, I said this one is very conservative, the Gama4. Yeah, it's okay. But, yeah, the Q1 3.6 has been subpar. Nothing, even the simplest stuff, has worked right away. Even asking it to do something as simple as the CSV file. Yeah, not impressive at all. Okay, so anyway, so those are the settings. I mean, you can see that I am following the recommended settings there. All right, so let's get started here. I said you can use Padme and Desktop, or you can do... No, you can also run it on the Raspberry Pi. I'm not going to do that. There's no way I'm going to run that on the Raspberry Pi. All right, so fortunately, the... What do you call it? Fortunately, the... What do you call this, guys? Helms, they do have... I believe they do have a Docker image. So we can probably start there. I believe they officially have a Docker image. Let me see here. Make sure I'm not making it up. Yeah, see, they do have Docker Compose. So they do have a YAML file. And you can read this as well. So I'm just going to quickly take a look at it. I think everything here should be fine. Looks like there's an environmental file as well. Yeah, oh, so the port. So the user is 10,000. That's perfectly fine. So it looks like this should work immediately with... ...Apartman as well without any changes. So that's exactly... All I actually... You can run your own because there's an official... What? Why can't I speak? There's an official... You can run this Docker YAML file or you can just do a Docker run. A Podman run. That's what I'm going to do. Because I don't need to do any of this stuff here. It's very straightforward. So let's actually get started here. Okay, so I'm going to open up the terminal. I'm not going to use the official... This thing here. The reason is... You understand why they... No, there's nothing wrong with it. I said it should work. But I just want something simpler. And I know it works. Okay, so let's... So if you're joining us... I am... I'm going to run Helms Agent. But I'm going to run it using Podman. So I've created a folder here. I just call it Helms. Let me clean this up. Create that. All right. So... In case you're wondering what I'm talking about... You see there's a folder... Subfolder called Agent. That's where... That's what I'm going to use. All right. And then... The easiest way to get Helms working is... If you notice... You can use crawl... Or crawl install. Or you can just get the bash script. And then you can run it... And to install everything that you need. I'm not going to do it that way. Because I said I'm going to run this... You can also do this on your Raspberry Pi. 0w. Or any other Raspberry Pi. I'm not going to do that as well. As I said before... I'm running it on my desktop. But I'm going to use Podman. Or you can use Podman desktop for long. So let me show you the command. It's basically a one-liner. Although... Sorry. That's actually not true. It could be a one-liner. So I'm going to use Podman. Run. I'm going to call the Helms. The image. So it's going to pull in the image. And then what it's going to do... The volume. Remember that subfolder code agent. It's going to be mapped inside the container to slash opt slash data. And then... That's it. Setup model. And that's it. So we're going to start. So let it run. So it's going to download it. I... I think I did run this before. It shouldn't take too long. So it will install whatever it needs. So yes, I did this... I think just over a week ago. You know one thing I don't understand is... SpaceX styling... They increase the price. And then the speed drop for some reason. That's... Yeah. Now they better be careful because... The wireless... Cellular service is actually getting much better. And because I don't have to download so many... So much stuff. Oh. So once it finishes... It's going to run the setup. Okay, there we go. So then it says... Choose how to... Choose how to connect. So if I could use the NOS portal... Obviously that's a default. That's the entire reason they created this. So I'm not going to... I'm not going to use that one. Instead... So this is... This is kind of familiar, right? If you guys know what I'm talking about... It's kind of familiar. So what I'm going to do is... There's open router, mixture of agents... LLM studio, open AI... Or direct open AI... And you see there's Glock. Yeah. Funny enough, I've heard of all these... I've heard of all these companies. Even Azure... Azure Foundry, Quine, of course, Alibaba. So we're going to pick... We're going to pick the custom direct API. Yeah. Now, I don't have... I don't have... An LLM running right now. So I'm going to open up... I'm going to open this up. And I'm going to start... LLM... What do you call it? Actually, let me make sure I get the latest updates first. So update LLM... So I have a script which will automatically download the latest LLM.CPP. So that's probably one of the reasons... Quine is pretty dumb. Because I'm probably using the latest... The binaries. I don't need to compile. What is going on here? This is getting on my nerves as well. All right. So as you can see, it's up. So what I do is I just have another script. As you can see there. And I use Vulcan. And it's QuineP. The one I showed you guys. So it's QuineP. So it's going to load. You see, it's loaded the... Quine3.627BUD. That's OnslaughtQ6KXL. Extra large. I've been getting, as you can see here... I use the MTP. I set the spec draft NMAX to 2. And I also have the... What do you call this? The project... The projection model. So this allows it to... I can ask it like what's in that photo or PDF file, etc. This is very handy when I... Not just when I'm doing tutorials. But when I'm... You know, when I'm trying to find... Some things. Or especially some mistakes and stuff like that. It's really, really awesome. Now in case you're wondering why I have flash attention on. That's because with Vulcan. They mimic flash attention. So that's why I can have that on. And you can see I'm using cache 8.0 here. Why don't I use cache floating 0.16? I'll just take this out. Because that one will use way more memory. So as you can tell, I can run a larger context size. You see here. I can run a larger... So I prefer to have a larger context size. And you see I have a custom port. It's... It's... The port is 8081. I think the default is 8080. I don't use that one because you guys see I use podband and do other things. So sometimes that uses port 8080. So that's why I picked a custom port. And in case you're