Build Anything with DeepSeek V4, Here's How...

Today I'm going to show you exactly how to build and automate anything with DeepSeek V4. So if you haven't got free access to it, or if you don't know how to access it, I'm going to show you exactly how to build anything with it. And we're going to start off with the chat, that's the easiest way. So what you can do over here is you can go to chat.deepseek, this is completely free.

You go inside the chat here, and then you can switch between instant and expert mode. Now if you want to build something instantly, you would use instant. Bear in mind you've got DeepThink and on DeepThink. If you want the search as well, if you want web search, so if you want it to be able to search for information on the web, then you can use the web search feature, right?

If you don't want it to search information, you don't have to use it. So, you know, just use it when necessary, basically, right? And then from here, what we've got is DeepThink, right? DeepThink.

So with this, what it means is, do you want thinking on or do you want thinking off? That's as simple as it can go. Now from here, if we wanted to build something, then we can just take a random prompt and start building with it. So if we say, okay, for example, create a Pomodoro timer.

One thing you want to do here when you're creating anything or building anything is just select how you want it to be built. So the easiest way, if you want to be able to preview this inside the chat, is to build it in single HTML. That's the easiest way to build and check it out. So that's what we're going to do right here.

We're going to type in single HTML. Then we're going to hit enter, like so. And from here, we can start getting the HTML. Now, one thing to note here, when you're prompting it, is if you're using the chat, single HTML will allow you to run a preview of it later.

Now, if you want to download it and upload it anywhere else, as a website, whatever, then you go to download over here, right? So you've got run, where you can preview it in a second once it's finished building. And then you have the download option. Now, you can have multiple tabs building with this stuff, right?

So if you want to, for example, try something else out, here's what I'd recommend, right? So you can select DeepThink over here, which will allow reasoning for the model. And you can select Expert, which is the best model on the platform, right? DeepSeek v4, best model you can use.

So what you can do from here is you can then give it another prompt. So if we say, OK, build a simple, fun pong game in single HTML. Again, that single HTML is important because otherwise what's going to happen is you won't be able to preview it. And then you'd be like, well, what do I do with my code, mate?

You know, how do I use this? So that's what you want to do. And then you've got the run option inside this. We've now got two versions of DeepSeek coding out side by side.

And again, this is all free, right? You can use the platform for free. You can code stuff out for free. You can run it for free.

Boom, shakalaka, we're living the dreams, my friend. Now, I did promise you that you can build and automate anything. So we're going to just wait for these to finish. And in the meantime, what I'm actually going to show you is exactly how to create one click AI agents with this as well.

So over to if you go to Olama, you can download Olama over here. And what this allows you to do is run cloud models like DeepSeek v4 Flash. Now, v4 Flash is not the most powerful version of DeepSeek. That's OK.

You know why it's OK? Because for most people, 99% of people watching this, you're not going to need the full power version anyway, right? And so if we pull this up, we've got, for example, DeepSeek chat over on the right hand side, sorry, the left hand side over here. And then we have Olama, which we're going to run with DeepSeek agents in a second.

Again, we're focusing on building and automating anything. And this is one of the easiest ways to do it. So what we can do here is you want to make sure that you open up your terminal and all you do is you copy that and you paste it over there. Copy that, paste it over there, right?

At olama.com. That's how easy it is to run Olama and set it up inside your terminal. So you don't need to be like a terminal wiz or a coding wizard or whatever. Don't worry about that.

Now, if we have a look at this particular example, you see that we've actually got Hermes AI agent running with DeepSeek cloud, right? That is amazing. Now, I want to show you exactly how to set up AI agents and multiple different methods you can use this for. So if you go to Olama, then you're going to go over to models, right?

So you'll see it on the navigation bar here, models, or you just go to olama.com forward slash search. And from here, you can start using DeepSeek however you want. So if you go to DeepSeek v4 Flash, this is free as long as you stick within token limits. Now, important to note here, when you're doing this, if you're using DeepSeek v4 Flash, number one, it's quite a lightweight model anyway.

But number two, it's designed to be efficient with tokens. So if you're going to use a model for free on the cloud system, then this is a pretty good option, right? It's a pretty good alternative. Now, I'll come on to exactly how to use this for five different applications in a second.

So you can build and automate anything. But from here, if we go back to the Pomodoro timer, like you can see, we've now got that ready to preview. And boom, look at that. We just built out this amazing tool in the space of like one minute, right?

How do we do it? We just typed in a simple prompt. So you've already learned how to build out basically many SaaS tools inside DeepSeek directly for free, right? Crazy.

So from here, you can switch between, you know, for example, short break, you'd have a long break. I just want to show you this actually works, which is amazing. And then you can reset it. You can start it.

You can pause it. You can reset it. Blah, blah, blah. Pretty nice.

So that is method number one. And it worked perfectly first time round. Beautiful. And bear in mind, that was with the instant model, the super lightweight instant model that is free inside chat.deepseek.

So that was pretty cool. Now, if you go over to this tab, now you can see we've got the ping pong game ready to go. Let's have a look at that. Boom.

Look at that. That's pretty cool, right? That is pretty nice. And again, you just learned how to build that and automate that in one single click.

