Getting Started With OpenClaw: Step-by-Step to Your First Bot

Hey there, it's Mike Stelzner. I was just in downtown Franklin having coffee with Michael Hyatt, and we were actually talking about AI Business World and how excited he is to be attending. And he's not just going to show up and teach, he's going to be hanging out the entire time, and he's really excited to be hanging out with 1,000 other marketers who are committed to come and to learn. Because the truth of the matter is that AI is taking the marketing world by storm. And it's an incredible competitive advantage. And doesn't matter whether you are 70 years old like Michael, or in your 50s like me, or in your 20s and just getting started. I want to encourage you to come and experience the best training you will ever get in your professional career because I've recruited the absolute best of the best. It's taking place in the Anaheim Convention Center right next to Disneyland on April 29th and 30th. Grab your tickets now and commit to taking your AI marketing to the next level. Go to aibusinessworld.live and get your tickets today. I can't wait to meet you there. Welcome to the AI Explored podcast, helping you put AI to work. And now, here's your host, Michael Stelzner.

Hello.

Hello. Hello. Thank you so much for joining me for the AI Explored podcast. Brought to you by Social Media Examiner. I'm your host, Michael Stelzer, and this is the podcast for marketers, creators, and business owners who want to know how to put AI to work. Autonomous agents are here. They have actually arrived, and today we're going to explore OpenClaw, the open-source solution for creating powerful AI bots. My special guest is an AI educator who helps non-technical people use advanced AI tools His AI-focused YouTube channel is Creator Magic. His membership is Creator Magic Premium. Mike Russell, welcome back to the show. How you doing today?

It's great to be on, man. Thanks for having me back. It's exciting to be here. I can't wait to get people into OpenClaw.

I am very, very excited to talk about OpenClaw. And really, before we get into what it is, let's just talk about what are some of the biggest misconceptions when it comes to AI agents as kind of a category? What are some of the things you're seeing? Maybe that might be not necessarily true, that a lot of people believe to be true.

A couple of the biggest misconceptions about AI agents is that they can't do significant tasks. They need to be babysat along the way and they mess things up. That's simply not true. A lot has changed since they first came on the scene, maybe a year or two ago now. They've advanced significantly. And the second thing is they're not to be trusted. You know, we've seen so many stories, particularly at the start of this year, that OpenClaw is, insecure and it's going to leak all your credentials. But actually, again, that's a myth and we'll get into that, I'm sure, more later in the show.

Love it. Okay. Before we describe what the heck OpenClaw is, I'd love to kind of explore what are the benefits or upsides to utilizing something like this. Said another way, when this is done well, which we're going to teach everybody in just a few minutes, what are the benefits?

The benefits are incredible. You can literally outsource parts of your workflows, the things that you're doing every day. So if you've got a system, if you're a a marketer and you're trying to figure out how to optimize, say, ads on Meta or X ads or something like that, you can actually have OpenClaw go off and do the research and autonomously plan things out. You can even get it to automate certain parts of your social media if you so wish, obviously while keeping you as the human in the loop and ensuring quality and not AI slop along the way. And you can really take systems you already have working for you automate them. So basically it takes a lot of the grunt work out and you actually, you find yourself more productive and with more time on your hands because these agents are running around for you and doing things and reporting back to you with information like fresh reports in the morning. Like, I found this, this is a great opportunity. Let's look at this. They can be proactive. And I think this is one of the great things about OpenClaw, and that's how I'm using it.

Love it. Okay. I would love you to describe what is OpenClaw And then we're going to talk about what it can do. And if you understand a little bit of its history and how it came about and maybe who owns it now, that might be some interesting insight to kind of open up this conversation that we can get into what you can do with it.

Yeah, absolutely. So OpenClaw originally coded back in late November of 2025 by Peter Steinberger. He's got some serious tech chops behind him, but he essentially made this and then was acquired by OpenAI to come on board and grow this project as an open source project. Now he has many contributors. Growing OpenClaw every single day with new releases and updates. So it's, it's pretty fantastic how it's grown. And even Jensen Huang at the GTC conference at NVIDIA actually said this is the iPhone moment for AI. And if you are a CEO or someone in a senior position, you need to have an OpenClaw strategy. So I think it's one of the most important projects right now and potentially ever in the AI space.

Well, and it's really interesting because there has been a lot of chatter. Now, for folks listening to this podcast, you might listen to other podcasts or get email newsletters, and people were going crazy. I don't know how else to describe it, right, Mike? They were going literally nuts when this thing came out, and many of them were kind of technical. But tell us a little bit about that kind of craziness, and then let's get into what it can do. Like, you had a chance to kind of watch it and ultimately create a lot of videos about it. Like, what was the fervor? Why were so many people going nuts over this thing?

