Transcription
If you're still using AI in the browser, you're doing it the slow way. You see, each of these apps has a terminal version, and they make me 10 times faster. I'm getting so much work done, and the AI companies are kind of quiet about this. They're marketing these tools to developers for code, but here's what they're not telling you. You can use them for everything, and it's way better than their apps. Writing research projects, working in the terminal is a superpower. I'm about to show you why I'm literally writing this video with these tools right now, and most people have no idea this is a thing. But I'm telling you, once you see AI in the terminal, you're never going back to the browser, so get your copy ready. We're going to my favorite place, the terminal.
But hold up, before we go there, I know what you're thinking, "Chuck, I use AI just fine." Do you tell me if this sounds like you? Because this is how I used to use AI. You're in the browser app, you're asking questions, research mode. You're diving deep into a project, can't even see your scroll bar anymore, and this is your fifth chat because ChatGPT lost its context, or it's mine. You also created a few more chats with Claude and Gemini to make sure ChatGPT wasn't lying. And yeah, you tried to copy and paste some stuff into your notes app to keep track. That never works. At this point, your project is a mess, spread over 20 chats, two deep research sessions, and scattered notes. There's a better way to do this. Hear me out, attend the terminal.
Now, we're not going to waste any time. We're diving straight into the terminal, and I know you probably have some questions. Put those in your pocket for a second, and we'll address those. But I first just want to show you what it looks like. I want you to try it so you can know it's not really scary. The terminal's a fun place, and we're going to play with Gemini CLI first. Why? Because it has a very generous free tier. That's right, you heard it, free. And I'll show you Claude Code in a second. That's my favorite by far, overpowered, but they share a lot of the same concepts. Let's just get you going. We can install it with one command.
Now, go ahead and launch a terminal. It doesn't really matter where you launch it. Mac, Windows, Linux, all these terminal apps work everywhere. Great. For me, I'm going to use Windows with WSL, or the Windows Subsystem for Linux. If you have no idea what that is, it's totally fine. I got a video right here. So I'll launch my Ubuntu terminal, zoom in so you guys can see, and we'll copy and paste this command installing Google Gemini CLI. Ready, set, go. Coffee break. While that goes, never checked our coffee. And if you run into a scary issue, run it with pseudo, just like this. And if you're on Mac, you can also use Brew. Brew install Gemini-CLI.
Now it's installed. Before we launch it, we're going to make a new directory. MK-DIR. Then we'll call it coffee-project, just like this. And let's jump in there. CD coffee-project. Now we can launch Gemini. You'll see why I did this here in a second. Type in Gemini. One word. Ready, set, go. We're here now. First, isn't that logo just awesome? I love the terminal. It's so nostalgic. Now, first thing you'll do is get logged in with your Google account. Everyone has a Google account, and yes, this can be a free, regular Gmail account. This can open your browser. Sign in, and I'm logged in.
Now, don't be scared. Go ahead, ask it a question like, "How do I make the best cup of coffee in the world?" I love the little messages that give us. Herding digital cats. It's doing a Google search right now, pretty sick. Crafting the guide, and then it responds just like we're used to. That wasn't so bad, was it? But notice some superpowered things. First of all, we got Gemini 2.5 Pro, the latest and greatest model. Also, the browser doesn't show you this: 99% context left. Every chat you have with AI has a context window. The browser hides it from you, and the terminal does not. Also, your browser can't do this. Watch. I really want you to find the best way to make coffee, research the top 10 sites, only reputable sources, and then compile the results into a document named "Best Coffee Method.md," and then create me a blog plan, just an outline. I'll do the writing. Ready, set, and go.
Alright, now it's asking us a question. "Do you want me to write a file for you?" "Do you want me to create a file for you?" "Yeah, dude, go for it." This thing can do everything a browser can do, but it has a superpower. It can access your computer. It can read and write files. Like, I'm not copying and pasting this. It's doing it for me. I mean, look, it actually made files on our computer. There they are. Best Coffee Method, coffee-blog. Think about that for a second. It can access your Obsidian vault, all your notes, because those are just files sitting there on your hard drive. It can run Bash and Python scripts. It can do mostly everything because we broke it out of the browser.