It was so easy, wasn't it? Super easy stuff. And so you now know how to build mini apps and also how to build games for free using DeepSeek V4 Pro. Now, we're going to do a lot more today, but I just want to make sure that you know exactly what you're learning to build here and how powerful this stuff is, because it's pretty amazing.

And, you know, again, you don't need to be a coder, right? You don't need to be a genius to use this stuff. It's pretty simple. You just type in what you want and explain it in plain English.

So that's exactly how we can use this. So from here, what we can start doing now is we can start building out AI agents. So if we go over to Olama over here, we've got five different ways that we can use this, which is pretty amazing. So I'm going to set up some different options here.

I'm going to see, you know, what do you like? What do you prefer, etc. So just to bear in mind, we have set up Olama. That is the most important step.

So we're just going to close that terminal window here, and then we're going to open this up and we're going to code out multiple things, right? So I'm going to just have a terminal window over here, and I'm going to have multiple different terminal windows. It's going to be fun to use. So I'll show you exactly how this works, right?

And again, this stuff is free as long as you stick to the token limits. And if you don't want to stick to the token limits, you can either run DeepSeq locally if you've got an amazing setup. If you don't have an amazing setup, no problem. You can just use the local models with something like Gemma4Flash, right?

And so you see how we're building out these terminal windows. The reason that we're doing that is because we want to have multiple agents building and coding and working together, my friends, in perfect harmony. So if you go over here, we've got these terminal windows, right? And we need five of them.

So I'm just going to make this even smaller. And bear in mind that one of them might not work, right? Let's be honest, one of them might not work. And then we'll show you how to set up five different applications at once using DeepSeq.

And you just use what you prefer, mate. That's what I say. So from here, we've got five terminal windows. Good stuff.

Now what we're going to do is we're going to start copying and pasting these into these, right? So just to recap on the steps here, you download Alarmer, you open up the terminal, and you don't have to open up five terminal windows at once. This is just something fun that I want to show you. So from here, we're going to open up Cloud Code.

Boom. That's going to work with DeepSeq, right? So we've now got Cloud Code working on Flash Cloud. And this is free as long as you stick with the token limits, right?

Now, if we go over here, we can also run ChatGPT codecs with DeepSeq v4 Flash as well, right? We just need to update this. So we're just going to update the codecs right there. We can also use this with OpenCode.

Sometimes this doesn't work so well. So I'm just going to see. Yeah. So we just need to install OpenCode first, which is OK.

So I'm just going to go to OpenCode.ai and then we just need to copy this terminal command, paste it in there, and then we can install OpenCode as well, right? OpenCode is another sort of powerful coding agent that we can use. From here, we also want to set up OpenClaw. OpenClaw is pretty fun to use as an AI agent.

And then we can also have Olama, right? Olama with Hermes running as well. So that's basically how we can use these. Let's just have a look here.

So you can actually see that this is typical of OpenClaw, but it has failed to run there. So we'll fix that in a second. But in the meantime, let's just get this terminal party started. That's what I say.

So we're going to open up these windows over here. And now we can start coding and building anything that we want with them. It's going to be fun. So there we go.

Here we go. All right. Then I'm going to take some prompts, which I'll come on to in a second, okay? So I'm just going to see why is that not working?

Now, Nino says, what's up, Julian? What are we building today? So, so far, I've already shown how to build for free a mini sort of app tool for productivity and also games for free with DeepSeek using the chat for free, all right? So, we've already covered that, but we're going to go into more depth in a second.

So, if we go over to these coding agents, right? I'm just going to grab some prompts that we can use. Now, you might be wondering, what prompts do I use? So, what do I even use with DeepSeek?

Like, I would look at where do you spend your time and then what do you need to automate based on your time? That is by far the best way to figure out what you need to code, right? And I'll show you a good example in a second. So, over here, inside OpenCode, we just need to run the terminal command because we had to reinstall it, all right?

So, we're going to go back to Olama, go to DeepSeek v4 Flash, which we've got over here. Bear in mind, DeepSeek v4 Flash is the more lightweight version of this, right? This is a more lightweight version of DeepSeek. So, from here, we're going to go over to OpenCode.

We'll paste that in. Boom, shakalaka, that's going to start running. There we go, all right? And we've got DeepSeek v4 Flash Cloud plugged into there.

And from here, we've now got multiple terminal windows. Now, if we were looking at Cloud Code, here's the way that I would use it, right? So, if I'm using Cloud Code, I would actually go inside the chat here and I'd be like, right, okay, OpenClaw isn't working with Olama locally, right? Because we just saw that a second ago.

You can see over here. So, if we run that, I'm just going to give it one more shot at trying to run this. Let's see if we can get that. We'll just try running it one more time.

And then we'll try this over here. Oh, look at that. That's working. All right, I'm just going to say working and just check it's working with DeepSeek.

Nino says, I'm building a game right now. So, things are much faster. Some bugs are annoying. Yeah, I mean, that's part of the game, isn't it?

But it sounds like you're doing good work there, man. Fair play to you. So, we're just going to zoom out here. It seems to be responding, but we'll come back to that in a second.

All right, in the meantime, let's start using OpenCode. So, if we, oh, yeah, is it working? Beautiful, beautiful, it is working, right? So, now we've got DeepSeek working as, sorry, DeepSeek working with OpenCode as well, right?