Yeah, it's a good question, Mike. And I think it's because it's the first time AI has been put very accessibly into the hands of the everyday user. So one of the biggest pluses is once you get it up and running, you can actually chat with your AI agents over apps you're already using, like Telegram, like Slack, like WhatsApp even. So this is a really fantastic way of doing things. And actually, since I've installed my own OpenClaw instance, Es with an S at the end, because I'm running many. I work with AI mainly on my phone and it's absolutely fantastic. But not only that, you can spin up multiple agents to do lots of different tasks for you. And I think this is a fantastic way of working with AI. And another key point that I want to absolutely emphasize here is that it has a phenomenal memory. It can remember things. Unlike ChatGPT or heading into the Claude text box where you're talking to a stranger every time and you have to reacquaint yourself and say, hello, I'm Mike and this is what I do and this is what you should know about me. OpenClaw knows those things and surprisingly well as well. It'll remember little small details I've told it a long time ago. Now, I'm not going to sugarcoat this because it also has moments of amnesia, and that's the kind of this is not perfect yet thing about OpenClaw, but it's pretty much the best we've had in terms of memory. It will remember things that surprise me and it will nudge me because I'm sure we'll give some use cases later on when I'm doing things in business. It reminds me of what my goals are and I even have a health OpenClaw that we can talk about that reminds me, did you do this? Did you take that supplement that you should be taking? It's incredible. It's not perfect, but it's, It's pretty good.

Okay, so let's kind of go over what we know so far. So far we know Peter Steinberger invented this thing literally like 4 months ago or something like that, right? Or, or maybe it was around earlier, but that's when it had its pop moment where it exploded. And this thing, what makes it unique, what I'm hearing you say, is that you can communicate it in the same way you communicate with a person, but you do not need to use a special interface like you do with ChatGPT, Claude, or Google Gemini. Instead, it interfaces with tools that you already use in your communication stack. Slack, Discord, WhatsApp. Well, you didn't say Discord, but I'm assuming Discord is in there. Is that correct? Discord is probably an option.

Oh yeah.

And Telegram. Okay. And a lot of people that don't use Telegram, it's kind of like WhatsApp or Messenger, and you communicate with it in a non-technical way, right? And as a result, it sounds like it's a lot like getting a text from a human. Is that correct?

It is. And it's like having an employee because you can actually, and this is where it leans into the memory, you can teach it skills. So you can teach it things you want to do and it it will remember those things as skills. So you can say, for instance, a very simple one would be, every morning when I wake up, I want you to source an inspirational quote and let me know about it. You could save that as a simplistic skill and actually also a timed job. So you can say, I want this at 8 o'clock in the morning delivered to me on WhatsApp or in my Discord channel or whatever. It's, it's really, really cool.

Can you give us an example? When we were preparing for this, you did a couple of cool things with it that I think would be really helpful for people. One of them had to do with your watch.

Yeah. So I— those of you listening, I'll describe. I'm holding up a Garmin watch here. Now, Garmin, fantastic for health tracking. I'm a stats nerd, but not so fantastic for getting the data out of the watch because they don't have a publicly accessible API. However, I just thought, well, that was one of the first tasks I gave the very first OpenClaw I set up. I started to initialize it and I said, listen, I want to improve my health as much as I can, but I don't want to spend time reading through Garmin dashboards. And I've got an Oura Ring as well. So I just said, can you just figure out how to get the data from my Garmin watch and do something with it? And it went off, it did some research, it came back and it said, yeah, I found this. And this is where it gets technical, but don't worry, you don't have to be technical. It said, I found this Python library that can access Garmin data. And I said, well, fantastic. Can you set that up for me? And this is like having this technical employee that works for you. It said, well, yes, of course I can. Da da da da da. Came back and then it said, okay, all I need from you is this information and then I can connect to your Garmin account. So I popped in my Garmin information and immediately It came back to me and it said, yes, I can see you had a terrible night's sleep last night and you need to work on your sleep patterns. And it broke everything down in fine detail. And then it said, but what you're doing really well right now is this, keep doing this. And I, I said, this is fantastic. Okay, so I've got a lot of Garmin data going back years. I said, can you go back like 5 years and give me like a, a timeline? And it went off for a while. It took a while. And this is the cool thing. It's like an employee that goes off and works. It went and did the research, pulled everything down, and it said, well, here you were doing really well and you were cycling lots. And here you went on a bit of a binge and your weight went up. And then here you, you went in and you did something else and then you moved to Cyprus and got on the Mediterranean diet. And that's probably the best thing you've done yet kind of thing. And it gave me this cool timeline and I was like, oh my goodness, this is amazing. So based on that, what should I be doing? What should I dial into? and it said, well, you've got this, this, and this you should be looking into. And then I made skills that can read all of the blood tests I've ever taken. I took a DNA test with 23andMe, fed it all into OpenClaw, and I said, right, go wild on this. Look at all of this information. And now because it's got such a good memory, every time it speaks to me, it says, based on your DNA, you shouldn't be eating that. Don't touch that. Avoid that kind of thing. Dial into this, lean into this. Based on those blood test results, we could certainly do a little bit of work over here. And it, it is just amazing and it reminds me every day, but in a really cool way. And I can say, but do you know what? I fancy this today. What, what can I do about that? And it will give me solutions. It's just crazy.