Anyways, back to the terminal. If we type in forward slash tools and hit enter, you can see all that Gemini's allowed to do. You can even add more tools. But this feature right here is what made me switch from the browser to the terminal. Watch this. Type in forward slash and it just like that. Go. What it's doing right now is something powerful. It's creating a Gemini.md file, and in the process, it analyzed our project, read our folder, read our files. And yes, go ahead, buddy. Create that file for me. What it just did there was create instructions for itself, context for what we're working on. Let's take a look at it. Cat Gemini.md. And while we didn't do much in this project, it knows. It knows what's going on. And every time you launch Gemini, it's going to load that file as its context. Like, let's test it. So we still have our Gemini session open. I'll open up another Gemini session in that same directory. This is a new conversation, fresh context, 100% left. Notice it's using our new Gemini.md file. And I'll tell it this: "Write the intro for blog post one and the coffee series." No more context, just that it should know exactly what I'm talking about. Look at that. Yep, go ahead, buddy. I didn't give it any context. It just knew. This is a new chat session. And as I work, I can just ask Gemini to update that file with my thoughts, research decisions, we made the progress of our project. And look, it added a summary to what we were working on to our Gemini.md file. I can close all this. Start up a new session. It picks up where we left off. No re-explaining the context, no starting over, no more 20 scattered chats. We just had this one file that helps keep us organized. Everything you need. You're never paralyzed again.
Now, when I saw this, I'm like, "This is it. I finally have control over my context, my files, my projects. They're not stuck in some browser chat session anymore. They're right here, sitting on my hard drive, mine, my precious." Now, I'm literally using the Gemini.md file right now for this project. This is a small example. Let me show you a big one. This is the Gemini.md file for this video. It tracks everything. It describes how I create project status, major decisions I made, even other documents should look at. Yeah, it can do that. I'll launch a whole new Gemini session in this project and I'll ask you for where we're at in the project. Are you seeing this? This has completely changed the way I create or do anything now.
And I don't just use Gemini. It's not even close to the best one. Let's look at Claude Code, my daily driver. This one's so crazy. Now, real quick, time for a coffee break and a bit of a PSA, thanks to our sponsor, Twin Gate. Now, I'm teaching you and millions of other people how to use AI and the terminal, which will also give AI access to your computer and all your files, and access to do pretty much anything. This is cool, but also really scary. Especially scary if they're a remote employee, because not only does your remote employee, or just remote person that you give access to your lab or your internal network, have access to everything over a traditional VPN, if that's what you're still using. I'm talking your servers, file servers, NAS. Because that's normally how traditional VPNs work. If someone has remote access to the network, their computer has access to everything in the network, which means their AI may have access to everything in your network. This is insane, and it's why traditional VPNs are dead, or at least should be. You should be using Twin Gate instead. Twin Gate is all about zero trust network access, meaning they zero trust people connecting to your network, and you should zero trust them too. With Twin Gate, your employee is given access to nothing unless you explicitly give it. So Beatrice here, she's on the marketing team and given access to the NAS, but only the NAS and nothing else. And definitely not access to the super secret file server where we keep all our coffee recipes. No, no, no, no, no, no. And Twin Gate takes this to the next level. This is one of my favorite features right here, trusted profiles. You can make it so that not just any device can connect to your network, and it'll do checks to see, like, "Hey, are they running Android?" Sorry, buddy, you're out of here, because iPhones are clearly better. Everyone agrees. Or if they have Windows, we can make sure hard drive encryption is required, and they have screen lock, and a firewall, and antivirus, and their OS is up to date, and we can block Windows 10. All these checks will happen before they're allowed to connect to your network. This is crazy, because it's also free for up to five users. If you're a business, it's a no-brainer. But for like a home laber, you seeing this, you can geek out with some enterprise security. It's just fun. And speaking of enterprises, if I create a trusted Windows profile, I can connect this to my CrowdStrike, Intune, Jamf One, whatever EDR or MDM I have, and have that tell me if that device is secure or if it meets my standards. Now, me and my team, we use Twin Gate every single day. If you're still using traditional VPN, you're crazy. Try out Twin Gate. You can set it up in about five minutes, and it seriously is the best, most secure way to access your home lab or your business network remotely. So go check it out, twingate.com/networkchuck, or check the link below. Thanks to Twin Gate for sponsoring this video and this channel. They help make my videos possible. I love my partnership with them. Anyways, coffee break over. Back to the video.