So, we can come back to that in a second. In the meantime, let's use code with DeepSeek now as well. So, if we go inside the chat here and we say, okay, build out a beautiful, and then what I'm going to do over here is grab some of the prompts from our group. So, for example, this Space Invaders game.

So, let's try that one right now. I'm going to try that. Then we're going to go over here and we'll take another prompt. So, for example, let's take this one.

We'll grab that and we'll plug that into OpenCode. All right, so now we've got Cloud Code working with DeepSeek V4 Flash and we've got OpenCode running as well, right? Pretty nice. So, it's super simple.

And then we're going to do this with, let's see what else we got here. We have Codex as well running with DeepSeek in the bottom left. So, this is Codex over here. And then for Hermes, the way that I'm going to use Hermes, right, and I'm going to just minimize OpenCode because we've got it inside the gateway anyway.

I'm just going to try something now. So, we're going to say to Hermes, go to the web and search Google for AI automation news. So, now we have five different AI agents all working together, right? All building crazy stuff.

We have Hermes over here, right? And this is basically browsing the web as an AI agent. We have Cloud Code building out a Space Invaders game. We have Codex running over here, right?

So, that's building out directly. And then we have OpenCode, which is already finished coding, which is crazy stuff, right? So, let's have a look here. So, we've got this page here.

I'm going to say, open it up in my browser so I can preview it. Boom. Look at that. Yep.

That works. That works, my friend. All right. So, we've built out a pretty cool little option there for creating and creating like sort of SaaS tools and that sort of thing.

We've got OpenCore as well. All right. But everything so far has worked really well. All right.

It is, do you know what? It's interesting. Like, if you actually go inside, I think it needs a good harness, right? That's what I would say with DeepSeek.

It's like, when I used it yesterday and I tested out for the first time, I was like, this is not that interesting, right? It's not that good. When I test it out inside the, directly inside the chat and I tried to create a website, right? But what's interesting is like, if you go to DeepSeek and you put it inside a harness, right?

So, for example, like OpenCode, then it's super powerful, right? As you can see right here. Same with ChatGPT codecs over here. And you can see sometimes it's going to fail as well.

I just want to be 100% honest with you. Like, you see this has failed right here. I will show you a more interesting way to run that in a second, right? But you can see how basically Claude code has failed on us.

It just gave up. It said, we have to log in. It doesn't work with the API. Okay, no problem, mate.

We'll shut you down, right? But that's okay. That's okay. Because we've got OpenCore over here.

We have OpenCode and we have Hermes, right? And they're all working nicely along with Codex as well. So let's just open up Codex, which we've got over here, all right? And so we have Codex building.

OpenCode seems to be the best so far. Like, I don't know how it's so good. OpenCode and DeepSeek, amazing. Absolutely amazing.

So what we can do also inside here is we can say, okay, based on what you know about me, what could you help me automate, right? And so we've got two different AI agents running at the same time with DeepSeek V4 Flash, right? We have OpenCore and we have Hermes. By the way, some people ask me, like, what's my favorite?

I would personally say Hermes is my favorite, right? It's pretty good stuff. Hermes is just a lot smoother. I mean, you already saw, for example, OpenCore break earlier today.

Like, you know, Hermes is a lot smoother, a lot easier to use, a lot more chill, et cetera. Way better in many different ways. So we have Hermes over here. I mean, do you know what the hardest part of this is?

It's keeping them going. It's like actually finding interesting stuff for them all to do at the same time. So let's take some more stuff here. I'm going to take this one, like so.

Anything like coding related, you know, if it's a landing page or a website or whatever, that's going to be best for OpenCode, right? Anything for like agentic tasks, that's going to be best for Hermes or for OpenCore, right? So it really depends. What do you want to use?

What do you want to build, et cetera? But that's basically how you can use DeepSeek with Hermes and OpenCore and also with OpenCode, right? So I'm going to close Hermes for now because I just think like it's almost, it's too much for my brain at this point to run through them all. But we'll have OpenCode running and we'll have ChatGPT codecs running and then we'll also have OpenCore running, right?

And so I want to like just break down what we've done so far and just recap on like how you can build and automate anything with DeepSeek, right? So you've got a few different methods and if you want like a full SOP on like how to build and automate anything with DeepSeek as a full guide, I actually got that inside the AI Profit Boardroom, right? So if you go to the AI Profit Boardroom, go to the classroom here, scroll to Advanced New AI Tutorials. We have inside the AI Profit Boardroom, we do have like a full five-hour course on DeepSeek already.

But what I'm going to do is if you want to see like the latest stuff on DeepSeek v4, I'm going to plug in this full guide on how to build and automate anything over here, right? Along with that will come with a video tutorial as well. And we'll fill that out with a full step-by-step SOP later. All right, so let's come back to this now.

So you can see, for example, here, I mean, this is crazy. Like basically it has, I'm going to show you something interesting as well that I've done to make these agents even better, right? So you can see here, for example, this understands, this has so much context, it's on me, it's insane, right? It is insane.

And I'll show you why that is in a second, right? So basically this understands exactly what's going on in my world this week, right? If you're wondering, okay, like how do you give it that much context? How do you get DeepSeek to understand you on such a deep level?