Okay. We're off on a little rabbit trail here, but I love this. I use a website called functionalhealth.com and they have DNA tests on there and they also have all the other kind of tests. And I would imagine if I had something like this, I could feed all my health data into it and it can help me understand my diet and track. It's super cool. Give us the example of what you've done with one of your social accounts, because you created a special social account. Talk to us about that.

Absolutely. And this is great. This is a fantastic business use case for Open Claw because people are like, yeah, I can use it to improve my life, but how can I practically use this in my business? Right? So I have a personal X account, but I also have an X account for my YouTube channel and for my business Creator Magic. And I just left it dormant, didn't do anything with it. So back at the end of February this year, I decided, well, I'm not doing anything with it. Let's put an OpenClaw on it and let's make it a build-in-public experiment. So not trying to fool anyone that this isn't AI. It's marked on the X account as automated, and every post now that goes out says made with AI at the bottom. So it's very clear that this is an AI experiment. I put it out, I mentioned it a few times on the YouTube channel, and basically everything about that account is automated. It's making posts. It's actually got a skill that I've taught it where it will take my YouTube videos, download them, then chop them up into clips and put them out as threads on X if I want. It will also research for the day's news and trends and post about them. And something that I've just implemented this week that I'll probably have a video on my YouTube channel about when this podcast goes out, but I haven't quite recorded the video at the time of doing this podcast right now, is I saw this great project by Andrej Karpathy, the former head of AI at Tesla, called Auto Research, where you can auto almost recursively improve something. And I took that framework and all I had to do— I didn't have to be technical. I took the link to his code that does that thing. Paste it into OpenClaw, and I said, how can we use this to recursively improve the posts that you're putting out on X, formerly Twitter? And it went off and it said, right, I can see what's going on here. Okay, we can run an experiment. We can look at how many views posts get, and then we can iterate on that experiment. And it's basically now continuously experimenting with posts and posting styles until it dials into what works for my particular account. And it's incredible. That's, that's another rabbit hole for you, but I love it.

Okay, folks, so clearly you can begin to understand that what OpenClaw has done is what we wish the other existing platforms could do. Like right now, when we work inside of Google or Claude or ChatGPT, even though we have connectors and all those other kind of things, there is kind of a limit to what it's capable of doing. And I think the secret sauce here with OpenClaw, and we're going to get into like how to begin in just a second, but I think the secret sauce is two things. Number one, it is got a communication layer that allows you and it to communicate. And we're going to see other solutions coming out with this communication layer, like Claude recently announced that they're also going to have communication layer as well. But this is one of the big advantages. And then the second thing is the fact that it can take over, for lack of better words, a computer, and it can do all the things that a human does at a computer. So these are kind of like the unlocks that I think make possible. So as you start to listen to what we're going to talk about today, just start to think about like, what are the tasks that you wish you might be able to hand off to an AI agent and literally do most of those tasks. And in some cases, all of those tasks. So with that, let's get into kind of like the groundwork because there's some things that everybody needs to understand. Full transparency, I've never used OpenClaw, so I'm going to ask all the questions that all of you wish I would have asked. Okay, so let's start with the beginning. Where do we need to begin?

So yes, you need to begin by finding a place to install OpenClaw. And the common two methods that are mentioned are find a computer, preferably not your own computer. And the reason for that is because then it, OpenClaw can access everything. So it can see every single file.

Wait, wait, wait. When you say your own computer, you mean your main computer, right?

That's a very good clarification. Not your main computer where you do all of your work on because OpenClaw can access all the files, every app on there, and it can also modify system files. So probably not a good idea until you build some trust. So if you've got a spare computer kicking around, or you've got a, like a popular thing has been the Mac Mini, if you've got a spare Mac Mini, but you don't have to spend, you know, $600 on a Mac Mini, you can spend $99 on a Raspberry Pi. It'll work on that too.

So will it even work on an old laptop or something like that, or does it have to have a certain amount of power?

Yeah, if you've got a dusty old laptop and the caveat there is, and this is as we go through the setup here, you'll have to choose an AI provider. Okay? So if you are using a very cheap or not very powerful computer, say a Raspberry Pi or a dusty old laptop, then you're gonna use OpenClaw with a cloud provider like OpenAI, Anthropic, Google Gemini, and it'll send off cloud requests and minimal compute is used. So that's fine. The reason you would want to consider spending more is if you want to run a local AI model, a model that doesn't touch the cloud and is your own AI model running on your own computer. And there are benefits to that because the cloud models sometimes go down, you get disconnected from the internet. If that's the case, you own the intelligence. So a lot of people are going that way. And some of the more recent models—

is Mistral the one everybody's using or what's, what's the one everybody's using?

