LIVE: Google Gemma 4 + OpenClaw!

Key Takeaways

Today we're going to be testing out Gemma 4 bite forbite. The most capable open model available right now. This just dropped from Google and you can actually run it with other models. For example, if you download it locally, you can use this with cl code. You can use it with all your AI agents and it's completely free to get access to. Supposed to be designed for agentic workflow. So it can call tools. It can be used. It can speak 400, sorry, 140 languages and Gemma 4 is available to get right now. So, this is an exciting one. Last time I tested out the previous version, Gemma 3, super powerful. It was really, really good. And so, we're going to test out Gemma 4 today and see how it performs and we're going to get straight into this. So, if you've never used this before, one of the easiest ways to get started with this is you can download from Olama. So, if you go to Olama here, you can run this inside your terminal. This will download OAMA and then you're going to go to models and then you can go to Gemma 4 and from here you can pull in the model from terminal using this command. So if you copy and paste that into terminal and then you go inside your terminal once you've installed and downloaded that then you can run some of these models right. So for example you could run claude code with Gemma 4. I'm not sure how good it would actually be inside claude code because claude code is obviously designed for claude as an API but you get the point. Like you can run this with multiple models. You could run this with open claw, right? And it' be pretty simple and easy to do. And you can see the different model sizes here, right? So you've got for example Gemma for E2B that's 7.2. Then you got the biggest model which is 20 GB, right? The one that it will actually download by default if you run this command inside OMA is this model here which is 9.6 GB. And if you've not tested out, let's start running it directly. I'll show you exactly how to get started with this. So let's say for example you want to run with Gemma 4. You just go inside your terminal here and then you're going to copy and paste a command, right? So you see that command from OAMA. You go into terminal. You paste that in. Boom. Now once you've done that, that's going to pull in the model. So you can see it's now pulling in the model. Now it's already downloaded it for me. If you've never used this before, if you've never downloaded it, the first time I did that, it took about, you know, maybe 10 20 minutes to download the full 10 gigabyte model. All right. And then once you've done that, it's good to go. So, for example, if we say, "Hey, are you working?" You can see here that it's beginning to respond. It's pretty fast. It's pretty fast for a local model. Honestly, I've tested out a lot of models. That is one of the fastest I've seen running. Now the other thing to note here is like this is really designed for mobile devices if that makes sense. So what Google are basically planning to do is they'll get this running on mobile. They get this running on you know your your iPad for example and you'll have like these local models running on your local devices. That's why it's designed to be so small. It's not designed to be like the best AR model of all time or the best API of all time. The whole point of this is like they're creating like these miniature APIs, these miniature models that are free to run, can run and be powered by your device and then run with whatever you decide to plug it into, right? So, for example, here we we could actually test like building out, you know, a website or something like that. So, if we go to the airprofit boardroom.com and we take this information from the page and then we go into terminal, right, we've got Gemma 4 running here. So now we can say okay code in HTML [clears throat] a page for the AR profit boarding and then we can paste in the information from our website here. Hit enter and you can see it's beginning to reply now right super fast. Now if you're wondering okay what's my setup like why is it running so fast? So I'm on a Apple M4 Max on a Mac Studio as you can see here. So that probably helps to be honest. It's a pretty good setup, but I know this can run like locally on pretty much any device. That's what Google have designed it for. Right now, if you're wondering, okay, like how do the benchmarks perform? So, you can see some examples here. Now, this is only going to be compared against like smaller models, right? It's not they're not going to compare it against like for example Claude Opus 4.6. And that's why the benchmarks are a little bit different, right? But you can see like if you compare this in total to total motor size versus ELO score, right? You can see it's pretty much on par with Quen 3.5 for the Gemma 4 thinking models. And then GLM 5, which just recently got announced. A lot of people loving Kim K 2.5. It's not far off right on the benchmarks here. Now, whether you trust the benchmarks or not, that's up to you. But you can see that it's working. It's It's moving pretty quickly here. It's actually responding with code. Honestly, I've tried stuff like Neatron Cascade 2 that just came out recently, which was a much bigger model, and that was probably worse than this. And I just know from the response from looking at the code, I can kind of get a feel for it that this output is going to be better. But we'll test it out and see what the actual code is like in a second. But that's basically how you can use it, how you can set up, how you can get it working locally for free, and how you can actually use it to build stuff, plus what it's being used for, plus the benchmarks of it as well. Right now this is a very very interesting one. This is one of the most interesting points about Gemma 4. So what they've said is this