Let me show you what I'm doing because it's pretty crazy, right? So what I've actually done is I'm using something called Omi and Omi records my screen. It records any sort of microphone or anything that I have going on, right? And so using Omi, I link that to Obsidian, right?

And every time I export to Obsidian, to the Obsidian vault, this updates my Obsidian vault with all the information it has about me, what's going on in my world, what I'm working on, right? Because it's literally recorded like everything that I'm working on, right? And so you can see here that it has all these memories on me, which is crazy stuff, right? It understands everything about me, understands what I do day to day, what I work on.

It has 570 memories and I don't have to update this manually myself, right? We're using something called Omi and here's where it gets interesting. So with Omi and Obsidian, when I have all those memories I export, I then link that to OpenCLR and so OpenCLR understands. understands everything that's happened this week, which is why it has so much personalized information on me.

Because, I mean, for example, it even knows I'm going to Hokkaido, right? And that's because it has, I didn't even realize that, but it knows that because it has access to OMI and Obsidian. And I didn't have to write any notes, I didn't have to work on anything, I didn't have to create anything, I didn't have to journal anything. It was fully created using OMI because OMI is automatically recording my screen and it's already taken in, for example, the microphone and everything.

So it's pretty amazing. Let's see what we've got in the questions now. Y says, I'm installing it right now in Brazil. That is something I need to know before doing it.

I don't know. Oh, yeah. I mean, you can choose between Hermes and OpenCore. I'd recommend trying them both and seeing which one you prefer.

But for me personally, I prefer Hermes, right? It's just, it's more, it's more smooth to use. It breaks less, basically. What else we got here?

Boom Shakalaka, I love when I catch your lives. Hello from Vancouver, Canada. Thank you so much, sir. Appreciate you joining.

Sorry, not sir, Michelle. Thank you, Michelle. I really appreciate that. And then Administrator says, wow, thanks, brother.

Happy to help, happy to help. I'm glad this is helpful for people. Now let's have a look here. So what we got here, Nino says, how important is the difference of plain AI versus same AI skills?

I think the best thing you can do is like you look at what you're working on day to day, and then you, you decide what skills you need to create. And that's how you can basically figure out, okay, what is the best way to use this AI? So the skills or no skills, I don't think it's a big deal. The main thing is, is it personalized to you?

Does it have context on you? Does it understand you, right? If it does, like it does with me with OMI and Obsidian, then it's going to be amazing, and also the thing, for example, if you're using Hermes is every time you use DeepSeek with Hermes, it's going to write new skills based on what you give it, right? Based on what you teach it.

So that's what you want to work on. Yeah. All right. So now we've got the bird game here, right?

So we just created this with OpenCode. I mean, DeepSeek is absolutely amazing, isn't it? Look how good this is. That is pretty amazing that we just built that using OpenCode and DeepSeek.

Wow. And then we've got this game here. So if we type in, for example, Dark or Click, right? Epic, this works perfectly, right?

So, I mean, it's just so good, isn't it? It's just so good. So, and then we've got the, you know, the personalized setup here. So let's talk about all the different ways that we can use this because there's so much here to unpack.

There's so much more you can do with it. I feel like I barely touched the surface teaching you, but let's have a look what we got here. What's the best use of DeepSeek? Everything that we're showing you, right?

So building out tools plus mini apps, building out games. We haven't even come on to building out websites, but we can do that in a second. Then we've got, for example, AI agents, right? So you've got AI agents, e.g.

Hermes plus OpenCode. And then you might be wondering how, okay, is easy one-click setup for everything above. And then also chat.deepseek, right? So if you're wondering, okay, like what are the best use cases, building out tools and mini apps, building out games, building out websites, AI agents like Hermes and OpenCode, OpenCode is amazing.

Cloud Code, I wouldn't recommend actually. And then also you've got, for example, what was the other one, chat.gptcodex with DeepSeek, super powerful. So they were some of the best use cases that we're talking about today. And then also, you know, going to AI agents, e.g.

OpenCore, like we were talking about here, and just asking it based on what you know about me, what could you help me automate, right? And then also the other thing I was going to say here was Obsidian plus OMI, right? Equals winning. We have inside the AI Profitable Boardroom a full, you know, like I was showing you Obsidian and OMI right a second ago.

If you go inside the AI Profitable Boardroom and you type in OMI, right, we actually show you how to set up, how to get it working, etc. So if you want to learn that stuff, you can learn it inside the AI Profitable Boardroom. And we also have loads of training on Obsidian, right? Obsidian is super powerful too.

And we have a lot of good training on that sort of stuff. But yeah, winning contacts equals, you know, I would literally say like, if you set up Obsidian and OMI, your AI agents are going to be 100 times better, plus more personalized. The other thing that I recommend, and we have a full workflow on this when you join the AI Profitable Boardroom, is like, you want to look at what you're working on day to day. And that'll help you figure out what you need to automate personally.

Now, Jacob says, you know, Obsidian versus Notebook LM, as you've seen, right, like, Notebook LM is not going to take automatic notes on you. Whereas, for example, Obsidian is. That's the biggest difference. Notebook LM is not going to plug as easily into all of your AI agents.

Notebook LM actually will, right? So Obsidian is way more powerful. Notebook LM is great, like, for creating research. It's great for media.