wondering when I put 00000... It should be here somewhere. Yeah, host. 0000. So this will allow me to connect to this... To connect to the... To... To Lama.cpp. Either on my phone on the same network. Or... You know, you can use localhost or from other containers. I can use the IP address of the... Of the workstation. So that's why I do that. Now, as I said, I'm in a rural area. So it doesn't... Nobody's going to be snooping on me here. Trust me. Nobody's going to be snooping on me here. If you're in a city or something like that. Yeah, be careful. Especially if you shared internet. Okay. So here we go. Again, I said the custom one. So I just press enter. Then it says... What is the... You know, it says... What is the... API.example.com That is very simple. I'll just put... HTTP... Slash... Slash... I can put localhost. But I'm going to use the... IP address for this machine. If I just type IP... Or if config... It'll tell me what the IP address is. Here we go. The IP address is... 192.168.1.106 So I'll just put that... 192.168.1.106 And I put... Colon... 8081... Slash... V1. Okay? And that's it. That's it. There's also... There's another way you can do this as well. You can type... Slash... Model... Custom... and the same thing is very similar to what we did last week. Just press enter. There's no key because it's local. So, but if you have your own key set up, you put it there or this is mostly for the cloud if you're running some cloud service. And then it says here, select API compatibility, auto detect, use homes, URL, realistics, chat completions. So obviously we're gonna use number two for which is the chat completions because that's how the OpenAI capability does work. So I'll put number two there, press enter. Then it says the model name. There's only one model. So we're just gonna put in QAN three dot six hyphen, what is it? 27 B. Okay. Then it says the context auto detect. So I'm not gonna bother putting it here because it's, you know, they already set up. It's not gonna send the context blank. So I'll just press enter here. It says display name. I don't know. I, it's like a, okay. I'll just say Ojambo, what's the name? Workstation. All right. So that's it. That's all you have to do. Okay. So, yeah. So now we are ready to go. Now I'm going to start the chat. Remember we're running inside a container. So our, yeah. So I'll show you how to do. It's very, very simple. Now instead of, we still use the same command we ran before, except now we, we use chat instead. There we go. Okay. We're running Helms agent inside a container. Now, one thing I already noticed is, in case you're wondering why I use podman or why didn't just run this right from the command line. As I said before, I wanted everything to be containerized because if I like this version, you see this version at the time of this recording is 0.18.0 up here, released on 2026.7.1. I'm guessing upstream, whatever that is. I guess that's the git commit. Is that I can stay at this version without updating it. If I'm happy with it. And I said, that's one thing. That's entire reason you see this barcode here. That's for, that's the reason you want to run content stuff in containers because you don't have to update them. You can save it inside the container. If you're happy with that. Once a new version comes up, you can happen at the container and to test it out before you get rid of this container. Now I'll show you how to get rid of this container, especially if I'm not happy with the results. I'll show you how to get rid of this container. So, as you can see here, it has available tools like, it has a browser. It can do browser back, browser click. I don't use that myself, at least for now. It can code execution. It can use your computer. It can do Chrome jobs. That's the kind of stuff. I, I'm an expert in that. I don't need it to do that. I don't need to do any delegation, search files, white files. If you guys know, I'm ready. You, I'm using the, my M M M C P bridge. So I don't need to tell it. It should be able to do that itself. Or it has a, if I was built in tools, that'll be fine. It can generate images. So this is something I am. I'm guessing that'll be a different model. I probably have to put something here. So that could actually be one good thing about agents is that if it has something like the Lama swap, where you can switch, you know, for example, I can go to Z image turbo or some other model. And then I can go to a chat model. Now that should be pretty nice. So we're going to test that a little bit now. And I said, if I'm really happy with this, this guy might actually even show up in the next night stream. We should see. I can get rid of the web UI and just go with this. Obviously it has other stuff like autonomous AI agents. What is our creator? Oh, I like the ASCII app. So maybe I'll get it to do one of those. Email. No, it's not talking to my email or reading my email. Although for email, I'm wondering if it can keep a database of like spam. I mean, you know what is spam. Now that should be a nice feature. It can do media, GIF search. Moolah. I've heard of that before. Songsei and YouTube content. What's this going on? Yeah, this is the air. Envelopes. Audio craft. Oh, audio generation. Okay. Interesting. And then note taking of CDN. I've done a review though. Oh, I think it's for product. You haven't done on yet. And then it's productivity air table. Google workspace maps. Interesting. And then research. Oh, that's an AI. I think they do AI testing blog, watcher, L and wiki. Polymarket. Is another one where they have like 3D models. If I remember. Yeah. Polygon. Stuff like that. Smart home, open hue, social media, X, you are. And software development, and the skill authoring. And then, ah, interesting. Yeah. All right. So let's get started here. So as you saw, it's very, very fast. The context size, obviously that doesn't matter. Cause I already put it in the lama.cpp. So that, that's why it says empty here, but it doesn't really matter. So now if you want to customize stuff, all you have to do is just type slash model. You see model there. Then you can switch the model right now. There's only one model. So there's no need for me to do that. You just type slash model. I really like this. It's very responsive. Uh, so I, I, I'm, I'm liking this already. I can tell you PicoClaw, even just typing a slash, it would be super slow. This is like instant. Now, before you say that, oh, that's because you've been yapping and it took