Now I use Claude Code, which is Claude in the terminal, for pretty much everything. It's my default, and here's why. It has a feature that changes the game: agents. Look at this. I have seven agents performing tasks right now in one terminal. Actually, there's 10. And listen, that's just one of the seven features it has that keeps me glued to the terminal. Now, Claude Code is not free, but I do have good news. If you already pay for Claude Pro, which starts at like $20 a month, you can log into the terminal with this subscription and use it. So yeah, you don't have to use API keys. And by the way, if you can only pay for one AI subscription, Claude Pro is the one I would choose, especially for the last feature I'm going to show you.
Now, let's get it installed with one command. We'll copy and paste this bad boy right here. NPM install. You get it. Now, I already have it installed. It should tell me that, or ask me to do pseudo. It yelled at me again. I'm sorry. And then we'll launch Claude very similarly to Gemini. Just type in Claude and your directory. That's it. It'll prompt you to get logged in and then ask permission to access this folder. Yes, of course. No. Again, don't be afraid. Ask Claude a question. "I need to find a NAS for my house. Here's my budget. What I want to do. Create a report for me." And just like Gemini, it does have some fun little messages for us. It'll search the web. It's got this nice TUI, terminal user interface. Oh, whoops. Our thinking is off. I can press tab to turn on thinking mode. Now we're thinking. And notice it is security first. It does ask you permission to do most things, and that's good. We can get past that here in a moment if you're feeling dangerous. I am. So we respond similar to Gemini. It can also have a context file too. If I run that same command, slash init, it will create what's called a Claude.md file. Noticing a trend here. And it's doing the same thing, looking through our project, trying to figure out what's what, and creating a context file. And also similar to Gemini, we can do something like forward slash context, which is so crazy. We can see exactly how many tokens we've used, how much we have left, and exactly what's using it. But what's funny with Claude, that doesn't really matter too much, as long as you know how to use their most powerful feature: agents.
Let's make a Claude agent right now. It's really simple. We'll do forward slash agents, just like this. We'll get a terminal menu, and let's create a "Home Lab Research Expert." And I'll just tell it that. So, create a new agent. Notice it'll ask us where do you want to make it? Because you can have agents that are tied to just this project we're working on, or personal agents that are tied to everything. You can always call them. We'll do just this project. We'll use Claude to make it. We can tell it what we want it to be. It's going to paste this in here and let it go to work. Notice you can give it access to tools or restrict access. We'll give it everything. Choose our model, Sonnet's great. Auto color. There's our agent. Press enter to save, and we'll hit escape to get out of that interface. And we just made our first agent, Home Lab Guru.
But what's the point of that? I kind of feel like we can just ask Claude to do research for us because we literally just did that. You're right, but watch this. I'm going to do a forward slash context once more so we can get our baseline. We've used 85,000 tokens. I'll give you this prompt, and I'm calling it "Home Lab Agents." It'll figure it out. I'll have it create a document, and I'll say, "Make sure you reference the research we made." I can just do an @ symbol and look at the documents in our directory. Is the NAS rec folder? There we go. Watch. It's going to use the Home Lab Guru agent. There it is. There's our agent researching for us right now.