Yeah. So a lot of people are using the Mistral open source models. People are also using Qwen. Qwen 3.5 has been a really popular one. I think Minimax as well was one. So a lot of those different models have been used locally, which is okay. So you can install it on a computer and it's fairly straightforward. The one technical thing you need to do is open a terminal window and type in a single line command and it will do the whole install for you. So that's local. But what I'd recommend for the complete beginner who hasn't even touched OpenClaw yet, and they're listening to this podcast right now and thinking, I want to get started, Mike, just tell me what I need to do. I would recommend while you experiment and find out what this can do, set it up on what's called a VPS. That's a virtual private server. And essentially it's a computer that runs in the cloud somewhere and it's hosted by a company. You can choose any— there's, there's so many companies that provide these virtual computers. Hetzner, DigitalOcean, Hostinger, Linode, or just do yourself a search and find the best company. I've got no affiliation or alliances with any of those companies, but a couple of them do have like one-click installers. And if you want to get really deep into it and customize your own version, you probably want to learn how to install and customize yourself. But to get started, the two companies that I know of right now, Hostinger and DigitalOcean, they have one-click installers. So you can basically sign up, create a new account, say with DigitalOcean, you can log in, add a credit card to pay. It's the, smallest computer that you set up in the cloud costs like something like $5 a month. And then you've got the ability to do all these one-click installers and they have loads of them. So do one-click install WordPress, one-click install content management system. And yes, they have a one-click install OpenClaw. So you can click that and it'll actually take you through a lot of the onboarding, sort of holding your hand as well. So if you're not technical, that's the way I'd start.

Okay. So Say again, was it Hostinger or was it Hetzner? Which one was the one that had the one— and DigitalOcean, I think, was the other one that had the one-click install. Yeah.

So the two that I know of, and I'm sure there are probably more because this is a fantastic service to offer, but the two that I know of are Hostinger and DigitalOcean.

Got it. And if we are to install it on our own computer, I would imagine there's a lot of fakes out there, right? So where do we go to actually find the code?

That's a very good question. I'm sure if you search for OpenClaw, this is actually one of the things the founder, Peter Steinberger, said. He had to change the name of it 3 times. And when they made the change to OpenClaw, they had to do it almost atomically because people were trying to snipe the name and get the domains and all of that. So you're right to, to raise that as a point. It's, it's very easy to come across a fake site. So the site is openclaw.ai.

Okay, got it. And then you have videos on your YouTube channel that kind of walk through how to actually do this, right?

Yes, absolutely. Yeah. So go visit those videos. I actually have a video installing it on a VPS in the cloud and installing it locally as well.

How long does it take to install?

So you're going to hit the install button, you're going to go off, make yourself a cup of coffee, cup of tea, let it brew for 5 minutes, come back and it'll be ready to go.

Got it. Okay, cool. You've heard me talking about AI Business World for a little while, and I know some of you that are listening probably don't have the budget to travel to Anaheim, California. And I want to let you know that AI Business World is a subset of Social Media Marketing World. And we happen to have something called a virtual ticket. And with the virtual ticket to Social Media Marketing World, you get all of the content from our AI Business World content live streamed. So we're talking more than 20 AI sessions, multiple main keynotes, including myself, but you also get the recordings of all the other content that's taking place. At Social Media Marketing World. So if you really, really deeply want to learn, but you just cannot afford to travel, grab your virtual ticket to Social Media Marketing World by going to aibusinessworld.live. Go to the register page, you will see the virtual ticket option there, and you have a chance to experience some of the best training that you'll ever experience. It'll be live and it'll be recorded. Again, aibusinessworld.live. So let's just talk about installing it on an old computer just because, right? Why? I mean, if you have an old computer sitting around, you don't have to pay any hosting fees and it's— there's still fees. We're going to get into that. But what do we do once we have it installed? By the way, is there any other advantage to hosting it in the cloud other than just not having the computer? Do they make it easier when it's in the cloud or is it just as technical? You understand what I'm saying?

When it's in the cloud, I mean, one of the biggest benefits, I suppose, to having it in the cloud is that it's always on. So uptime is pretty much 99.9%. Whereas if you host it on your computer, if you close that laptop or pull the plug from your Mac Mini, the OpenClaw instance goes dark. And that's not very good. And that actually cancels out one of the benefits that you can be on the bus or you can be on a jog and you can pull out your phone and Telegram message and your agent can get on the case for you. Right. But that's no good if your laptop went into sleep mode and it can't wake up. Right. So, okay. VPS always on. Secure, ready to go.

Okay, so we've got OpenClaw installed. Let's get to the install on a local computer or on the cloud. Like, let's assume we're using a third-party AI tool. Like, what do you recommend? What are the options? Let's talk about that just a little bit.

Yeah, it's a really good question because once you've run that installer and it goes to the onboarding, it's going to ask you a set of questions. And the first one will be, which AI provider are you going to use? So obviously the two most popular are Anthropic and OpenAI. Now, been publicly stated and actually banded around a little bit of controversy that Anthropic don't like you using their monthly subscription accounts, the Pro and the Max accounts to use OpenClaw. There's even been some speak that some accounts have been banned for doing that. So Anthropic are kind of like, please don't do that. But some people still do. And Claude Opus 4.6 currently is hands down the best model for it, in my opinion, and not just my opinion. But I've seen many other creators speak about it and say the same thing. It's been my experience. So Claude Opus, Claude Sonnet is a good backup. Opus if you're doing really complex work, but Sonnet is a good everyday driver. For instance, I'm running my X account on an OpenClaw that runs on Claude Sonnet right now. Or you can use OpenAI. So their latest model, ChatGPT 5.4, would be absolutely fine as well to get you started. And OpenAI do allow and encourage you to use their OAuth, their pay monthly pro or whatever the big $200 one is a month that you can use that to.