has industryleading capabilities of mobile first AI. Now the whole point of this is like Gemma 4 out competes models 20 times its size. 20 times it size. Right? So this is a pretty small model uh like 10 GB but you can see here that it's outperforming stuff 10 times it size. So uh sorry 20 times it size. Now you can see the actual HTML is ready to go here. So we could test that out in a sec. We've got it back from the old Gemma 4 inside terminal here. Bear in mind like the reason that you would install this on a laptop is like let's say for example you're on a flight something like that and you want to keep working, you want to keep getting stuff done. you want to be productive, but you don't have you you don't have internet on the on the phone, right? Well, what you can do is you can run this local model and still get help. Now, it's not going to do like crazy tasks. It's not going to be the best in the world at coding, but it'll be comparable to like some older models of like Claude, for example. You can see that it's coded out the page here, and it it's it's not amazing, right? It's not as good as Claude Opus 4.6, which I coded with before, but we've done this for free and we've done it locally, right? And you can see the page is is fully coded out directly here. Right? So if we go through this page, this is the HTML that we got back. Again, not like the best. Just going to be 100% honest with you. And I don't care whether it's good or bad. I just want to show you the truth. But you can see here like it has coded out locally for free. And we've just got that that running directly. Um and and it works, right? It actually works. So that's pretty cool. Now there are different there are four like different sizes of this. So you got 2B, 4B, mixture of experts and 31B as well. Let's see what we got in the questions here. Working at 4:00 a.m. I think we got different time zones around the world, mate. It might be 4:00 a.m. where you are. It's not where I am. And then what else we got here? Ashik says, "Good morning." Good morning to you to sir. Can you do probably won't do that and then S says cool yes cool all right so let's get back to this so if we if you have any questions just post them in the chat I'll be happy to help you so if you go inside terminal now we've used it for code it's done the job it's pretty fast to reply um and you can use it offline locally right now if you wanted to run this for example and I will do another video on this in a sec right in terms of like how this performs versus something like uh you know open claw and that sort of thing. But if we wanted to run this inside for example clawed code we can just run this model here right so we copy and paste our command and then we can start using gema 4 with claw code. So if we go inside here and we say okay you working it looks and it feels like clock code but it's running with Gemma 4 but it doesn't seem to reply to us. So it it may be set up, but again like Gemma 4 is not that great at gentic tool use. So that's probably why it's not replying, right? You can see it's stuck at mulling. I just want to be 100 uh 100% transparent with you. Like it's it's not working with uh Oh, it's starting to think now. Oh, there we go. All right, it's starting to work now. Can you create an SEO calculator for me in HTML? Plus open it up. Let's see if it can do that. You can actually change the effort levels as well, which is cool. Yeah, this using Gemma for it did take a little while to respond there, which is interesting, but in the end it did reply. So, let's see how it does on the SEO calculator. If you can actually use claw code with it. It seems to be slow, but it's replying now. So, you can see like there's a little gap. Probably takes about 30 seconds to reply. And then it says, see, this is interesting when you're using local models. I find this all the time. It's like they struggle with using the tools inside Claude. And now Claude is like a harness. It's an agent harness designed for Claude. And that's probably why it doesn't work so well with local models. You can see here, for example, it's like I can create the HTML file. Um but since I can't directly open files on your local machine, I'll create the file and provide you with the complete code which you can say right and it's it's I'm creating the the calculator now but it's not because it's finished replying and it's it's not doing anything right now. So you can see how inside claw code this would struggle. I wouldn't use this with claw code. You could try and you could probably get it working eventually but I just don't think it's that great. [snorts] Um, it does say cooked for 1 minute 20, but I haven't got any information back. Wouldn't recommend that. All right, let's try the next test, which is open claw. Let's just see if it works. So, we're going to run this on open claw. Now, bear in mind, like if this works, what you could actually do is you could get gemma 4 working with open claw on your phone, for example. So, let's see how we do. We got the local host link here. Let's try this out. And we'll just say are you working inside the chat? And you can see like it switched the API provider inside open claw to gemma 4. But let's see if it actually works. So it it can respond. I think it can use gem 4, but the main point here is like does it actually work well, right? Can it use the tools inside openclaw? And that's what we're trying to test right now. And also you can see it's quite slow to respond to. So we'll see how this goes in a sec. There we go. All right. So, it can reply inside Open Claw and it can respond. It just it is a little bit slow, but you could get for example OpenClaw working locally on a phone now using Gemma 4 in this local model. So, that will be an interesting option. Bear in mind as well there's a smaller version of Gemma 4 which is 