Now, here's why this is amazing, actually kind of insane. So Claude was like, "Cool, I've got a task, but it's not for me. I'm going to delegate this task to one of my employees or one of my coworkers." And this is another Claude instance. It's like a guy sitting over there. He's like, "Hey, buddy, are you busy? Here's some work to do." He's giving him a fresh set of instructions, and get this, a fresh context window. You saw just now we have 200,000 tokens in our context window. We used 42% of it. This guy, he's got a fresh 200. That means the conversation we're having right now, me and the main Claude guy, it's protected. It doesn't get too bloated. I can give tasks to other sub-agents and never have to leave this conversation. Claude just delegated this task to a new agent. He's got a fresh pot of coffee, he's ready to go. He just walked into work. And you can create a ton of different agents for different purposes. I use them all the time to protect my context and avoid any kind of weird bias. For example, here's my YouTube outline for this video. I'll normally be working with Claude on an outline like this, and often the context window can get pretty big. Like this is where I'm at right now. And let's say at this point, I want a comprehensive review of my outline. I have a "Brutal Critic" agent. I can just deploy. Hey, let's have, I'll just do @ and call him. There he is, agent, brutal critic, review my outline. And by the way, this is a fun one. It's designed to just completely roast me according to the framework I have and the scripting goals. And it just deploys it. My conversation is still sane, but I have a new energized employee, just got a new sub coffee in hand, fresh eyes ready to go. Look at him go. And you can run multiple agents, like I showed you at the beginning of this segment. Let's go check in our Home Lab agent. Actually, he's been kind of busy. Oh, he's still going. We're going to have the best home lab ever. And by the way, we can hit control O to see what he's doing. Oh no, I interrupted him. I interrupted. Sorry. Keep going. Don't mind me.
Actually, let's open up a new window, jump into our project, which we still have coffee-project. It's okay, it'll figure it out. Launch Claude in there. He's got his Claude.md file. He knows what's going on. And let's run this prompt. We'll launch our Home Lab agent to research the best proximate servers. At the same time, do a general agent that'll just search for the best pizza place in Dallas, Texas. And then another Home Lab agent that'll find the best graphics card for gaming right now. And then put it all together in a comprehensive report. Let's go. Now that's doing its stuff right now, deploying new employees over here. I've got the Home Lab Guru working, finding the best network equipment. I feel so powerful right now. This is so fun. Yeah, they're the three agents go. Of course, we got to give it permission to search the web and stuff. A lot. Actually, I'll show you how to get past that here in a moment. Actually, I'll show you right now. I'm going to actually interrupt them. I'm going to hit control C to get out of there. Control C twice. We can run this switch and actually I'll do Claude-R so I can resume a previous conversation. Yes, you can do that. And then I'll do dash dash dangerously dash skip permissions. This is Claude without training wheels. Ready, set, go. We can choose that previous chat we had. Let's do this one here. Yes, I interrupted you. I'm sorry. Go ahead and do it again. And notice right here, bypass permissions on go. Now we're cooking, man. Look at this. Three employees, fresh cups of coffee, just going. I told you the terminal was awesome.
Let's get crazier. I'll create another tab because no one can stop us. To our project, launch Claude in dangerous mode. And what do you see? We have Claude use Gemini for research. Yes, because we can run Gemini and Claude and all these terminal tools in headless mode, meaning you don't jump into a TUI, you just run them with one command. It's like Gemini-P and then here's your prompt. Find pizza. We can just create an agent that does that. Let's make it happen. Create a new agent. I'll give it a basic prompt. "You are a research expert. You will use Gemini in headless mode to research. Use it like this." That should do it. Let's check in our Home Lab Guru. Yeah, he finished up. It's looking back at our Home Lab Guru. He used 40,000 tokens just himself, fresh employee. Let's check our context for this conversation. We only used 4,000 more tokens for that. It was 85, now we're at 88. Alright, now let's find the best AI terminal videos on YouTube. Use the Gemini research agent. And look at that. It's using Gemini to search. We're having an AI use an AI right now. There's the prompt. I want to hit control O. Let's check on our multi-agent, see how they're doing. They're still working, man. Alright, it finished up. Our agent used Gemini to find the top 10 videos.