Okay. So for people that have cloud accounts, can they connect their existing cloud accounts or is it recommended that you set up some sort of an account where you can get API credits?

Absolutely. So yeah, you can do that. You can actually set up API access with your main cloud account if you want. If you just search for Anthropic API, if you do a search for that, it'll take you to a page where you can put in a card again to pay for API credits. So you absolutely can do it on your main account. If you're a business, that kind of definitely makes sense. If you have a business account with Anthropic, there is— yeah, there's no difference. You can still use your, your main Claude account as you would with the chatbox at Claude. But then you can also get API tokens that you can put into OpenClaw.

What do you recommend? Because it sounds like you could burn through some credits. You could probably hit limits on your regular account with something like this pretty quickly. Is that fair to say?

Yeah, it's, it's very costly, especially to use the expensive models and especially if you're doing deep work. So actually for me, I have a stack of multiple things that I'm using. So I use Claude Sonnet pretty much across the board because it's like something like 3 times cheaper than Opus. and it can do very competent work when I want to do orchestration. So when I want to set up really complex workflows and I want to really dial in, I pay for Claude Max, the $200 a month plan twice, cuz I, I use a lot of AI and I use Claude Code to do that. That's a little bit technical, but essentially I install Claude Code on the same machine as my OpenClaw is running. And then of course Claude Code can see OpenClaw. And when I want to do something, I'll talk to Claude Code. The first thing I'll say is, read the OpenClaw documentation and tell me the best practice to do thing I want to do. So I almost use Claude Code as the harness that wraps around OpenClaw and allows me to do super powerful advanced use cases.

And for my non-technical people whose eyes rolled backwards during all this, we have had a lot of other people talk about Claude Code that have been on this show, and you don't have to really get into that. But I think the key take-home here is that if you do anything of substance, you're probably going to need to set up what's called an API key in order to be able to get more out of Claude. And Claude seems to be the predominant tool that most people are utilizing for this particular application for OpenClaw. Is that fair to say?

Yeah, I've split tested both of the main models. So Claude Opus, Claude Sonnet, great personality, really easy to talk to, feels like you've got an employee there. OpenAI's model is a little bit more clinical. It doesn't really inject too much personality, but I guess that's a prompting issue, right? And that's how you set up your agent, because that's another thing, Mike, that we should get into is like, what's the first thing you say to OpenClaw once it fires up? Maybe that's the next thing we should talk about.

Well, yeah. And before we go there, can Gemini work with this too? Is that an option?

It can. So Claude has a great personality, really great to work with. Can do lots of work. It's, it's fantastic. OpenAI also very capable but less personality. Gemini I have not used myself, but from what I've heard from the content I've, I've watched and consumed, I've seen people talking about Gemini sometimes having problem executing tool calls. Now it's not an issue I've had myself, but that's obviously super important if that is an issue.

Okay, so Any other security stuff we need to talk about? We've already talked about the fact that it can connect with all these apps, and I would imagine that's pretty straightforward. Do we need to talk about either security and/or connecting in apps before we get into any other stuff here?

Yes. So in terms of security, yeah, you should definitely make sure that your OpenClaw instance is protected. And again, it's going to need a little bit of a maybe technical mind to think about that. So if it's running on your own computer at home, that's fine. That's a lot safer when it's on a VPS in the cloud. OpenClaw actually, with every update, has hardened security, and you would be surprised actually at how secure OpenClaw actually is. As long as someone doesn't hack into the VPS, the computer you're running in the cloud, you're pretty safe. But actually, you'll find most providers that I've mentioned earlier actually will allow you to run something called a firewall. Around your computer, which basically closes down all of the access doors, if you like, and only keeps the ones open that need to be. And then you authorize that with the credentials you set up. So yeah, looking into having a basic firewall would definitely be a good piece of security advice.

What about the communication apps? What's your preferred app and why?

Telegram all the way. And I, I say that as a non-Telegram user until OpenClaw came along. I was just like, why would I use this? You know, I was like the rest of us.

I've been on Telegram for a long time, folks, because the Web3 world, which I used to be part of, was active on there. And I see all my friends coming into Telegram and I message them. Let me guess, OpenClaw, right?

You're an OG, Mike.

Why don't you explain what you like about Telegram?

There you go.

Yeah.

