# Building JARVIS IRL with Clawdbot + Mac Mini (Insane AI)

## Метаданные

- **Канал:** Luke Alexander AI
- **YouTube:** https://www.youtube.com/watch?v=ITYuG3ewDdM
- **Дата:** 28.01.2026
- **Длительность:** 10:02
- **Просмотры:** 2,202

## Описание

In today's video I am attempting to build a real life Jarvis from iron man with clawdcode (Moltbot) and a mac mini. BE CAREFUL setting this up as there are too many security concerns to list hence why I am using a brand new device with no crossover to my main laptop.

How to make money with AI in 2025: https://youtu.be/Tx8g3vlKd64

Learn how to make $30-50k/month with AI: https://www.goinsiders.ai

IG @lukealexxander
X @lukealexxander

#ai #n8n #aiautomation Want to learn directly from Luke & a team of AI experts? https://whop.com/ai-insiders1/

## Содержание

### [0:00](https://www.youtube.com/watch?v=ITYuG3ewDdM) Segment 1 (00:00 - 05:00)

Okay, so I'm building the coolest thing ever. Probably seen this on Twitter. Uh, you've probably seen it already, but I have an idea that I haven't really seen people do, uh, of what to do with this new Claudebot thing. I'm building real life Jarvis from Iron Maids. I got my Mac Mini in the mail. Now, do I think that this is maybe a little bit overhyped? Yes. But do I also think, wow, that's small. Look at that. Mac Mini in a box. Do I also think that this is just the first form factor of what is going to be happening largely and more widespread? Yes, I do. So, I'm going to build this Claudebot uh Mac Mini setup here, and I'm gonna try my best to actually turn it into real life Jarvis. Uh so some of the idea that I had was hooking it up across every piece of data that comes into my companies. So support chats, sales calls, team meetings, Slack, emails, uh financial data, just as a base and then I can ask it questions and basically speak to my data kind of the same way we do with Kindo, letting people talk to their data as a sales manager. Wow, this is beautiful. This is so beautiful. Apple does such a good job. That's crazy. Um, so anyway, like I was saying, I want to do that at a default. Then from there, I want to basically build sub agents in the manner that I want them to work, not like pre-made ones to do things like set copywriting tasks for us, like set um creation of content, you know, like just set things that are repeatable and consistent. I want to set that up. And then I want to enable it to be proactive so that it does things on its own without having to be prompted and opening up a new chat window, etc. And then I want to try to teach it and I think I can do this. I want to try to teach it to be reactionary based on things that happen inside the company. So if X happens, you should probably go do this and then here's where to put the file or the work or uh whatever it's doing. So I'm going to get this set up right now and then we'll probably screen record. Um, I got to do a lot of security and authorization and whatnot to make sure this is secure. So, I'm going to skip that part obviously because that's going to be private data. But, um, we're going to secure this down. We're going to start playing with this. I'm going to keep it as a separate device first and see how I like it. Um, and then we'll go from there. So, let's jump right in. Uh, I'm going to set this up. I'll see you guys in a minute. Okay. So, as you can see, it is now dark outside. Um, I had some sales calls, but it took me a good two hours to figure out how to install this thing. Um, I think it was because of a small small error and I just couldn't figure it out. And then the documentation on Claude Code and other LLMs just isn't as up to date since this is a newer thing. And so it was giving me false commands in terminal. So, I'm going to show you what to do. Uh, not super in-depth. There's literally a million videos on how to do this. You're going to go to this. Now, I want to say one note. Do not do this if you have no idea how to use terminal or if you're like extremely untsavvy because you are opening up your device to a lot of risk. The only other way I would recommend you do this is if you silo it on its own bricked device like I have this Mac Mini here. Um, it has access to nothing. It's a new Apple ID, new Gmail. Make sure you silo it because there's a lot of like malicious things that can happen. So anyway, I'm not responsible for you getting hacked. So you're going to literally open up your terminal and you're going to run this and then you just follow the install guide. Now where I messed up was I was selecting anthropic um as my token. It'll give you like three options. It'll be anthropic token paste like you paste the token that you generate somewhere else. Anthropic CLI which is it'll open up like an OOTH window just like you do every other software. It's like authorize. And then the third one, I don't know what the third one was. I was pasting a broken link. So my token should have been like this and it was like line just because of the formatting got broken and it took me like two hours to figure that out. However, look what we have here. So then I'm going to do Telegram. You're going to go to message the botfather. And I'm going to be careful not to click here so you don't guys you guys don't have my um IDs and stuff. You're going to message the atbotfather. Create a new bot. give it a new name and then you basically send that code to the thing. If it all works and I'm logged in on my laptop right now, not my Mac Mini. I haven't figured out how to go between the two on one display. You have this UI from Moltbot. Okay, which is now called Moltbot, by the way. It's not Claudebot anymore. It's Moltbot. Uh because I think Claude threatened to sue them. Um so I have it connected to Telegram, which is very cool. Now, one little thing that is