But this next feature is tied with how awesome agents is, and I'm embarrassed to say I just found this out while making this video. Now, Claude Code, it was built for coders. It's in the name, but I hacked it, kind of. Claude Code has a feature called output styles. Look at this. Forward slash output style. We got a few defaults in here, and these essentially control the system prompt for Claude Code, how it's going to respond to you, the persona it has. The default is code, and that's what we're using, but we can change that. Let's create a new one. So I'll go back here, type in output style again, and do output style new. It's actually an agent. I didn't know that. Let's just be simple. "You are a home lab expert designed to help me create the best home lab possible." That's it. Now we can get a lot crazier than that, but you get the idea. And that's it. If I type in output style, there it is. I can adopt that right now, or when I back out and I jump back in, I think it should adopt that style by default. Yes, it does. You can create output styles that are tied to just your project, or as a whole. Whatever project you jump into, you can switch to those output styles. I'm using the output style right now to make this video. You want to see it? Script writing. Output style. And this is what it looks like. It's pretty intense. Optimize for what I do.
Now, honestly, this entire video could be about Claude. There are so many features. Like, you can do shift tab to cycle between modes. One of my favorites is planning mode. Hit shift tab. There's plan mode on. This will put a very well thought out plan together, and then you approve it, and then it just does it. Like, there it is. And no, keep planning, or yes, let's keep doing it. And there are so many things I can't dive into them. They have prompts, hooks, custom status lines. You can paste images into your terminal. But here's the thing, I don't just use Claude Code. I use Gemini. I use Claude Code, and I use Codex, which is Chat-GPT's terminal tool, all at the same time. Let me show you how. Gemini, Claude Code, Chat-GPT's Codex. I'm using all three right now to work on this video script. How two steps. First, as long as I open up Claude, Gemini, and Codex in the same directory, they're all using the same context. It's the same project. Let's do that right now. Claude, Gemini, if I spell it right, and Codex. The second thing I do is I make sure my context files are all synced up. They all say the same thing. So Gemini.md, Claude.md, and agents.md, which is what Codex uses. And they're trying to make it a standard. They're all the same. And I usually have a terminal open for each one while I'm working on a script or any kind of project. Watch this. I'll tell Claude to write a hook for this video, authority angle. Write it to authority-hook.md. I'll have Gemini write a hook on a discovery angle. Write it to discovery-hook.md. And then I'll have Codex review it. And that's normally what I do. I find Chat-GPT very good at analyzing things from a high view. Gemini and Claude are very good at the work, the deep work. And look at it go. They're all using the same context, different roles. I mean, have three different AIs working on the same thing at the same time. No copying and pasting. They can see each other's work. They're working in the same directory. That's awesome. And it's done. Looks like Claude won this one. Good job, Claude.
Now, are you seeing what's happening here? This is the craziest part about this. Everything I'm doing, talking with these three different AIs on a project, it's not tied in a browser, it's not tied in a GUI. It's just this folder right here on my hard drive. I can copy and paste that folder anywhere. All the work, all the decisions, all the context. It's mine. And that's the difference. Nothing annoys me more than when ChatGPT tries to fence me in, give me that vendor lock-in, so I can't leave. No, I reject that. I own my context. If a new greater, better AI comes out, I'm ready for it because all my stuff is right here on my hard drive. I will use all AI. I'll use the best AI. No one could stop me.
So I got a little excited there, and that's what I want you to take away here. Leading the browser, going to your terminal, puts you back in control, and it gives you better features. But now I want to get real practical. I want to show you exactly how I run a project like this. How I keep things in sync, how I keep my Claude, Gemini, and agents files in sync and work on a daily basis using a system like this. I'm going to refill my coffee. I'll see you there.