So Telegram is fantastic. And why is it fantastic? Because if you want to— at first I thought, well, this is great. I'll stick it inside WhatsApp because I'm inside WhatsApp all the time. So that would just be another thing that I'll I'll talk to. And then I soon discovered that was a really bad idea for two reasons. First of all, if you want to set it up optimally on WhatsApp, you need a dedicated phone number for your agent, so you have to acquire one. Now, there are ways you can use things like Twilio. Do not give it access, I would definitely say, to your main WhatsApp account because you have to authorize it, and then basically it can see all of your WhatsApp messages. That's not necessarily a bad thing, but it means when someone messages you on WhatsApp, they might get a reply from OpenClaw, which should be undesirable. So with Telegram, it's fantastic. For me, it was a clean slate, but also the big thing for me was obviously WhatsApp have the KYC rules, know your customer. So you have to authorize and have a phone number and they have to know who you are. With Telegram, you can spin up multiple bots, as they call them, using the Bot Father. Mike, I know you're familiar, and you spin these up. So each agent you create on OpenClaw and OpenClaw can have multiple agents. It's not just one agent. You can have many of them. Running and doing tasks for you. And you can, you can spin them up and put them inside Telegram. And it's a lot of fun. And I actually give my, my agents personalities. So the main one that's orchestrating everything for me is called Zeus because I'm in Cyprus now, home of Greek gods. And, and I have a server maintenance mechanoid. And this is going to be a very local joke, but all of your UK listeners will get it. But you won't, Mike. I don't think you will. His name is Crichton. From Red Dwarf and he's a mechanoid. And for the UK listeners who are of a certain age, they'll go, yes, yes, Mike. Yes.

I only know Michael Crichton, the author, so it's all good. So, okay, so does Telegram cost you anything?

It's free.

Okay.

Yeah. And I think there is an upgrade path if you want, like, some premium bells and whistles, but—

okay.

You don't need to pay for it to use it with OpenGPT. No.

Okay. Before we get into how to create your very first AI agent or bot, I would like you to talk about skills a little bit because you mentioned the word skills quite a few times. And I can discern through having heard you say it and having heard lots of other people talk about skills that it is somehow a way to train up an AI on very specific tasks. But I'm curious, what do skills mean in the OpenClaw environment? And what do we need to know about skills?

Yes, exactly. So imagine you have a human employee and you want them to do a thing for you every day. That is essentially, that's a skill. So you can go ahead and you can say, okay, we're talking about humans now. Okay, human employee, every day I would like you to look through the papers that I put into your inbox, and I would like you to sort out the ones that are most important and put them in the right order. That would be a simple skill. You could translate that into the OpenClaw world by saying, okay, OpenClaw, I would like you every day to check my email, maybe via Gmail, and I'd like you to look at everything that's come in. First of all, I'd like you to discard all the noise and the spam and the pitches, and I would like you to just get the things that I need to reply to. And then I'd like you to triage them and like put the most important fires at the top and everything else down like that in a nice list. And it works incredibly well. And then what you can do, a skill can be constantly improved, just as with a human employee, right? Now, don't do it like this. Do it like that. You work with it. Essentially, you tell it what to do. You say, right, can you get into my Gmail? And it'll say, okay, finding the Gmail skill. Okay, connecting to your Gmail. It's going to ask you for some credentials to do that, and then it's going to get your email and say, right, I've got some unread emails there. Now, can you take a look at them and tell me what you think is most important? And it will come back and it'll say, right, that's clearly a pitch, that's spam, but this one is important. This email from Mike Stelzen is very important. Okay, so what should we do with it? And then you start talking. And obviously this is where OpenClaw gets to know about you as well. So on the initialization, you would actually say, well, for instance, I'm Mike from Creative Magic. I'm an AI educator of a background in audio as well. And so say it finds this email from Mike Stelzner and he's like, hey Mike, got a problem with my audio, can you help me? It'll be like, this one is very important and you also know all about this. So boom, this is going right at the top of your triage list. And this can happen every day. With new emails and new outreach, and you can save it as a skill and get it to do things. And if then you see it goes through your inbox and it messes something up and surfaces something that shouldn't be there, or hopefully you're bootstrapping it to start with and looking at what it's like marking as red and you're like, hang on, you mark that email as red, but that one's actually very important. You educate it along the way. So it's constantly improving a skill. It's brilliant.

Okay, so it sounds like you need to tell it, make a skill out of this thing or it won't do it. Is that correct?

Yes, exactly. So yeah, you'd, you'd sit down with OpenClaw, you'd basically work through how you're gonna do the thing. It does the thing, you're like, that's cool. And then you can just tell it, and this is the beauty of OpenClaw all along the way. You can say, that's brilliant. I love what you've done there. Can you save that as a skill? And it'll look at what you've done and it'll say, okay, let's call this the Gmail Triage Skill and it'll save it.

And then in the future, do you have to remind it to go to the Gmail Triage Skill or is it smart enough to know to look up that skill before it does things for you in the future?

Yeah. So most of the time it's smart enough to actually know. You can actually be quite loose depending on the how good the AI model is that you're using. But if you're using a frontier level model, You just say, hey, what are my emails doing today? I'm very lax with my agents now. I'm just like, hey, you know what's happening on X? And it will go above and beyond. It'll be, oh, I'm pulling the analytics and whoa, look, and this happened. I'm like, great, thank you. So it definitely knows about it, but I'm not going to say it's 100% perfect. Obviously these AI models are not deterministic, so they can make calls and decisions and sometimes they don't do what we think we should do. But that's the same with a human, right? So he's just like, no, no, no, I want you to look at my X analytics and tell me.