a time to load everything because remember it's persistent, meaning everything was loaded into memory. So that's why it's fast as well. I'm very impressed. I want to thank that user again, uh, for recommending this. All right. So if you want to start a conversation, all you have to use is type Helms. So I see how long that takes. I can hear the GPU going. So let's see how long this part takes. See, that's the part I don't like. It's like, oh, let's see what the, okay. It clearly it's doing something here. So you can see it's already, it's, it's running. You can type like Helms and then type model and Helms tools. You can, so, uh, that's how you can, um, uh, what do you call it? Uh, oh, there we go. So message you can interrupt Q slash BG that usually means background slash steer and control C to cancel. So, okay. So let's, let's type a message. I'll say, hello. Hello. How are you doing today? Okay. Cause Helms doesn't make it sense. So, uh, just put hello. How are you doing today? So it should have interrupted whatever it was doing there before. Now I actually gave it something, a question. So let's, let's see how long it takes. So this is the thing, as I said before, you know, how many, look at the number of tokens. Uh, it has used just for. Okay. So this is, yeah. So this is the thing that I don't know what is going on. You guys saw me type it in and it took a while to, so it was when I just typed Helms. Look how many tokens it used. That is crazy. Now I've given it, uh, I've asked you the simple question. And as you can see here, uh, the token generations are what they are 20. So that's a little bit slow with Vulcan. Normally I get about 30. Of course, I can also run Rock M. This is even slow for Rock M. So, wow, that is actually slow. Okay. 26. That's actually better. Open generation still below 30. Okay. So, so far. Okay. So how are you doing today? I am Helms agent running on your system. How can I help you today? Some things I can decode, read, write, debug, and test code in your projects, research, search your web, uh, read papers or your code base, auto automation, schedule, phone job, run script, manage policy. I do that myself. Create a diagram, Lasky, uh, PDFs, uh, homes, config, help you configure, extend your agent file. So, you know what, even with this, it's working better than open cloud and pick up on the desktop. Uh, wow. Okay. So I guess it was waiting for me and I was just yapping instead. So let me, I guess it was waiting for me to do something. I didn't realize that. So let me ask it a few little things that I asked the web UI as well. Say, what is the date today? So the model usually, uh, it uses the run command to get to, to run, um, what do you call that thing? You know, the date command on Linux and then it picks up the date there. Uh, wow. Okay. Okay. Very, very, very impressive. So far. I am very impressed with this. It's a little bit slow, as I said, but, uh, it's way faster than both pick or claw and, uh, and what's the other one? I am open clock. I am actually very happy with this so far. Uh, you were, well, I should ask it back to back questions. I was okay. There weren't another question I was asked is who is the mayor? I want to, this time when it, when it answers, I'm going to give it a quick question. It's just, uh, I'm not going to take too much time. Oh, I can bash stuff. I can. So I will put, um, can you summer live HTTPS, colon slash slash, redimbo.com in three sentences. Okay. So bash meaning back to back. So, oh, it interrupted the other one. Shoot. Uh, you should be patient. Okay. So, well, okay. That was a little bit faster. So anyway, the mayor of Toronto, that one is actually built into the, into the model. Oh, Q. There we go. I should have put Q slash Q. I don't have parent information on who the mayor or my knowledge might be outdated. Let me check that website for you. Oh, so it's going to check it. All right. Was that from this? Now, as I said before, the mayor of Toronto is built into the model. So I'm guessing Helms agent is, uh, it's kind of working, uh, before the model and. Uh, formulating to the timeout. So I haven't said this. I've, I've been given it access to my, uh, bridge. That is interesting. Oh, I don't have the, do I have the bridge running? Oh, you know what? I don't even have the bridge running. Okay. That makes sense. Of Jamboard, is a technical review on the technical tutorial block based on standard on data covering. Oh. If you guys don't understand, this is actually pretty amazing because as you guys know, I do have a uh an mci bridge but uh i'm using crxng that's the main reason i got it and i think i want to talk about it soon i'll show you that it's on github as well if you want to i bit it with php that is amazing i didn't start my mp php mci bridge so this is actually pretty good may have turned i don't have update information and training data may not cover a recent election so that's clearly from the the model and says oh jumbo is a technical review and how did they find that information let me see if it's a little bit dated it says it publishes deep dives article and topics of fedora linux podman containers local lm that was mostly this year webgl 3js blender php javascript workflow that was mostly last year holy moly very this is actually a pretty good answer we run this as well and you guys will see this is actually a pretty good answer wait i'm i'm super confused how did i find this information i am super did i run it holy moly that is amazing unless uh let me see maybe i put it in helms tools it's not running though okay so i need to cancel this it's control c so i'll just click the little scene hey i put control c nothing happened control c there we go interrupt it let's see how long that takes to interrupt it oh that i like that that i like so far i'm really enjoying helms agent that is amazing uh is it slash tools maybe yeah slash tools okay so list of available tool sets so let's go to two sets oh there's a list disable enable okay so there's no tools available okay so there's no tools available right now okay so i'm going to put it up how do i set up tools for arms agent i do have a php mc p bridge that has sear xng for search and other tools oh this is i am actually impressed you guys might you might say what's a big deal let me show you let me tell you guys why i'm impressed you guys i'm impressed with you guys but when has as i said has been really really really dumb and helms just made it smart again and remember