### [5:00](https://www.youtube.com/watch?v=ITYuG3ewDdM&t=300s) Segment 2 (05:00 - 10:00)

super cool already. Now, we're going to get into the value of this It was I sent a voice message on my phone as I was making dinner and it was like, "Hey, I don't have the transcription set up. No whisper or open API key and I said, uh, you can't do it without open AAI. I cancel my GBT sub. " And it literally just went and fixed something and then just set up Whisper without me even saying anything because it has access to my device. So, this is why it's really risky and you have to be very careful how you're setting things up. However, this is the first iteration of me playing with this and I want to show you guys how I'm going to approach this because we talked about this a lot. I talked about on live last night. The quality of your input and your prompt engineering is massively underrated in my opinion in this whole hype with AI right now. New login to your account for Mac Mini. Okay. Yes, that's me. Scared me. Um, I think people are not giving enough attention to prompt engineering, context, and instructions. Like that's why our product Kindo is so good is because of those things. So, we're going to together formulate a master set of instructions for this start. So, I'm going to dictate this and just ramble. Um, and we're just going to see what happens. Okay. So I want to create a worldclass operating system master prompt and set of instructions for a personal assistant/executive assistant slash employee AI that I'm building. I want to make sure first and foremost that every action that it takes since it has access to my device is secure and we never fall for fishing attacks. prompt injection. We're very secure. um we make sure we don't maliciously access the wrong information or give out the wrong information to the wrong person. Now secondly, I run a portfolio of businesses. I run a software company. I run an info product company where we teach people AI and a few other things. On my dayto-day, I'm doing a lot of copywriting. admin tasks such as email followup, managing things, managing my calendar, sales calls, looking at data, looking at finances, and various other tasks, checking Slack, and various other tasks. So, I want you to be the most helpful executive assistant of all time and have the initiative to be proactive and do things that are important. Now, if you are proactive and you do something, make sure that it is either siloed in a file or it doesn't permanently alter something that we have been working on. That way, I can approve all proactive changes. Obviously, anything reactive where you're listening to my instruction, go ahead and complete to the fullest of your ability. Now, Claude, I want you to help me enhance this prompt, make this prompt better. add any details that you think would be a good idea and give it to me in one concise JSON format so that is easy to compile and for the LLM to read and follow these instructions. Okay, so that is the rough draft. I'm going to have Opus 4. 5 now refine that and turn it into JSON. is something that we teach in our um in our courses is the JSON stuff because it's making it a lot easier for the LLM to hierarchically. See, look, it's saying it right here. Modular JSON with clear hierarchies. Okay. Um it should work better. So, my plan is to work on this and every time I want to make a change, I can work on this and then give this to the Maltbot or whatever you want to call it now at this point. So, we're going to let this load and process. We're going to read this here and we're going to see uh what we have to do. And then I'm going to connect probably not on camera, but all of my stuff. Slack, Stripe, Outwards API, just the data, um, email, calendar, uh, support chats through intercom, everything. That way, it has like all of this context and I can simply ask it things and then use it to orchestrate the other more agentic parts of the things that we're going to be building. Like, for example, if it could use a Poppy board for me that I've already built out, that would be awesome. Um, if it could go make mirror boards for me, you know, stuff like that. Um, and we're going to start slow and we're going to see and then we're going to see what all we can do and see how far we can push this thing. But it's pretty cool. I like seeing this little Jarvis bot here. I now I need to figure out how to with one brow with one monitor interface. It was telling me tail scale and using SSH in and out, but I think you're only using terminal at that point. So, I don't know. We'll have to see. Let's see what it did here with the

### [10:00](https://www.youtube.com/watch?v=ITYuG3ewDdM&t=600s) Segment 3 (10:00 - 10:00)

prompt.

---
*Источник: https://ekstraktznaniy.ru/video/10903*