So we should tell it to make a skill after it's done something important that we anticipate we'll use again. And we should also tell it to update its skill. Or will it be smart enough to just know to update its skill? I'm just curious about that.

Yeah, I do think at this stage, maybe we'll get to the stage where that's not needed. You know, infinite context and infinite memory. That's the dream, right? But for now, I think it's a good practice just have those things baked into your head. It's going to make your life easier. Make this a skill, update the skill. And another key thing to remember is obviously the memory is great, but if something is super important to you, tell it, say, commit this to your memory. So if it's like critical to your business, like say, say you have a partnership with one company and you can never mention another company, that's probably a good idea to tell it like Commit to your memory, I don't wanna mention this other opposing company kind of thing, so.

Love it. All right, perfect. So let's talk about how to actually create our very first bot or AI agent, whatever we're calling it. I know you have a great example of one that we prepared. So this is something that everyone who's listening can go ahead and actually try on their very first installation of OpenClaw. So let's hear it.

Yeah, this is the exciting bit. So you've done the install, you've selected your AI provider, and then it pops up up with a chat window, very similar to what we've experienced before. And it says, who am I? Who are you? What are we doing here? Kind of thing. That kind of message, initial Tamagotchi moment, right? And then that's where you educate it. So I went off and I basically created an AI version of me. That's the first thing I did is I wanted to get it up to speed with everything I'm doing. And I think the more you feed OpenClaw about who you are and what you're doing, the more you're gonna get back because the more it can be proactive and think for you about what your next move should be. So basically said, who am I? Who are you? And I said, I'm Mike Russell from Creator Magic. Also got a 20-year history with Music Radio Creative. So go out on the internet, research that person, research these companies, find out as much as you can, and come back to me and tell me who you think I am and what you think I do. So that was my first prompt. Okay.

And what happened?

And so it took some time. It went off, it did some web searches, it like looked around on the internet, it found some social profiles, it found some blog posts, it probably found one of your podcasts and listened to it or found the transcript to it. And then it came back to me and it said it just gave me this summary and it was the most incredible summary. It was like so incredibly accurate. And actually, I really encourage anyone listening who sets up OpenClaw after listening to this podcast, make that your first prompt, providing you have a public image and you've got some kind of presence online. And also I would suggest to direct it. So Mike Russell, there's quite a few Mike Russells in the world, so I had to be specific about which Mike Russell I was. And it comes back, it's insanely accurate, but it also comes back with details you wouldn't expect. Like it already knew I was in Cyprus. It's like, oh, just made a move to Cyprus. Like, wow, that's really cool. And then I said, okay, so based on all this, what do you think I'll want to do and know about? And it said, well, you should probably watch some Cyprus expat news videos. You should probably set this up and you're following AI tools, AI automations. So how about we get this kind of trends list set up? And I'm like, brilliant, can you go and do it? And it started doing things. It's, it's really cool.

Okay, well, a couple of questions. What do you do if it gets some of its information about you wrong or it confuses you with another person? Did it get 100% accurate on the first go-round with you? And if it did, like, it's not going to necessarily get that with someone who has a very common name like like, like Bill Smith, you know what I mean? There's probably a million Bill Smiths in the world. So do you have any advice on how to train it to accurately understand who you are if it doesn't get it all right on the first go-round?

Yeah, that's a really good question. And yeah, that certainly can be a problem. So yeah, obviously you can tell it after it gets back to you if it's incorrect or inaccurate. You can correct it. You can say, no, that's not right, that's incorrect. You can go that path. Or if you feel like, yeah, you have a super common name or it's, it's really hard to pin down who you are, just handhold it as you would with someone you're trying to introduce yourself to. You'd say, well, you know, here's my LinkedIn, this is my YouTube channel, or this is my business website, whatever. Think about some of the places online where there is definitely some information about you and point it to those things because OpenClaw does have the ability to access the internet and look online. It's not just pulling from the AI memory. It can actually make tool calls to access the internet. So yeah, correct it and give it as much information as possible, I'd say.

Now, do you tell it to make a skill at this point, like make a skill like I am Mike Russell skill, or like talk to me a little bit about like once it's done its research on you and you feel good about it, like What do we do with that information? Tell it to commit it to memory, or what do we do with that? Right?