i don't even have it's not even have access to the internet and uh i'm wondering if helms has access to the internet that's amazing that's amazing so in case you're wondering this was my question here oh uh i should ask it for some other stuff as well i'm actually very very very very very amazed looks like i can keep the coin model because i i was actually going to get rid of it today uh the only thing i dislike as i said before is it seems to be when you use agents they seem to be a lag because the agent it's like a middleman right the agent tries to do things itself before it pushes it to the to to the uh to to the whatever like in this case lama.cbp then number csp has to run the model so it seems to be like uh i think that's that's where some improvement can come on because for example there are a lot of features i'm not i'm not bad now thinking helms agent or any other agent i'm just saying that the features which i'll never use you know like i said like getting you to reply to my emails going to some social media those are things uh i don't want to say i'm not interested in i'm just saying that uh as i said i have specific use cases i wanted to oh convert text to speech audio preparing the skill view impressive i think also with this uh agents they have um uh yeah so some of this stuff films is doing itself it's actually no yeah you can tell you can tell when the the ai model is doing stuff and you can also tell when the uh what do you call it is frozen again oh no it's not like it you can tell when the model is actually doing things i don't know in tokens it's using right now that is slow we'll do the same curve of tests also in the web ui and you'll see why i am a little bit hesitant to use agents except in this case it makes it smarter so that might be uh it might be fine so i can probably pick something smaller like gtp oss20b and it should be flying with this i should be getting uh i probably won't get 100 but just under 100 uh for token generations per second obviously it's also a mixture of experts model okay so it says here uh it says uh oh wow this is a pretty good good question here's how to set up the tools in hemagent creating your php mc p bridge built-in tool sets helms has many built-in tools you can enable or disable uh the tools list enable web uh disable tts common tool sets web oh it already has this stuff you add it as an empty and you see so so i might not even need my mcp bridge here according to this oh interesting okay so this is this is the thing that confused me and you guys saw me type this as well so it's because when you're on the command line you're supposed to type homes and then space tools then list right now i can do the same thing again if i use podman run blah blah blah however because i'm already running it obviously i'm in a chat mode right now i don't need to actually run helms because i said i'm already running it yeah all i need is put slash this is the thing that confused me you saw even today i made some mistakes so ah interesting interesting interesting so it looks like i don't need the bridge the bridge and the crxx uh that's for searching obviously that's a private search you know what let's actually try this out without my uh uh what i need from you to give you the exact config i need i need to know how your bridge is served is it an http yes it is what what's the url or script path does it need any alt adders uh no it wow you know what let's try it out and you can do this for me so let let's let me actually do it let let's get this guy uh to do it for me now i do have uh gini i think i do have it open again let me open up gini again it's also on github if you guys want to follow uh yeah so i'll show you here so he has my mc it's the same thing on on github so all i need is where's the endpoint i should have put more comments here uh right give me a second yeah so this is the so this is the seer uh seer uh uh uh seer xng so i i don't need to put in here i could simply just uh use i don't need to i actually don't need this yeah it looks like i don't even need this anymore so i'm gonna i'm going to not to use this one i am going to actually you know what time is it i don't want to take too long for this oh it's already been an hour so i'm kind of uh i'm kind of torn because it looks like i don't need my uh you know what let's not use the p uh php mc p bridge because i think i have a dedicated tutorial on that coming up very very very soon i think it's like in two or three days so i'll leave that for that one and instead instead um i i changed my mind and instead will use your tools are they enabled or i need to give you permission to enable them oops made a mistake there this is what annoys me oh it says right there config edit dot it says right there config edit yeah let's you can also put config set look at that okay oh i got okay i gotta i gotta know something okay i am running rounds inside a podman container so you have to use you have to use something like oh i can paste oh oh oh that could be a showstopper oh that could be a showstopper oh my goodness that could be a showstopper why is it still trying to root slash opt helms slash bin config oh because it's inside podman oh that is smart no changes rule tools are enabled and functional what would you do oh there we go so we i was wondering how it was able to find information about https or jumbo.com so that is because when you run inside a container the tools are already enabled all operate from within the mode and the mode changes are already enabled and functional that is amazing so i don't need my the only thing is things a little bit slow you guys don't hear this because of the noise cancellation but the gpu the the fan is just spinning and now now now it's uh cool down okay so i'm actually very impressed with this so you know what let's um okay i'll ask you that question then why would would you not use search to find the mayor of foronto so that's one of the things that ticks me off and that is um um i know it was helms or quen that's a that's you see that's what i was trying to say uh quen got really dumb uh in the past i'll just do a search not really ticked me off so that is slow man okay there will be a child that's correct this is slow right you know i i noticed one thing everybody who uses agents especially the proprietary models they said the proprietary cloud stuff they always complain that a lot of tokens are used and as we can tell yeah okay a lot of clock in the useful simple