You know, I've got loads of videos, tutorials, and it's a great tool. But it's nowhere near as good for context and understanding you personally, right? So best use cases, building out tools and mini apps, building out games, building out websites, AI agents, open code, chat-tributed codecs, going to AI agents, for example, open quote and say, based on what you know about me, what could you help me automate? And then actually implementing it, right?

The big important thing there, actually implementing it. A lot of people won't, but that's super important right there. And then you've also got Obsidian OMI, right? Get that amazing context right there.

And then just look up what you're working on day to day. That'll help you understand what you need to automate personally, because you know where your time goes. And then how do you do these things? I mean, number one, easiest one is DeepSeek with chat.

Second best one is Terminal with Olama. And then you set up open code, codecs, open core, Hermes, right? Those are some things that you can use DeepSeek for. Now, also what you can do here is you could actually go over to, for example, Hermes with DeepSeek.

So if we go over here, right? Let's open up a new thing here. And then we're going to open up Hermes with DeepSeek, which you've got over here. Open up a new terminal.

We're going to run Hermes with that. And once you've set this up, then you can go inside the chat and you start going from here, right? So from here, now you can go, okay, every day at 8am using DeepSeek, research the latest AI automation news for me, right? And so scheduling tasks is a big one as well.

So you want to make sure you schedule your tasks, right? Anything that can be automated, do it. That's what I say. That's what I'd say.

You know, what else we got here? You're right. Yeah. Jacob says, notebook is more powerful than PowerPoint.

Yeah, I'd agree. I'd agree. Do you have any video talking about typing minds? Not come across that before.

Julian, where can I go to talk to you? Go to the airport for William. That's where I connect with people personally. That's the only place I talk to people personally.

And you can just DM me inside the airport for boarding. All right. So you can see here that we've scheduled this tasking. So this AI agent now with Hermes is going to run scheduled tasks for us each day, right?

And so that's, that's running now daily. And it will just search the latest AI automation news with us as well. Now, if you've never used Hermes before as well, let me show you some other cool stuff that you can do. Some other use cases with DeepSeek, right?

So if we go to Telegram, ah, that's a good one. Jake, LearnStrong says agent zero as well. Yeah, we can come on to that in a second. So if we go over to Hermes here, I'm just going to check it's working.

So what you can actually do with Hermes, for example, is you can connect this to Telegram and then you can use it on your phone as well. Here's some examples, right? So this is me chatting to Hermes right now with Telegram. And so you can link this inside Telegram as well, right?

And then just operate it from your phone. So you can manage all your automations remotely from your phone by connecting to Telegram as well. So that's 11 different ways to use this, right? And, and I've shown you proof that every single method works.

We've used every single method today, which is pretty crazy, actually, when you think about it. So if we now come over here, ah, we got OpenCode, we got Chachibity Codex, Hermes, we have OpenClaw running the cloud too. Let me show you another one. So if we say, okay, open, this is good as well.

So you can actually use this for browser automation. So you can actually use your browser, right? So if we say, open up my browser and go to Google, and then you see how it's navigating to Google here, it literally, it just opened that up, right? It said, done, opened Chrome straight to Google, you're in, right?

And so browser automation is another one as well, right? Browser AI agents is a big one too. So that's another method for using this. So that's 12 different ways that you can use this, right?

So for example, if we go over to OpenClaw here, and we're like, okay, open up a browser, go to Google, and research the latest AI automation news, as an example, right? So we can actually get the agent to control our browser. So if we type that in now, that's now searched for the latest AI automation news, as you can see here, right? So it searched it on Google for us.

And, and yeah, it can operate our browser too, which is pretty crazy stuff. You know, like you could say to OpenClaw, okay, go and speak to ChatGPT. and have a conversation about life. All right, so it's open up chat GPT here, as you can see.

And that's basically how it can start interacting with your browser and start using it for you. So browsers use AI agents. Pretty easy to organize directly from OpenClaw. It's pretty good for that.

OpenClaw is good stuff. So let's have a look. Jacob says, for basic business daily tasks and automations, what's the most cost effective combination? So I mean, you can use local models, right?

So local models are a good option. Otherwise, you can get, for example, Kimi coding. I mean, even DeepSeek, to be fair, DeepSeek is pretty reasonable compared to, for example, Claude. Let's compare these.

So compare in a table. And you're talking about the API, right? So let's have a look at this. Compare in a table API costs of DeepSeek versus Claude versus chat GPT.

And we'll do 4.7 versus GPT 5.5, right? Versus V4, Flash and Pro. Bear in mind, there's Flash and Pro, right? So there's the model, which is Flash, which is the fastest one.

And then you've got Pro, which is a bit more heavy. But it works really well as well. So DeepSeek V4 is free inside the chat. If you're using it on API, no, it's not free.

And if you use it on Olama, then it's free as long as you stick within token limits, right? How is DeepSeek Flash compared to Minimax M2.7? So if you were comparing Minimax M2.7, you would use DeepSeek Pro, right? You're not going to compare Flash versus Minimax M2.7 because Minimax M2.7 is a much more powerful API, right?

Versus Flash. But if you're comparing, for example, Minimax M2.7 versus DeepSeek V4 Pro, well, we can do that in a second, right? Let's actually do that right now. So DeepSeek V4 Pro versus Minimax M2.7.