Okay, this video was made with this process. First thing I want to show you is how things are synced up, specifically my Claude file, my Gemini file, and my agents file, which is Codex. It's an agent. That's all it is. So if I get back to my terminal here, where I was showing you earlier, when we had our hooks developed, Claude made one, Gemini made one, and Codex critiqued it. I rely on Claude to run my agent that will close out everything. So I'll just normally do this when I'm done for the day. I'll go, "Hey, let's close this out. Run." And I'll mention my agent script session closer. Let me show you what that looks like. This is one of those agents I keep as a personal agent. I use it for many projects. And these agents are just files. They're files inside an agents folder. I'll cat that file here, script-session-closer. And here it is. It does a lot of stuff, but some of the key things it does. And actually, while I'm talking with you about this, I'm going to go and run the agent. We did some stuff, didn't we? I'll run the agent. But first, it'll gather everything we talked about, everything we did, and do a comprehensive summary. It'll then update a session summary file that's specific to just updating what are some things that were done in the past sessions. It'll see if any core project files need to be updated. And if I'm talking with Claude, it'll update every context file: Claude, Gemini, agents. And then this is probably my favorite part. I commit my project to a GitHub repo. So normally you would use Git for code, for writing and deploying code. I treat my scripts and pretty much every project I work on in my life like code. We commit that change, give a reason for that change so I can see a history of what I did and why I did it. Maybe something breaks, I can go back to that change and reinstate it. That's the power of using GitHub with all your ideas.
Now, this is killer for me because I'm really bad at documentation. I'm really bad at keeping track of things. But now I have this help me keep track of things. So when I'm really tired at the end of the day, and I've been working on a video and my mind is fried, I'm like, "Okay, tell you what, I'm done. Close this out." It will look through all this stuff. It will figure out where I'm at, end the project, end the day where I need to be, and then I can start fresh the next day and be like, "Hey, where are we at? What are we working on?" They can tell me, "Hey Chuck, you finished the script. It's time to record. We made these three decisions. Go for it." That's it. This is the tool I've been looking for. The syncing is probably my favorite part. No, no, actually this thing's my favorite part. How it roasts me. So I don't only use these AI terminal tools to help me create. I use them to critique me and make me better. So here are my critics. These guys are so stinking mean, and I designed them to be. They're agents. I get the brutal critic. I told 'em to be mean. So I had an issue where my AI was being way too agreeable. I'd write something and be like, "Oh, Chuck, best thing you ever wrote." I'm like, "Ah, you're gaslighting me. Stop it." I wanted something to be super mean. I want it to be hard to please so that when it did tell me I did a good job, I knew it was good. And that's what this thing does. So actually, I'm curious if my script right now is any good. I actually have not used it yet on this, which I normally don't do. It's bad. Let's see. I'll say, "Use the brutal critic to absolutely roast and verify my stuff is good." And this script, I'll just reference my script again. Everything's a file local here on my file system. I mean, you can just move so quickly and nimbly when you have a system like this. And the agent's going to go, it's going to do its thing. And again, what I love about this is while I've been talking with this AI for a minute, right? We've been going back and forth doing things. If I asked that session to review me, it would have a ton of bias coming into that based on what we've been talking about. I don't want that. I want a fresh cup of coffee, credit coming in, going, "Here's what I know Network Chuck needs to have, and I'm going to roast his current script on what he's doing." That's what I need. Notice it was really quick. Editor free stream on that. I have inside its instructions to read some framework documents I have that describe how I create and describe what my thing is, who my audience is. Again, I don't copy and paste. I got context already here. You can reference that. I love this so much. It's my favorite thing. I had to make a video about this. I wouldn't stop telling my wife about it, and she was going crazy. "Whenever he talks to me, it just goes in one ear and out the other." I tell everyone about this. I'm actually really scared about what it's going to say. Oh no. Okay, let's see what it said. Said, "When I fix it, it'll be good." What did it tell me? Oh, it's an 8.2. It's not bad. Now, I forgot I did this. The brutal critic has three personalities, or three people that come in and roast it from different angles. I forgot I did that. It's insane. Okay, I don't like this. I'm not sure I agree with this. We'll see. It says, "Segment five is a feature dump, nine features, zero depth, five to 8% retention drop, which means you'll stop watching." Let's prove it wrong. Segment five, five's coming up. We're currently in segment four. Jump there if you want to. I don't care. Tell me if it's bad. It's actually really cool. I hope. But seriously, doing stuff like this saves me hours.