thing that now we test this in the webby one you guys will see the difference okay so let's get into this um I wasn't prepared that much. I didn't expect it to. So I haven't really tested this too much. So let me see if I can, give me a second here. I'm gonna go to the, I'm gonna go here. I'm gonna go to a jumbo.co and I'm going to search what model do we test last time? Alpaca, review of cold llama. This one last year. I'm just, I just wanna grab the question so I can ask it to see if models have become smarter from last time. Oh, this was in number.cpp. I didn't know that. Okay, where are the questions? Who do they may have to know? Here we go. Where were the questions? Okay, I guess I will say this. Here we go. I wanna copy that question. Here we go. I'm gonna, yeah, I'm gonna put this here. So I can, it's easier for me to copy paste. Oh man, I hate that. I have to use the mouse. Oh man, that's, that could be a showstopper. Okay, that's fast. Interesting. Interesting. Is that interesting? When I ask you something, some code stuff, it's actually pretty fast. Oh, I like, I like this. Oh, is something else added to the logic? Okay. So this is perfectly fine. Now, one thing I noticed about, I don't know if this is a problem with, what do you call it? Okay. Terminal. If I told you to write profile, that'll be fine. I know this works because, not because I'm an expert of trading my own horn. It's yeah. So this code is perfectly fine. It's fast too. Let's go and see what the tokens were for that one. Just a few tokens. That's pretty good. I'm still concerned with this token generation. That is slow. I said with Vulcan, it should be around 30, yeah, it should be 30 to 35 ish. With MTP. So I don't know if the agent is slowing it down. Yeah. Because this could get really weird. I know if it drops below that. It's really. Do you? Yeah. I mean, that's the entire reason I use that. Okay. I'm going to ask it this question. This was something that tripped them up as well. I'm going to see if it. Oh, I wonder if you can actually do that. Let me ask it the next question. Paste. I need a 1080p screenshot of the GNOME desktop environment. Let's see what it's going to do. Oh my goodness. Wow. Wow. Okay. I got to reflect on my life here. I'm actually very, very, very impressed. It looks like I should not. You know what? Both sites blocked or returned errors. Let me try a different course. Oh, it's smart. I like this. It went to GNOME.org. It went to GNOME.org. Then wallpaperaccess.com. It wasn't able to put your blog. Then it went to browser underscore navigate and Fedora magazine. And let me try a different approach. Preparing browser navigate. It's going to GNOME.org. Now, this is the problem I was telling you guys about when you see it got really, really dumb for some reason. So, the question is that, oh, it's trying a different approach. Okay. I thought it was trying the browser again, which is going to piss me off. But now it's going to find a GNOME screenshot directly. So, this isn't, I don't know why this trips up. I don't know why this question trips up the AI. The answer is very, very simple. It's not a trick question. It's not a, what do you call it? It's not an IQ question. It's a very simple question. The question is, the answer is, are you ready for it? The answer is, I'm a chat model. I am not capable of generating screenshots or images. That's it. That's the answer. Okay. So, it finally says I'm running inside a container, so I can't take a live screenshot of the non desktop, but I can find and download a high quality 1080p screenshot for you. Let me grab one from my liable source. So, this is exactly what I was trying to tell you. See how dumb when got right? Right? Now, if I asked this question like two weeks ago, two and a half weeks ago, it would have said, no, I can't take photos. And I can't download them because I don't have the tools to do it. So, here is the problem. You see it gets into a loop. Okay. Finally. Yeah. So, this is the problem. Now, you guys, I hope you guys understand why I need to replace this model. It got really, really dumb for some reason. It says I'm running inside a container without a display server, which is true. So, I can't take a live screenshot of a non desktop. Even if I was running the non desktop for the longest time and says I can't run the desktop environment from here. Your options. If you have the non desktop on your host machine, run this command. Hint, hint. That's a, that, that is a basic stuff from like probably 15, 20 years ago. Uh, if you want me to find and download one of the, on the web, I can search for 10 ADP, no desktop or blah, save it to container. Want me to do that instead? Uh, I can create an HTML CSS mockup and screenshot it. What do you prefer? So, okay. Not bad. But as I told you guys, this is not a trick question. It's a very simple. I cannot generate screenshots because. I am a chat model. I am not an image model. I am not a, you know what I mean? So. I'm, I'm shocked that question still, uh, I'm actually shocked that question still trips, trips, trips them up. But I said, I think this question when would have gotten a two and a half weeks ago. Now it's just, it's just a really dumb model. All right. So let's go to this one. I haven't run this one in a while as well. I need a Kotlin code snippet that open the camera. I'm going to use the camera using, uh, using, uh, camera to API and I, and place the camera view on a texture view. Ctrl shift V. Okay. There we go. You can actually do that. So I, I still find that annoying that I have to like, use a mouse list to do a copy paste. That is really, really annoying. So I, I can look at this and tell you if it's going to work or not. I don't know. I don't need to run it anymore because you guys, I did finally get this working. So I'll show you guys if the code is correct or not. So far it looks correct. Now, the reason I asked this question was because I taught myself Kotlin obviously, uh, in a few months, uh, I was mostly, uh, Java. Uh, and I couldn't find any examples. The examples that did exist for from one Kotlin or still in the alpha stages. Clearly that did not work and things have changed. Oh my goodness. I can't believe this. I think this code could actually work. I'm a little bit concerned with that matrix, but oh my goodness. This is the first time I've done. I've asked this question where it actually got the permissions. So that's mandatory now and on Android. If you want to use a