Right. That's an excellent question. So at its very core, OpenClaw is just a bunch of text files, or they're known as markdown files, but they're basically human-readable text files. The most popular and famous one is the soul.md, which is literally the essence of who the OpenClaw is that you're creating. So for someone like me, it would write to its soul.md file that I am an AI version of Mike Russell. He's an AI educator and does this, has a background in audio, and, and that would be the identity there. It also has another file called identity.md, and that's like, you know, how it behaves, how it talks to you. It has lots of memory files as well. And the beauty of this is if you, if you get deep into it and you actually look at the files that OpenClaw creates, you can just open these in a text editor on your computer and view them so you can check that it's accurately committed to its memory and its soul and its identity exactly what you want. And I occasionally— I wouldn't say often, but I occasionally just take a look at some of those files to see what it's remembering about me and to see what it's got committed to its memory. It's pretty interesting stuff because a lot of those core files are very critical. As to how OpenClaw behaves for you. So if you're getting, say, if you get down the line and this is where you always find issues that you need to overcome as you get further down the line, you get over the initial dopamine hit of, wow, this is amazing. I can do so many cool things with this. Once you actually get into the weeds of working with this, then you'll say, it keeps doing that thing. I wonder why. And you can open the files. Or actually what I get now, if something is not quite right with my OpenClaw, As I said before, I just put Claude Code on it. Or you could also use Claude Cowork, providing Claude Cowork can access the directory where OpenClaw is writing its files. That's a much more non-tech-friendly way of doing it. You can just say to Claude Cowork, okay, I've got OpenClaw installed. Here's the directory. Can you look at what it's done? For some reason, it keeps telling me like I'm 6'7", but I'm actually 5'4". Can you find wherever it's remembered that detail about me and correct it kind of thing.

So now that we've created kind of a virtual equivalent of ourself, does OpenClaw take over your computer browser and do all the things that you would do if you had a human sitting next to you, or is it more work with the APIs? Because I know that Gmail triage is something that you've kind of already alluded to that you've used this skill for, and I'm just curious, like, do we have to technically connect these things behind the scenes, or does it just literally open up a browser and log and you log in and it just acts as if it's you in your email? You understand what I'm saying?

That's a brilliant question, Mike, and I absolutely love it because it can do both. They can do either of those and it can also do those significantly well for non-technical users, which is the great news. So yes, it can open a browser, but that might not always be optimal. So you could say, open the web browser and navigate to my Gmail and log in and look at the emails. But that would be slow. That's gonna take multiple minutes to then look at your Gmail on the website and open each Gmail. Uh, so there are faster ways. But yes, anything that can be done on the web, OpenClaw can certainly do it. It just might be slow, but that's okay. If you've got a web task that absolutely can't be done in another way, As you mentioned, APIs or interfaces like that, you might be happy if you're on the move to say, hey, can you go to this website and do this thing and just let it go off and do it and hopefully get back and the thing is done. But when you want speed and efficiency, so say, you know, every morning I'm like, hey, what's new in my inbox? I don't want to be waiting 3 minutes for it to check my email because by that time I could have made my cup of coffee and looked at my email already and got everything I needed, right? So So that's where you'll have pre-baked skills. You can create your own skills, or you can just tell it to do things. So there are APIs and, shall we say, programmatic ways of doing things. So Google actually created a set of command line tools for Google Workspace. That's basically an API, text-based API for reading your emails. And you can tell OpenClaw to set that up. You can say, can you find the official Google command line tool for Workspace. It can access Gmail, it can look at your calendar, it can play with any of the apps inside Workspace, and it's the official Google thing. They actually only launched it, think about it, only just over a month or so ago. So it's, it's really cool. So Google are clearly, you know, leaning heavily into this. And then OpenClaw will go off and, and do the work and figure out how it works and do the thing.

Folks, we have just tapped the surface of the brain of Mike Russell. Now I'm going to ask Mike where you can discover more about him, but I am going to share he will be attending Social Media Marketing World and AI Business World. So if you happen to be there and you want to talk open claw with him, I can assure you, you will be able to find him. Now beyond that, where can people discover you? Tell us about your YouTube channel. Tell us about all the things. Where's your preferred ways if people want to directly connect with you?

Thanks, Mike. And I'm super excited to be coming to Anaheim. I can't wait. I'm gonna try and squeeze in a day in Disney as well. Nice. So if you wanna find me online, a best place is YouTube Creator Magic. That's Creator Magic. Can search my name and AI, whatever. I'm sure I'll pop up. And there's Creator Magic Premium, which is my community. And there's a group of us in there that are building exciting things with AI.

Mike, thank you so much for demystifying this concept of OpenClaw. Like, it's so exciting. Now I know a lot of people are going to go try it right away, and I cannot wait to see you soon.

Can't wait to see you. Thanks.

Hey, if you missed anything, we took all the notes for you over at socialmediaexaminer.com/a102. Be sure to follow this show on your favorite podcasting app. And if you've been a listener for a while, we would love a review on whatever platform you're listening on. And also do let your friends know about this show. You can tell Tag us on Facebook, LinkedIn, or on X, me personally. And do check out my other show, the Social Media Marketing Podcast. This brings us to the end of the AI Explored Podcast. I'm your host, Michael Stelzner. I'll be back with you next week. I hope you make the best out of your day, and may AI help you become more successful. The AI Explored Podcast is a production of Social Media Examiner. Do you want to go deeper in your understanding of AI? You've been listening to this podcast for a while, but did you know that we have a membership with lots and lots of marketers, entrepreneurs, and creators who learn every single month? Every single month we do live meetups. We have professionals coming on who teach training, and this is exclusive content only available in our AI Business Society. If you're ready to begin committing to ongoing development Join the AI Business Society right now by visiting socialmediaexaminer.com/ai.