7.2 and you could potentially use that 7.2 GB version instead. which might be better. Also, something to note here, if you look, there's a difference in the context windows between them. So, you've got gem for latest, which is a default model, and that's 128k token context window. So, it's not massive. It's not going to be it's definitely behind like a lot of the frontier models when it comes to context Windows 2. And why is that? It's because it's designed for mobile devices. It's not designed to be the most powerful AI in the world. And then if you look later, you've got 256K context. Now, that's interesting because for example, if you're using something like um I think GLM or maybe even Miniax, I think Miniaax's latest model comes with like 200k token context window. Let's have a look at it. Yeah. So, look look at this. This is interesting. So, Miniax M2.7 cloud has a smaller context window than Gemma 4, which is a local model and more lightweight. really interesting to see that. So like in some ways, you know, you've got to download the 18 gig or the 20 GB model there, but you can see that you could interestingly you could run this with a bigger token context window than Miniax 2.7. Pretty crazy. So that's basically it. That's how to get access to Gemma 4. How to use it, how it works, how it performs. I would say don't use it inside Cloud Code. I would say download it locally. like it seems pretty cool for that. And then also it's worth testing inside OpenClaw. Just be prepared to understand that it's going to be a little bit slow to reply but at the same time it's got vision, it's got tools, it's got audio, it's got four different models and you can easily install it. It's super easy to get started set up. Right. So thanks so much for watching. If you haven't already, check out the AR profitab link in the comments description or go to the arpufferboarding.com. And if you want to get all of my best trainings, automations, workflows for using tools like using local models and growing your business with AI, then you can get the AR profit boardroom link in the comments description or go to the AR profitable.com. And inside here, you can ask questions, you can get help, you can get support, you can connect with 2,700 members. We have a calendar with weekly coaching calls. You get four weekly coaching calls where you can connect with other people doing the same things as you. Connect with people doing AI automations. Inside the classroom here, you can see all of our best training on like how to go from beginner to expert with AI automation, how to use the strategies I use to grow my business with AI and also how to get the latest updates on all of the stuff that's just recently come out, plus a six-hour course on open claw. So, if you like this stuff, feel free to get the air off boarding. I've got full trainings on like how to grow my YouTube with AI and also how to rank number one with AI SEO and that's all inside here. So, thanks very much for watching. I appreciate it. Let's see what we got here. Is it a good choice to use a local LM as main brain for open claw and make it use other LLMs with other agents? So I would say like for the main brain, you know, and I I don't see anyone doing that. No, I would say you you want to switch it the other way around. So like you have your best API as the brain of your open claw. So for example, you might have like Claude as like the operator right at the top and then what that is doing is that's controlling your local LMS. So for example, we could have claude opus 4.6 is the main brain of open claw or Hermes and then that controls everything underneath it. So for example that controls your sub agents that are running with um Kimk 2.5 or for example uh you know Nvidia models whatever sort of local AI you're using directly. I think that's a better choice because bear in mind like local LLMs are not going to be as powerful or as smart as something like Claude and so you want the brain the host of your operation to be the smartest AI possible. Which one is that going to be? probably going to be something like Claude or you know you could have something like Miniax 2.7 running the show as well. Also is this maybe improvable on lowerend hardware with Turboon? Yes. So for example there I actually saw a video of someone running Gemma 4 on a mobile with open claw and they were running it with turbo quant and so that was a smarter way to run it for sure but again it's never going to be as powerful as like these frontier models it's just a way of making like local models run faster and and smarter and more efficiently right so that's the way you can look at So today we are going to be running and testing out Gemma 4 inside open claw. Right. So we have gemma 4 which is the new local model from Google. It's a free local model and you can install it in one click with open claw if you've never done that before. I will guide you through the whole process and we're going to be testing Gemma 4 with open claw and seeing how powerful it is. So how does this all work together? So there's three simple steps. Number one, you're going to download. How do you download? You copy this command. You go into your terminal, you open up a new window, and you just paste in this command to update to the latest version of Volama. Bear in mind, if you're on an older version of Lama, if you haven't updated to today's version, you can't use Gemma 4 yet. So you must update to the newest version of Lama which you can do by just running this simple terminal command from.com. Now once you've done that you can go over to models and then inside models you go to Gemma 4 and you copy and paste this command. So if we go inside here like so and we run this command oama run Gemma 4 that will pull in the latest model from Gemma 4. Now I've already got it installed. If you haven't installed