And by the way, if we just go over here, let's have a look at this table, right? So if you're using the API from DeepSeek V4, right? You can see the cost here. So look at, this is ridiculous.

This is absolutely ridiculous. So look at V4 Flash, right? That cost versus Opus 4.7. Insane.

Absolutely insane. Same for GPC 5.5, right? Like it's literally like a fraction, a tiny little fraction. So big difference right there.

Can you use OpenCloud to browse the web for free? You can with local, again, you can use local models. You could use a free API, like for example, Elephant Alpha was free as an API recently on OpenRouter. So you just want to figure out, okay, like where, if you want to do stuff complete for free, like browser automation, where can you get the free API keys?

You know, and then that could be OpenRouter sometimes. Sometimes it might be just running stuff locally, you know? So these are the differences. And then if we scroll down here, these are the differences between DeepSeek V4 Pro and Minimax M2.7 RAM.

So DeepSeek V4 Pro was released on April 24th, 2026. Minimax M2.7 was released on March 18th, 2026. If you look at the total parameters, so 1.6 trillion versus 230 billion, right? This is five times smaller as a model.

License is different as well. So the license for Minimax M2.7 is proprietary. The license for MIT is open. Now Minimax did release like an open source version, but it wasn't for everything, right?

It's not MIT license. And then you also have context window. So a million tokens versus 200K tokens, right? Big difference right there.

And if you scroll down as well, you can compare, for example, the differences here. So I mean, Minimax M2.7 is a lot cheaper, but it has a much smaller token context window, right? If you wanted to compare, let's do this then. So let's do DeepSeek V4 Flash Plus Pro versus Minimax M2.7 versus GLM 5.1.

That will probably be an interesting comparison right there. So let's have a look here. DeepSeek is crazy. The difference is ridiculous.

Yeah, I would agree with you. DeepSeek should compare to Kimi K2.6 and GLM. All right, let's do that as well. So we'll make sure we compare them versus, let's do that up here, versus Kimi K2.6, right?

So we'll compare all those models here and we'll see how they perform. Now, in the meantime, some people are asking about, by the way, out of all of these models, OpenCode has been my favorite by far. Wow, what a harness. What an ancient harness that is.

The actual sort of like harness that something is stored in, I'm going to close OpenCore as well in a second. The actual harness that something is stored in is, it makes such a difference on the outputs you get from the API, right? So if you're using OpenCode, you're going to get better outputs for coding than if you use the chat directly, if that makes sense. So it's something to understand there.

Something big to understand. Is there any equivalent to OpenCore for an iPhone that is free? So you can actually use KimiCore on your phone or you can connect OpenCore to your phone. You know, that's one option.

You can connect Hermes to your phone too. I better remind that OpenCore actually has an Android app as well. And then the other option is you can use something like PicoCore or NanoCore, which are super lightweight options, super lightweight versions of OpenCore and they run on your phone as well, right? So that's another option too.

So if we have a look at the artificial intelligence index here, V4 Flash, V4 Pro, M2.7, and you can see here the versus all of them, KimiK 2.6 is winning there, right? Number one open model, according to the numbers from Claude. And then if we have a look at the total parameters here, so V4 Flash, much, much smaller than V4 Pro. M2.7 is smaller than all of them.

GLM 5.1, sort of in the middle, 754 billion parameters. And then you got KimiK 2.6, which is 1 trillion parameters, right? If you're not sure what all of these are, these are all Chinese models, right? They're all from China.

And then active parameters here. So we've got 13B, 49B, 10B, 45 billion parameters, 32. If you're not sure what active parameters is, basically a lot of these models, particularly if they're like, you know, open source, what they'll actually do is they will have a mixture of experts model, which means like, for example, you may have 1.6 trillion parameters on a model, which makes it big. And typically like the bigger, the more powerful, but the thing is, right?

Actually, in reality, you're only going to use the experts in that model related to the tasks you're doing. So it will activate 49 billion parameters, even though the whole model is 1.6 trillion parameters. It's not going to use all of that. Because if it uses all of that total parameter, then the problem is that it's going to be very slow and it can get quite expensive and you won't be able to run it locally and it wouldn't be open source anymore.

So that's the biggest difference between all of these. Now, if we have a look at context windows here, so context window is 1 million token context window. So DeepSeek still has the biggest context window here. And then modality, you can see they're all text apart from KimiK 2.6.

And also Minimax, I think can do images, but it's on a slightly different model. You see the release dates here. So KimiK 2.6 just came shortly before DeepSeek. And then M2.7 came March 18th and you got GLM 5.1, which came on the 7th of April, right?

Now, if we look at prices here, you can see that V4 Flash is the cheapest option. So that's the most cost efficient here. And then you have the output as well, right? So if you look at who's winning, DeepSeek Flash is less powerful, but you can already see today, like it's done pretty well.

And KimiK 2.6, I've done, you know, a lot of tests on that. It is a pretty cool model. It works really well. It's good.

And then if we look at coding benchmarks here, you can see that V4 Pro is crushing, right? It is a beast, right? 80.6%, 67.9%, LiveCodeBench 93.5%, GDPFileAA 1554 ELO score, right? It's beating Minimax, it's beating V4 Flash, it's beating KimiK 2.6 by a long way, right?