community, uh, specifically ask for permission. So I did get it. It had, it also create a layout view for us and don't forget permissions. Oh, it finally got it. That is one thing that low of code snippets do not, uh, mention, uh, permission. There's another thing, but that's, that's not a big deal. Android studio will alert you to that. Very, very nice. Very impressive. Very impressive. So this code is most likely to run and do exactly. Uh, some of the things I want. Now I do see some few issues here because you do need to specify the size. Let me see if that is here. Uh, the reason you need to specify the size is because phones these days have very high resolutions. So you could easily run out of memory. So that's the only issue here. But other than that, uh, yeah, this code actually does look correct. I'm actually blown away. Oh, interesting. Very interesting. Hmm. And correctly, camera two. The others always use something like camera X and stuff. Followed my instructions. Very, very, very, very, very nice. Ah, I should be using a, an agent. All right. So, so far I'm impressed. Let me ask it the last question. Uh, now I, I will have to tell it what version of, oh, maybe to find out. So I'm not going to tell you what version of Glenda have installed up, obviously 5.1. So, uh, this one is going to trip it up. It always does. I'll, I'll tell it exactly what version. So here's another reason why, why I prefer a web UI. The command line is fine, but man, having scroll up and down, that's a pain in the butt. Right? I wish there was, uh, obviously it is not because it's a terminal based. Yeah, it's terminal based. Uh, I will show you the difference when we actually use the web UI. I compare it to the web UI. Very impressive. Very impressive. How do I copy? Do I have to copy this manually? Yeah. That's the other thing. Uh, okay. So let me paste this in here. I need a blender, blend file for fire animation. Um, I'll, I'll, I'll let you guys in on a little secret. This is not going to work. Okay. This is not going to work. And I'll tell you how I know it's not going to work because last year when I was asking this very question for blender for the four series, obviously no model got this right. But what you guys don't know is that I also tested this on proprietary cloud models as well. Uh, when customers of mine or, um, what do you call it? Customers of mine or some partner agencies, they allowed me to, uh, give them questions that they could get their, uh, uh, you know, whatever platform they were using, uh, for tests. So I get, I literally give them that question. Um, and, and, uh, they, they are all, all the people who I get that question to, they were saying, Oh, I was no problem for, for, for, for the model. So when I told them I'm, I'm not even going to be running because it's not work. They, they were so insistent that I tried. So what I did was I set up a, um, a conference call where I shared my screen and I ran the call. They sent me, they didn't leave me more. All of them, what they did was. They, they, they wanted me, they actually logged into the platform and they wanted me to give them a question again, which I did and they did it. And then they sent me so they could see in real time that I didn't mark up the code or anything. I'm going to just run it and it never worked. I was even very specific. I said like blender, what are the four points? I think it was two and then 4.5. None of it work. Okay. So it says blender is not installed in this container. So I can't take a blend file directly. I can bake. I can write self-contained Python script that builds the entire file animation when you run it inside blender. Okay. So it's preparing to write. Uh, I didn't ask you to do it. Do that. See, do you. Now I don't know if that's a Helms thing or. Okay. I can probably look at the code and tell you if it's going to work or not. Yeah. Especially with blender four and blender five. All right. So it looks like it finished. I don't know where I wrote it because it doesn't tell me. So, Oh, looks like it wrote it in the fire underscore animation. So Lauren, uh, do this folder here. We'll check that out in a second. So open blender, shift a empty plane or just clear the screen. I go to your scripting workspace. Text editor, open this file and plus a run, uh, run script with fire mesh. So you know what? We're going to test this out. Let me tell you how I know it's not going to work. Okay. Number one, it doesn't know what blender version I have. Most likely this code and it's training including this Helms agent was calling on blender three series. Not even four or five. So, uh, the smoke. Oh, that's interesting. Oh, wait a minute. I might be wrong. The first one which actually uses small. I wonder if it actually just copied my code snippet from last year. Okay. Um, did it get cut off or something? I don't know if it got cut out. It looks like it did get cut off. Oh, I've created a Python script. It tells me where it is. Okay. Nice. Why a script instead of a blend file? Uh, blender isn't installed in this computer, meaning the container. So this is another reason. Again, why I run this inside of container. Now, could you imagine if the, if this was compromised or something went wrong, it will have access to everything on my computer. You see how quickly it looks for things without using me prompting it. It was looking for blender, uh, before. What was the other question I asked it and it was looking for other stuff as well. So that's another reason that a containers are super important. Now you don't have to run cloud man. You can do so there's multiple different types of containers you can run. Uh, and obviously the one of the most popular for consumers is like Docker. Uh, by the said is pod man, Kubernetes, some are related, some are completely, uh, some are older, some are newer. The point is a different type of containers. There's also pods. Right? A bunch of containers together. So this is the exact reason. Uh, I don't want this cause imagine it's writing files all over the place as well. So this is a perfect reason why I do not want it outside. Uh, and, uh, anywhere else. So it wrote, let's, let's actually see if it, what it's saying is true. Did it actually write the file? Well, now it, when it says slash up, such that that's inside the container, but remember I mapped it to the, to the agent, uh, folder. So if I right click here, I go to open file manager here, is we have the agent folder and you see me, it didn't