Now, also, I want you to know this. It's not writing for me. I'm doing the writing because I like to keep that. I think that's important now with AI, but I do have AI roast me, help me stay on track. I get distracted. I might make some really bad creative decisions that aren't in line with what we're trying to do. Now, the point I want to hit home is that I made this for me. This is my own personal software, exactly my use case. What can you build for you that's just for you and your niche and whatever you're trying to make happen? And this is just one tool I've made. Look at the other stuff I've done. That crazy project I showed you in the beginning was running like seven agents. This is a custom-built project just with Claude Code, and I'm going to use it right after I record this. It's going to go through every file, it's going to transcribe it, read it, and give my editors notes. They need to be able to edit this video. Like, "Hey, Chuck repeated himself 17 times in the fourth video, and it's a disaster. Ignore that one." It will do that. Seriously, the tools I create are so powerful for me. I wake up every day feeling like I have superpowers. I want this for you.
Now, speaking of superpowers, let's talk about segment five. There's a tool that's actually open source. Now you can use any model you want with this open source alternative, and it might be the best tool of all of them. I'm still testing it. You also get Grok free, which is pretty sick. And a really powerful part of this is you can log in with your Claude Pro subscription and use it like Claude Code. Let's play with it. It's called Open Code. We can install it with one command. I'll have all these commands below. Installing. That was quick. And that's it. If I can just launch Open Code, just type it in like this. That does it. Oh, let me do a source because it edits to my bash RC file. You might have to do this too. Source bash RC. Now let's try it. You can open and close your terminal and do the same thing. Oh, we're here. This is Open Code. A nice TUI terminal user interface. Couple of things real quick. They launched this straight into Grok Code. Fast one. They have a deal with Grok AI that allows you to use this for free for a while. So same story. Hey, Grok Code's helping us. Even knows like, "Hey, can I help you with your coffee project?" No mask stuff. So it read the document and it can do all the stuff. Okay, cool. I'm going to exit out. We can use local models. This is the killer part. I don't think any other tool does this. But we have to edit a file. Talking about nano, and I think it's inside config, open-code. Yeah, that's the directory. And then I think the file has to be open-code.jsonc. I'll paste this in here. I'm going to use, let's do Llama 3.2. I think that's what I have installed. We'll find out. I'll save that. Let's see llama list. See what my models are. Do I have 3.2? I'll pull it down if I don't. And then when I launch Open Code again with that config in place, oh wait, what did I break? Let me fix that file real quick. There we go. I should be able to go forward slash model and change to Llama 3.2. Hey, cool. Llama works. If I back out, I can log into Claude with this command. Open Code auth login. I can choose aaro with my Claude Pro. And I'll launch this browser, paste this code in there, and now log into Claude Code. I just launch Open Code now and switch models to Claude Sonnet 4.5. And we can pick up where we left off because again, all our files are local. It doesn't matter what tool we use. It's all ours right here. I can switch models midway. Let's go back to Grok. Cool. What's our next step? Cool. While it's doing that, I can do new session. This is a new session. There's models to Claude. I can do sessions. See all the sessions I have. Why do I have so many? Oh, these are my past sessions. We can also share these sessions. Forward slash share. And watch this. This is so crazy. I think I was going to do it. Oh, the URL is copied in my clipboard. So I can go to my browser, paste that URL in, and I can, I mean, look at that. That's awesome. Wait, is it live? You can share your session with people. That's pretty neat. Okay, we can do timeline. We can jump back in time and restore. I guess if we like how far we've come in our conversation, we just go back. I want that in real life. That's awesome. We can also, if I jump out of there and do Open Code help, they've got agents. You can start a headless server. You can attach to headless servers. Export your session as JSON data. I mean, this tool is pretty cool. And the fact that you can log in and use your Claude Pro subscription, that's next level. Because otherwise, you're putting in an API key and you're paying per use, and that's a whole nightmare. I'd rather pay upfront. What's fun is I've been following these guys on Twitter before they started making this code, Open Code. This guy Dax, these guys are killing it.