And so V4 Pro, if you want like the most powerful model, then you go with something like this, right? Malertz is, technically you could connect OpenClaw or Pclaw from your phone directly to post on social media for you. You could, the way that you would do that is you basically, you would use a website in between, like an API in between, right? And so you can use these APIs in between to post social media content for you.

I've got a lot of training on that inside there, apart from border. So over here, we have speed as well. Let's have a look here. V4 Flash apparently has the worst token verbosity and the highest hallucination rate.

KimiK 2.6 has the best velocity and the best hallucination rate, right? 39%, pretty low. What else you got? I don't know if those numbers are 100% correct, but there we go.

And then unique capabilities, you can see the differences here. So for example, V4 Pro, biggest open weight model ever. Yeah, that's basically it. So if you ran through all of those models, well, KimiK 2.6 is the overall winner on paper, right?

Scores 54 in the AI Intelligence Index, by highest of any open weight model, whilst generating outputs at 112 tokens per second, well above that. of average it reduced its hallucination rate as well. DeepSeek v4 pro is a coding benchmark master right so it's crushing on benchmarks for coding but it uses quite a lot of tokens so that's the downfall. I mean all of these have sort of trade-offs right so you know that's the difference here v4 flash is the cheapest minimax is pretty good for agents and then glm 5.1 is like the the best for long horizon coding tasks so it can work independently and continuously on a single task for more than eight hours autonomously which is pretty cool so those are five different models and they're the comparisons of each all right pretty cool pretty cool all right so we've covered DeepSeek v4 flash let's talk about agent zero now so a few people are asking about this so if you wanted to use agent zero with DeepSeek I'll show you exactly how so we can connect to it here let's just go to that so from here what we can actually do is we can take the quick start from here open this up open up a new terminal window and we're going to install agent zero right so you can see it's already working right here and then we're going to open this up inside docker get this set up this is running now so you can see it's setting up agent zero inside docker if you're wondering okay what is agent zero it's kind of similar to ai agents like hermes or for example something like open claw but it feels slightly different it feels like more like almost a website you're using i'll show you what i mean in a second and why that's important so you can see we've got these running right now and these are getting set up step by step so we'll just wait for those to run and then what it should do in a second is actually set up this directly with docker so you can see we're getting that set up over here let's put in the layers for it another option you got as well is that you could use this with something called space agent right which i'll come on to in a second if you've never seen space agent it's really cool it's totally different to open claw and these other ai agents so that's the main thing so if we go to agent zero we've now got that ready to go right so that's starting up here and so we're gonna hit play on that and it should be running let's just check this is working here so that's beginning to load and then we're just going to open that up right so you can if you've never set this up before basically just follow the instructions on agent zeros github if you need more help on that stuff then just check out the our profit boarding and from here if we go to the local host we've got this running right pretty nice so that was super easy and simple to set up and then what we need to do from here is actually connect deep seek right so the way that you do that is you're going to go to your settings agent config over here and you're going to change the configuration settings so that you're using deep seek okay so you might be wondering okay well how do i do that and so what you're going to do over here is you just configure the models that it's going to use right and so if we go to open router we then need to make sure that we have deep seek running with this right so right now it will be running with anthropic we need to change that how do we do that i'm going to go to something called open router if you're not familiar with open router basically it's kind of like a website with all the apis collected on them right so from here what we can actually do is we can create a new api key and we can plug that into agent zero so i'm going to do that right now we'll click on create here let's create that like so and then we're going to copy that i'm going to go to agent zero here and we need to plug in our api key all right and so you might be wondering well how do i set the api key for this i'll come on to that in a second so just to recap what you want to do is make sure that you set up go to github for agent zero follow the instructions on that page then you're going to go to docker and get that running right once you've done that then go to go to your settings over here and you're going to go to agent config and scroll down right so inside your settings agent zero you're going to go to configure api keys inside configure api keys you're going to go down to open router and plug that in hit save from there right and now configure model and you want to change this one so the main model so we're going to go to open router now type in deep sync and you've got two options here you could use v4 pro or you could use flash so let's take an example here so if we take v4 pro we can copy the model name we can go back over to agent zero we can paste our model name in here we can hit save save and just check that works sometimes doesn't always work so i just want to check it works so let's just double check it is working yes it's working boom so now we have deep seek working with agent zero right so that's another option as well agent zero so if we go back over here we can go to agent zero and start using it jacob says i don't get why using so many different cli's because everyone has a different preference everyone likes to try different things and i have promised in this video to show you how to build and automate anything and so to deliver on that i have to make sure that i show you you know every way that i know to use this so that's why i'm showing you these different ones you could just pick one right so for example you could just go okay well i like all these options but you know the one that i'm going to focus on is open code or for example i like all these things but i need an agent so i'm just going to focus on hermes and just pick one right just pick on one you know use that for a week and then go from there right so from here we're good to go uh milad says could install pica claw on iphone if you want to use agents with your iphone i would recommend doing it the right way which is really going to your terminal on your computer setting up an ai agent and then connecting it to your phone and if your if your computer is not powerful enough to run ai agents on your laptop then just use something like manas and connect manas agents to your phone as well right i'm big fan of the manas agent it's pretty good pretty reliable it works pretty much every single