it before, it's going to be about 9.4 gigabytes. Right now, Gemma 4 is powerful because it's local, it's free, there are no limits on using this, right? You can use it as much as you want and it runs directly with Open Claw in one click because we've covered step number one, which is downloading OAMA. Now, we've covered step number two, which is installing Gemma 4. And step number three, the last step is you copy and paste this command. You go into your terminal and you you open up a new terminal window and you paste that command in like so. Right click proceed. Boom. Shakalaka. You've now got this starting up with your assistant. Right now, I did do a quick test run with open claw. So, if you go to openclaw here, I did do a test run before. So if we and it was a little bit slow, but that's okay. If we're getting it for free, there's going to be some sort of compromise, some sort of price to pay, whether that's time, whether that's energy. And today maybe a little bit of time, but let's see how it performs. All right. So we're going to say, okay, create a SEO calculator in HTML and just see if that can actually do it. So you can see like it is not as responsive or as fast as any other AI model, right? And why is that? It's because it's running locally. Now, if you're wondering, okay, what is my setup to run this? You can see it here. So, I'm running this on a Mac Studio with a Apple M4 Max, and it's now responding inside Open Claw, right? So, it is a little bit slow. Now, you can always, for example, what you could do, I think this would be a much smarter way of running Open Claw is that you've got Open Claw, right? And then what you can have is for example Claude or Miniaax as a brain of this and then that can delegate tasks to local models. That could be Gemma 4. It could be another model like Neatron Cascade. It could be a different model like for example Neatron 3 Super you know all the Nvidia models. But basically you give a task to open claw that gives it to the main brain which is claude and then claude delegates the tasks to your sub agents right your local sub aents running with oama and I think that will probably be the better way to do it but you can see here that it is replying it is doing its magic is working and responding and it's actually we're using this 100% free, right? Like there there's no cost involved with this. Now, it's actually responded whilst I was talking to you and so it's already built the HTML page. I don't expect this to be the best code in the world, but let's test it out and see how it goes. So, we can copy that. And also, it's not responded in a weird way. I've seen more I've seen APIs struggle with responses like this. Gemma 4 has actually used the tools correctly. It's actually done the job properly and it's actually worked inside open claw which is pretty amazing to be honest. So if we go and have a little cheeky preview, we'll go to live weave and just put this in like so. I don't know why open claw is super buggy. I click copy here, it doesn't work. It's okay. We'll just grab the code like this instead. Do it manually. And that is not to do with Gemma. That's just open core directly. So we paste it in. Boom. Page works perfectly. Right. Yeah. Actually works. So it's created like a working page. I mean this is this is not bad. I thought this would be way worse, but it's actually worked perfectly with open claw using gemma 4. So, so far the tests have worked. It is a little tiny bit slow, but at the same time, it it created the code and it worked perfectly and there were no bugs. That is amazing. Let's see what we got here in the questions. So you only use free models to generate your formula. No, because you're using an image model, right? There's no um I I don't use local image models. So something like Gemini flash preview, sorry, uh Gemini 3.1 Nanoban 2 would be a better model for generating the images. But that's something totally different, right? Like that's that's just calling an API to generate an image. What if you can load just one model? Can you switch between them in local? I think you could, but you'd need both models running at the same time, right? So, for example, if you had Gemma and you had Neatron 3 Super running at the same time, then you could switch between those models, but again, there's not going to be like anything super intelligent that's directing the sub agents. And so, it's going to slow your laptop down a bit or your setup down a little bit. And also, there's no brain in that operation, right? You're better off having clawed as the main API and then you run local LMS underneath it. All right. So can't open claw unload one and load the other. Like it's not just going to switch between them and start one and start the other up. Right? I mean it could but that would be an inefficient way. What you want is you want both local models running at the same time 24/7, right? because if it start stops start stops bear in mind you're going to lose context inside the local lm as well if you do that whereas for example if you've got lama running 247 then you can go from there I've been able to have a second model as a fallback if open router failed yeah so I think openclaw just fixed that yesterday in the 4.1 update and so failbacks are working better than they were previously And yeah, you could have probably you could probably have like OAMA as the the backup if that makes sense. But yeah, we are talking about local models inside this video. All right, so it works inside Open Claw. We've set up the free model. It's easy to do. You can do it in three steps. I've shown you exactly how shown you proof that actually works. I've shown you that it's all free right now. If you wanted the best model inside openclaw using only Gemma 4, you would go with either the 18 gigabyte or the 20 gigabyte models. The reason for that is