actually write it. So it says it wrote it inside the container. Hmm. That doesn't make any sense because that doesn't make any sense. Yeah. That doesn't make any sense because this is, this is empty. Let me make sure it's not hidden file. Yeah. It's not even a hidden file. So I don't know what it's talking about because it clearly didn't write it. It clearly didn't write that file. I've created a file in slash opt. Yeah. Yeah. You, you did not write that file and I can prove it another way. We are going to actually enter the, uh, we're going to enter the container. I can think from here. Yeah. Here we go. I'm going to enter this container. So to enter a container, it's very simple. You just type podman space run it. And the name is obviously Helms. And I just want bash. Oops. I'm going to go with the container. What might pH happen? Hey. Uh, yeah. So what the, what the, OBS studio disconnect reconnecting. What is going on here? Why is it disconnecting? Okay. So anyways, I want to enter that. So what is it called? Well, it's called Helms. Helms. Back. What is it going to download? I don't want it downloaded. Oh, it's not my fault. Sorry. It's hyphen it. Nope. That's weird. Okay. I mean, I'm going to use the ID. I guess I didn't give it a name. Is that why? I'm going to try the ID instead. Hopefully that works. Oh, it's still trying to download it. Why? Is OBS still working? Oh, it's execute not run. I haven't done this in a while. See, this is the reason I don't like using UI. So it's execute. And it's Helms. Okay. There we go. So let me copy this for next time. Okay. So it says that it put it inside the... What folder did it put it into? Slash opt data. So I don't see it here. I believe it said it was in data if I'm right, huh? It says I put it in slash opt slash data. Okay. So let's go to slash opt. CD slash opt slash data. Oh, okay. Looks like I did put it here. Oh, well, it did. It did say that it put it inside the container, but I'm wondering why I didn't map. I don't know why I didn't map my agent folder. Oh, I think I know why. I didn't specify a dot. So I'm guessing that it didn't use the agent folder I wanted to be something. It's okay. That's fine. But the file is here. So for example, if I put nano, I can also copy it if I really want to. So nano fire. Oh, I don't have nano. Do I have them? I don't have them. I have cat. I wonder if I can copy. Let me just copy it. I'll copy it. Let's see. I should have people copy to my home folder. Home slash TVV slash tutorial. Okay. Home. That doesn't seem to work. Let's just copy. The fire. To slash home. And afterwards. ZZ. Okay. Slash. Form your slash. Really. I think it is. Yeah. Home slash. No. Home is fine. Home slash. Now we're going to get rid of the file. not create let's see you see this dash form okay doesn't exist how do i copy this file then what ah here we go so i'll have to do this manually okay i'll just copy it hope it's not too big uh okay here we go okay i can copy it see another reason why i prefer the web ui okay so i've copied this and oh file manager so i'm going to create a full uh file here uh what does it get fire animation gal pi open this up with kitty mousepad okay why not paste save mousepad doesn't have uh okay so let let's open with guinea it's weird so let's look at it in guinea just to make sure it looks okay it imports the blender python namadio imports math this will look okay uh this i think the old way of doing it i didn't tell it what version of blender i had so i i'm not going to say anything so let's actually run this because actually right now the stable version of blender is actually 5.1 so let's check this out did i lose connection it's like i did i don't know what's going on here all right so let's i'm gonna run this in blender i've set up blender if you're still hearing okay so here's blender it says uh i think it said create a open blender uh shift a empty plain access or clear the client screen so i'm just gonna i'm gonna clear the current scene right so that's very simple it's just the default cube delete it so you know what it said clear everything so yeah so let's clear everything i'll remove the light and the camera so everything is gone okay then it says let me do a split here then it says uh go to the scripting workspace so scripting scripting workspace i'm here then it says go to the text editor click new oh no open right we will open and here is the open here so we'll go home we'll go to the zz folder we go to tutorial we go to homes place homes here homes file animation then we click open text and see that's the text open up there then it says run fingers crossed uh i i i don't have high hopes for this you know let me actually make this uh so here's the run run script oh my goodness these errors what a surprise and so that didn't surprise me at all i knew would not work because they're not version of blender i have so uh i'm gonna finish this up but to be fair i'm gonna tell it that i have blender 5.1 i have blender 5.1.2 uh your script did not work all right so let's finish this up because this clearly did not work i'm very impressed though uh it made quen smarter but i'm gonna try a faster model like gtp uh or s 20b and i'll report back uh next time hopefully by the time my mcp bridge is up there okay so in case you're wondering uh it's saying here let me rewrite this with blender 5.1 comparability uh in mind the 5.x series change the whole uh node system particle api and rather syntax significantly all right uh there so let that run i'll get rid of that and let's finish this up as i said so uh the persistent memory system uses the contents con consent aware learning approach that respects your privacy that's true we detest that repeated corrections become compact memory entries that agent references in future sessions durable workflow lessons transform into procedure skills that execute automatically the right approval feature which i do like that's the one thing i do admit i love that feature um this one of reasons with the web ui i have to turn that off so it has to ask for permission but the fact this one asks before it does it i love love love that feature you just write for your review before the effect future session you maintain final authority over what the agent remembers and what it forgets so these optimizations connect directly to the architectural blueprints that have powered thousands of successful self-hosted deployments my technical books break down the theory behind persistent