So how do you feel about your browser-based GUI AI now? Pretty bad, right? Kind of feels like hammer and chisel, or chisel and hammer. I never know how to say it. Because now you can control your context. Break out of that browser, that chat window, and don't let the terminal scare you. I mean, I know it's kind of intimidating if you're not used to working in the terminal. If you can get past that, this tool is for everyone. Everyone should be using this. Make everyone use this. I'm going to make my kids use this. I don't know about my wife. I'm not sure if she's going to do it. Seriously, nothing is stopping you from trying this right now. Gemini CLI, that's free. Open Code, you can run local models if you're worried about that. And while Claude Code is paid, it's overpowered, like you saw all those features. I use that every day. Dude, just you got to try it. Dip your toe in the water. It's fine. It's awesome. You will feel like you have a superpower and build whatever you want.
Now, do you want to go even deeper? You can add tools, MCP servers to any of these terminal environments. I made a video about doing MCP servers here. You can add it to all these. Check it out. That's the video. I'll catch you guys next time. Oh, by the way, have you hacked the YouTube algorithm today? Let's make sure you do. Hit that like button, subscribe, comment, give a doc, notification bell. You got to hack YouTube today, ethically, of course. Oh, and also let me know which is your favorite terminal AI tool, or if you even care. I'm curious, do you care about this? Because I care a lot, obviously. Tell me below. Yeah, that's it for real.
Hey, you made it to the end of the video. This portion of the video, I like to reserve for praying for you. You are a person. Well, actually, I can't say that anymore. It might be AI. I don't know if you're a person. This is for you. You have a life, you've got goals, you've got things you're trying to do. You're watching this video because you want to learn something new. I want to pray for that. I want to pray for your life. This may not be your thing. Totally cool. You can click off, that's fine. So I put it at the end of the video, but if you stay, thank you. The reason I'm doing this is because I do believe in the power of prayer. The reason I do anything on this channel to help people, or all the things I'm doing, is because of Jesus. He saved me. He's empowered me to do everything I'm doing. And I don't know what your beliefs are on that. Regardless, I just want to pray for you. That's it. So let's do it real quick.
God, I thank you for the person watching this video right now. I thank you that they are excited about tech and their lives and wanting to do something cool. Otherwise, why would they be here? I pray that you would just fill them with joy about the future, that you remove any anxiety, that you give them peace, that you would bless their families and let their families be healthy and strong. And if there's anything going on in their lives right now that are really hard, that's just really hard to get through, divorce, or just any kind of sin that's really bogging them down, Jesus, I pray deliverance over them right now, that you give them a path forward, and just let them be excited about their career. I pray you go before them and make their path straight, God, that you would create opportunities for them, and that they would take these skills they're learning and do something really fun and cool with them that will provide for their families and give them a bright future. I pray this over them, God, and ultimately, I pray that all of this will lead them to maybe one day just knowing you more deeply, knowing you better, and knowing that having a relationship with you is the greatest gift, the greatest thing, that there's nothing greater you can attain apart from knowing you. Jesus, I thank you for this person, and I just thank you for this community and for letting us do this cool stuff. It's a privilege. In Jesus name I pray. Amen.
Alright, thanks guys. I'll see you in the next one.