because they have larger context windows, right? So you can see 256K 256K whereas the smaller models or the default model have lower context windows. Obviously, agentic models like open claw need big context windows because you've got a lot of back and forth and it the the beauty of open claw is the memory itself, right? And so you would probably do better having this set up with something like that. But yeah, that's basically how to use this directly and how to get it all set up. Super easy. Super easy. You can go inside the open claw on boarding as well and switch there, but I think that's a lot more work. And then if you wanted to change back inside your dashboard here, you can switch between these models, right? So we can always switch back to a different model later if we want to. We don't have to stick with OAMA. Now, one of the problems that I found though is when you switch from an old local model to one of the other APIs, it doesn't work. Let me just test if that works or not. So I'm going to just run a little test here and see if this actually works. Sometimes you have to like restart the gateway for example and that sort of thing. So we'll just test if open claw uses the new model or not. So it's not working there. Just replies with heartbeat. Okay. Let's try another one. See I've seen this before. this error here. Like you can see. Let's just test. So I've now switched to open router instead of gemma 4 and it's struggling. So I think what you would do in that situation is you would switch over in the on boarding back to the original model you were using. So you can see here is like struggling since we switched back from Olama to open router. The the way that I normally fix that is I just go into claw code and ask it to fix it for me. But yeah, you get the point. So that's basically how to use gemma 4 with open claw for free and to run commands locally with it. Pretty easy and simple to set up. If you want to get all of my best training on openclaw oama local models, if you want to connect with a community of awesome amazing people, one of the best communities in the world for AI and you want a community where you can ask questions, get help, get support in real time, then you can get that all inside the AI profit boardroom link in the comments description or go to the arprofit.com. We have four weekly coaching calls. We have a map where you can connect with people in your local city. Inside the classroom, you can get all of my best trainings and courses on exactly how to use OpenClaw, O Lama, Hermes, and all my best agentic training inside the SOP update section here. You actually get a full six-hour course on OpenClaw and another three-hour course right here. And this is all inside the AR profit boardroom link in the comments description or just go to the arprofitboard.com to get access. Thanks for watching. Loserless claw helps with the context window. Yeah, I think so. But I I also think like if you're calling in another tool to help bear in mind like a model like Gemma 4 is not going to be great at calling tools. So for example, it struggles with claw code even. If it struggles with claw code, then it's going to struggle with calling tools. So, I don't think it'll be that effective for having long-term context. Have you gotten ACP working with coding harnesses like open code? Um, [clears throat] so have I like for example, I've used like open code directly with anti-gravity and claw code before, if that's what you mean. Yeah, a lot of uh information on the AI profitable on on how to do all that sort of stuff. Ang says, "Hi, good day." And then Sporter says, "Am I the only one here who learned everything I know about AI from Julian?" Thank you very much, sir. Appreciate that. be with you in a minute. Just wait for me. Thank you, peeps. Today we're going to be running through the latest update from OpenClaw. So, OpenClaw just released 4.2. OpenClaw 4.2 came out on the 2nd of April. And you can see here that there's five big headlines to durable task flow orchestration, better native exec defaults and approvals, co-pilot and Kimmy and provider hardening, tighter plug-in activation boundaries, and hardened provider transport and routting. If you haven't seen the full guide, right here you can see all the details on in terms of the change log. So we've got the announcement over here and then we can have a look at the change log and check this out. All right. So, you can see all the details like so a lot of different fixes. I don't think there's any like big headlines from here, but you can see all the details in terms of what's just come out. I'm going to talk you through it right now. So, this is supposed to make it faster and leaner than ever before. We'll test this out and see how it performs. So, first of all, OpenClaw works with 23 messaging apps. So if you're wondering how it works here, here's a a diagram of how it works. You go from your messaging to the gateway. The AR responds back to you, right? Everything runs on your computer. So what is new inside version 4.2? Well, basically number one is durable task flow orchestration. What does that mean? What on earth does that mean in English? Let me explain it to you in plain English here. So before this update, if you gave openclaw a big job, for example, like research this topic and write a report and send it to my team, it could get confused or lose track halfway through. Now it remembers every step of the job, right? So even if your computer restarts, it can pick up where it left off. It's got better context. It's got a better persistent memory. Now, why does this matter? So it means we you can give open call bigger multi-step tasks and trust it to finish them. Right? So if we go inside openclaw we can actually trust it with bigger tasks. we can give it more tasks and we can give it more information to to basically