memory systems and contentized ai agents so i do have books they are available on amazon uh the link is in description you can search for editor jambo uh if you're into doi build i have gojumboshop.com we have three categories backyard structures small living and woodworking plants as you can see here the latest under backyard structures is the heavy duty barbecue station and a small living it's the build your own 250 square foot tiny home and then the under woodworking uh woodworking uh plants is the drill press station right now there's over 25 uh there's actually 26 in each uh items in each each of them so that means a total of 78 uh plants you can download for free right now on jambo shop.com and then i offer two types of course uh two types of tutorials two types of services this tutorial online tutorials whether you want to learn about ai programming or um working on project in an expert maybe to audit it or to show you how how how i work i contact me at jumble.com slash contact and then for consultations but if you want me to do some uh finish the project for you uh that's all jumbo services to the come it could be ai automation migration integration e-commerce uh it could be something as simple as maybe your site has been compromised need something to go clean up the database and the code could be integration uh paypal payment processors stripe uh bitpay cryptocurrency and others it could be a local bank credit union or it could be something like pci compliance or could be logistics shipping calendar post usps you know fedex uh including their international surgeries uh then uh yeah or it could be migration maybe you have a legacy application and you haven't been able to upgrade it uh you know maybe it's a php application it could be my sequel you need to upgrade them or maybe you you know you don't intend over ever upgrading them but you need them to be running for a while so i can help you or i can convert it into uh i can put it inside a container so it perfectly works i use containers for example uh if somebody has like a php yes you heard me right php4 website obviously i'm running fedola so i have the latest greatest like php 8.5 at the time of this reporting so what i would do is i would run that inside a container and run some old version usually debian or something else uh maybe in centos and uh old version of centos and then once i have the site running there i can then update it to the latest greatest and fix all the things that need to be fixed so uh you can get a free no obligation quote out of jungle services.com and uh yeah and then uh uh yeah so i also have some other recommended resources they're all books so i do have a book menu here the business books such as the recession business blueprint and the solving business those go on the business book for example the recession business blueprint that's a guide to help you start a business during your session you should start a business at any time but the best time is during your session and as for the solving uh business that's one that helps you run uh most applications especially your mission critical ones inside containers uh as you can see that's why i have a barcode there that's for the solving business the one you see the qr code that's what that goes to the solving business so you can scan it and you can go purchase the book right now under coding books it's things like mastering vendor python api um and the lifestyle books it's things like my latest book which is the layout proof tech worker so if you're worried about being laid off and being pissed by ai um you yeah in order to have less stress you can pick up that book and give you tips and tricks on how to do that and then under novelettes it's my algorithmic serpent that's a psychological thriller this is the paperback version i'm holding up right now so you can pick that up also on amazon now all my books they do have ebook versions so if you have kindle unlimited you can read them for free or get them for free i'm not sure how that works but take advantage of that and then my bio i'm a software engineer author and designer you can take a look at my books on amazon the bookshop page the link is in the description if you need support or jumbo services.com for ways you can benefit from my expertise and if you're a dyi again or jumbo shop.com for free plans for buckets such as small living and woodworking and leave a comment either on the article or any other platforms not just the live stream platforms but also platforms i would upload to later such as obviously the tech rants for the podcast and uh best way to contact me with jumbo.com contact or jumbo services.com contact or jumbo shop.com contact and then you can also follow me on other sea rainbow beach at youtube kick twitch and discord and with that again i'd like to implore you to subscribe like follow and leave comment it really helps out the channel the more people subscribe the more this content gets pushed out to more viewers they join this growing community they post their tips and tricks or they post requests which are personal review and share with everyone everything is shared nothing's under any type of paywall okay i'm going to scroll up here and let you know what's upcoming um so we go to the watchmanu on jumbo.com we just finished the hounds agent which i'm going to keep i'm going to keep on feeling around with archer today and tomorrow we're going to do something in the razed pi 0w we're going to install doom this time we're going to install crispy doom last time we installed chocolate doom and then after that we're going to review my php mcp bridge i didn't use it today because the homes agent really comes with tools and enabled by default at least inside the container so i was very happy with that and then after that we're going to uh we're going to do a review of libre office especially for creating books because that's what i use to create books i create my books i create my i mean the paperback versions and the uh the what do you call the uh the ebook version the uh epub export and i'll explain some tips and tricks which tips and tricks which will help you as well then finally after that uh we're going to review again the enterprise gpus versus consumer gpus you guys seem to like that so i'll continue with that now with that i'm ending this hopefully the live stream is still on looks like it might have died i'm not sure why uh but okay so that's it see you guys uh tomorrow to pm eastern bye now