time right so there we go from here now milad says what would you say is the most underrated powerful free tool so out of everything i've shown you you know chat.deepseek is pretty awesome uh it's pretty awesome for doing this right so chat.deepseek is free to use i've shown you it works it builds some pretty awesome stuff so you can just go from there you know that's the option that i would recommend so from here this is finished building out as you can see and then what we're going to do from here is we're going to start going out um and start building with them jacob says yeah i get your point a nice pdf on of which of them would be good would be type of gold yeah so i actually i i have pdfs inside the ai profit boardroom if you check out the bonuses section i actually show you like how to get how to use this stuff and uh and what's best for which one right so i would recommend starting there if you're like i don't know which tool to use or how to use them etc right yeah so if we go back to agent zero we can still start building stuff so what i'm going to do from here is i'm going to grab another prompt and for example let's see what we got here let's take this one right so we're going to plug that into agent zero and it's going to use deepseek v4 pro to start building out right can i run agent zero with llama cpp local models yeah you can you can run it with local models so for example if we go to the settings whilst that's building out then we go to models and configure uh configure models is what you want to go to there we go and then you can select for example uh these different models so you can use like lm studio you can use hug in face you can use olama so local model and anything opening i compatible you can use with agent zero so so many different options right there but yeah if you want to use local models with agent zero you got lms bear in mind like deepseek itself is a local model as well but you could use lm studio you could use hug in face and you could use anything that's open ai compatible right which one would you recommend here for a beginner to start automations and autonomous agents so i just go with hermes if you want to learn agents go with hermes that is the best option i would say is agent zero better than open code does it have anything in memory so agent zero is a lot more autonomous than open code like open code is great for like small coding tasks like building our website landing page or whatever but for actually building out for example um you know if you want something with memory then what you can do is you could link open code to obsidian right so for example we go to our terminal here or let's say for example we want to link agent zero to obsidian what we can do here is we would start a new chat and then we're going to say use my local obsidian folder for your context memory all right and then we can use that directly so what this is going to do is it's going to link to our obsidian vault right and so if we open this up for example and we open up obsidian on the memories section you can see that it's going to take the local setup as you can see right here all right now we need to get the path for using this and so what we can do is we can copy the path right as you can see right here from the vault folder so we copy that path then we paste it into here as you can see the other option that we have is that we could grab a url for it or we could actually just copy and paste some memories inside right so that's basically how you can connect it and it'll do it in the background right now in the meantime you can see that this is beginning to build out here what else we've got here a snippy says this channel is a cut above the rest for all the latest ai automation for sure i'm a newbie in the field of ai the channel was poor quality uh pure quality not poor quality thanks for this channel i appreciate that thank you so much i appreciate that that's super kind what else we got here eos says i recently switched to hermes and it's night and day compared to open claw it's so nice i'm setting up with paperclip right now yeah i would agree it is so good hermes is beautiful if you haven't actually seen it um by the way this seems to be struggling so what i'm actually going to do is make sure that we have the right model here so we're going to switch that to deep seek you can see that it's hit rate limits on deep seek v4 pro and i think it's just rate limited right now because so many people are using it so i'm going to switch to deep seek v4 flash so if we go to open router we can switch to flash let's do that now there we go all right we'll copy that i'm going to change that for the lm so if we go back to our settings now and then we're going to go to models configure models and just change this to deep seek v4 flash right and then we'll say try again that should actually be a lot faster now because we're using v4 flash yeah so you can see it's now built the game so if we open that up boom look at that we just built this with deep seek and so that was super easy to build again you can use agent zero to help you with that it's pretty fun to play that actually it works really well yeah so you can see how it works right how easy it is to build i mean that literally took a couple of minutes and then if we go back here we've now got this access in our obsidian vault so that's how you can give context and memory to your ai agents what else we got here hello julian and peeps the capybara is watching what else we got here easy says what kind of usage limits you get with open router it depends what model you're using right uh but i've had no problems it's just because v4 has just come out right it's just because it's literally new so that's basically why so that's essentially how to use it i mean we've we've covered so much there it's absolutely wild right that was a full guide on how to use deep seek how to to build with it again we have a full course inside the airport for boarding on deep seek so if you type in deep seek here you can see we have a full course right here on exactly how to use it how to build anything with it etc um we've got loads of stuff on agent zero for example open claw right so if you want to get new daily tutorials like this feel free to check it inside the airport for boarding that is how to build and automate anything with deep seek exactly how it runs um exactly how it works etc you can post questions inside the community this is my ai automation community focus on helping you save time grow and scale with ai automation so if you're into that stuff um if you want to learn more if you want to connect with me personally you can do it all inside the airport for boarding with tons of training on for example obsidian and stuff like that as well inside the classroom you get access to all of my new advanced tutorials um you can see loads of training right here you get access to all this you can also meet people in your local city who are doing similar things to you so if you're working on air agents you want to connect with people who are using them too you can connect inside the map here and then also inside the calendar you can see that you can jump on three or four weekly live coaching calls where we go deep on stuff like deep seek ai agents how to use them how to set them up and also you can ask any questions or anything like that so feel free to get it all inside the ai profit boarding appreciate you watching and i'll see you on the next one cheers