work with, right? Whereas before it would struggle. I've seen Hermes struggle with that as well recently actually as well. Better native executive defaults and approvals. So exec means execute, right? Which means running commands on your computer. Before open claw could run commands a little too freely, which isn't good for security. Now it asks for your approval before doing anything important. It's like adding a are you sure pop up before deleting a file. So why this matters is your computer stays safer. You stay in control more and the AI can't accidentally break something without your permission. Then we have feature number three which is copilot and Kimmy and provider hardening. So in English that means openclaw now works with Microsoft copilot and Kimmy as the brain to your operation. So you can use Kimmy as the main API or you could use Microsoft as the main sort of controller here. Before you were mostly limited. So now you have more choices for what an AI powers your open claw. Right now provider hardening means all the connections to those AI brains are now more secure and stable. Why does this matter for you? Because more AI options equals more flexibility. And also you can have fail back uh failover APIs. So if one API fails, no problem. You switch to the next one. You switch to the next one, right? And it keeps running without any sort of disruptions. Then we have feature number four, tighter plug-in activation boundaries. What this means in plain English, plugins are like apps inside OpenClaw. Little add-ons that give it new powers. Before plugins could sometimes activate when they weren't supposed to. Now they only turn on exactly when they're needed. Why does this matter for you? Because your AI assistant runs cleaner and faster. So no random plugins jumping in. By the way, if you like this stuff, check out the AR profit boarding link in the comments description. Inside the classroom, we have tons of training on OpenClaw. And we actually have a full six-hour course on how to use OpenClaw and another three-hour course right here. And we have tons of training on how to use OpenClaw to get this working with your AI agent along with an amazing community. And it it's all about helping you save time and grow with AI automation. The final feature is hardened provider transport and routting. What this means, transport is how data travels between open core and the AI brain. Routting is which path that data takes. Both are now much more secure and reliable. So, it's like upgrading from a dirt road to a motorway. Why? Because you get faster responses, you get fewer drop connections, and your assistant feels snappier and more uh reliable. And also, they've removed a lot of unnecessary code and features that were slowing it down. If you've seen that recently, it's been a lot more buggy. It's been struggling. It's been slower to respond. They've sort of debugged it and reduced some of the code inside there as well. All right. And that's basically it. That's exactly how you can use it. You can see some of the forward slash commands. I think this is useful to recap on. So, you can see what model you're currently using and how many tokens you've used by using the forward slashstatus command inside openclaw like this. So, we can type in forward/stus and then it will give us the status right here. So, for example, we're running this with open router directly. You can use new to start a fresh conversation. Reset as well to wipe the memory. Compact to squish the current conversation down to save memory. Think high, which makes the AI think harder before answering. Verbose on if you want to add extra detail inside the replies. Turn verbose on. And then usage for this shows tokens and cost info after every message. Right? So if we put usage full like so, you can actually see the cost and the context of everything that's being used so far. And then they've added some other cool stuff like voice wake words which is pretty cool. You got live canvas so you can actually draw and write on a visual workspace so you can see in real time Android as well. It works on Android and browser control as well. Right. So that's basically it. That's how to use it. What the update is. Um, just to recap, OpenClaw is a 24/7 AI assistant and the new version adds durable task flows, approval gates, copilot, Kim support, and a fastelina engine. You can update very quickly by just selecting the update button that will come up on your dashboard. If you don't have that or don't see it, you can just ask it to update itself inside the chat. Simple as that. Simple as that, my friends. All right, thanks so much for watching. If you haven't already, check out the A prof link in the comments description if you want to get access to this. You can ask all of your questions inside the community. You can get access to all of my best trainings inside the classroom. You can also go inside the map here and meet people in your local area who are um in your city right now, which is pretty cool. And then inside the classroom, you get access to all my best trainings. So, if you go inside the SP update section here, you can see all of my latest tutorials and trainings on all the new stuff that's just come out with AI. like so. And this is all inside the AI profiter board link in the comments description or just go to the proferborn.com to get access. Let's see what we got inside the questions here, peeps. Nice to have Gemma models that can use the tools. Absolutely. This AI is moving so fast. 100%. Do you have trainings on how to make your 30-day plans so as I'd like to do them? Yeah, if you direct message me inside the AR profit boardroom, then I will send you the prom that I use for my 30-day plans. So, yeah, you can get that all inside profitable. Thanks so much for watching. I'll see you on the next one, peeps. Cheers. Bye.