# 149 OpenClaw Masterclass for Beginners (full tutorial)

# 149 OpenClaw Masterclass for Beginners (full tutorial) Speaker 2: OpenClaw is everywhere right now, but most tutorials only show you how to install it, but they leave you stuck when something breaks. So I brought on Nick Spisak, who's been using OpenClaw since day two. He's used it more than probably 99.9% of people out there. And in this video, Nick is going to walk you through how to set it up. And more importantly, how to troubleshoot it when it breaks, as well as how to maintain your OpenClaw instance using nothing but natural language, no coding required at all. And if you stick around to the end, Nick is giving away a free skill that he built that turns any coding agent, Claude Code, Codex, you name it, into an OpenClaw expert and it's yours for free. So stick around and enjoy the full episode. Speaker 1: All right. Speaker 2: So, Nick, by the end of this video, what will somebody have learned? Like, what's the main takeaway that they're going to get? Speaker 1: Yep. So, if you watch the whole way through the video, you will have an understanding not just what OpenClaw is and how to be able to install it, but for a more practical level, how to be able to use it through using natural language. Because technology changes, but being able to actually have this be outcome-oriented is the ultimate skill to have in the AI world. Speaker 2: Perfect. Well, let's jump into your screen and check it out. And you are the go-to guy for this in my mind. You were on, we were using OpenClaw on day two. It's like, what, three months old at this point? And so you probably have used it more than 99.9% of people out there. Speaker 1: Yeah. So I think the best way to kind of go through this, right, especially from a non-technical lens, I'm going to do a screen share so that way we can basically go through this setup together. Unknown Speaker: So right here. Let's do our screen. All right. Speaker 1: So what we're going to do is we're going to use a couple different technologies for today's demonstration. And I'm going to walk you through exactly how we utilize these, and in this case, to be able to do a setup of OpenClaw. So in preparation for this, what we're using here is this is actually Claude Code. So for Claude Cowork, one of the best things to be able to do is to use it as a partner to be able to come up with visuals, diagrams, and to really think through different areas. So what I've done here is actually connected a diagramming tool known as Excalidraw. And the Excalidraw itself will go over, I used it to prompt it to say, hey, what are we going to cover today? So right now we're talking about Cowork and Excalidraw. For setting up an OpenClaw instance, we're going to have to get into what is known as a terminal interface. And I know for if you're slightly less technical, that might be something that would be a little bit more daunting the first time you see it. But the goal for today is to show that it's not really as scary as what it looks like as we go through. So we'll do a setup of how to be able to get your terminal. We're actually going to install a skill. So if you're unfamiliar with skills, you can think of skills as a end-to-end business process. So we're literally going to show a skill that we have here in OpenAI to be able to go ahead and have access to the OpenClaw infrastructure. And then finally, we're going to wrap up here at the end. We have an instance that's known as Annie, and Annie is the personal assistant that I have running on OpenClaw. So I have her off right now. So we're going to do a little mini demonstration of how you can use this to fix your personal assistant. So as technology changes, by having these primitives, now you'll be in a position to be able to go through and be able to solve any issues that you have, even if the technology changes. Speaker 2: And that's huge because so many people run into issues with OpenClaw because of the fact that it's not as cut and dry as maybe some of the other agent tools out there. So I think if you're going to be teaching people how they can troubleshoot, that's going to be super valuable. And just to clarify too, Nick, so when you were sharing your screen there, the Excalidraw, really that whole diagram that you had Claude build, that's kind of like the, almost like the step-by-step We're going to make that available to the audience, right? I think we're going to make it to where they can go in the description or the show notes and basically download that whole diagram if they want just like a step-by-step version of this video. Speaker 1: Yep, of course. Speaker 2: Awesome. Speaker 1: All right. So what we're going to do here, let's go ahead and walk through each of the steps. So I'm going to do another quick screen share. And we're going to go ahead and we're going to start with some of the documentation for the different tools that we're going to use and you're going to hear through today. So the first one, you're going to hear me talking as opposed to typing. And the technology or the tool that we're using for this is known as WhisperFlow. So it allows you to use natural language to be able to speak. And when you speak, it'll go through as part of its typing. So this will be the tool that will be how we're able to communicate with the AI. Now, the second one that we talked about, right, this Ghost TTY, this is the terminal-based experience. So, to kind of connect the dots here. OpenClaw is, generally speaking, more geared to a slightly more technical audience, where Claude Code is geared towards someone that is traditionally less technical, but wants to learn how to use AI effectively. So when we get into anything with OpenClaw, we do have to jump into the terminal for some of these actions. So I'm going to walk you through how to be able to use this in a way that's way less intimidating than what you may think. So we'll go ahead and we're going to start with that. So you're going to use either the macOS or the Linux installs for these. And basically what you would do there is you would go ahead and do just a normal download like anything else. And when you have that download, it's going to create this terminal experience here that I have up on the screen. So the first thing that we're going to do is we're going to go ahead here and I'm going to make it bigger at this point. So that way we can have a couple different screens that are there. Now when we set this up for OpenClaw or any type of development on the terminal, a couple things that we can do is we can have different screens. So there's an option here where we can actually split the screen. So we'll split it right and then we can also split it down. So basically what this is going to do is it's going to give us the ability to do multiple types of actions all at the same time. So what we'll do in this demonstration is we're going to go ahead here and we're going to use OpenAI's Codex, which is the equivalent of Claude Code, to be able to interact with the documentation. So we'll type in Codex. And we'll put that on each of the screens so that way we can interact with it. Basically, what we have here is I'm going to go ahead and I'm going to set a model. This is just saying how much reasoning to use. We're going to use the current version and we'll just use medium for this. Basically, what we're going to do here is we're going to go ahead and use a skill that we wrote and it's called OpenClaw Prime. All that means is where this skill, when it runs, is going to go out to the OpenClaw documentation. Which is over here at docs.openclaw.ai. And this is the official documentation that you can use anytime you're looking to do any use case with this type of technology. And what you're going to find here for anyone that is slightly more technical, there's a file that's called LLMs.txt. What this means is this is actually how AI agents are able to find the documentation and to be able to read it. Why this is super important? Now we're getting into a stage where we don't necessarily need to know every single detail of how a particular product works. We can go ahead and point it to the documentation and then ask our normal questions to learn more about the features, the functionalities, and the benefits that are there. Speaker 2: So that OpenClaw prime skill, is that something that you built or is that something that is available? Can anyone just go and download that and just give it to their Codex and then use it? Speaker 1: Yep, yeah, so we can put it in the show notes as well for those. So anyone can build this. This is a skill that I wrote that is set up to go ahead and just go through this one specifically, right? Look at the docs, prime. Prime is just saying, hey, I want you to go become familiar with OpenClaw information. And at that point, you're just using WhisperFlow inside of the terminal experience to say, hey, go tell me how to install OpenClaw. So now you don't necessarily need to know all the low-level technical details. You can just tell it and prompt it for what type of outcome you're looking for. Speaker 2: Wow. So that's something that you should be selling, but it sounds like you're willing to give away to the audience for free. So for the folks listening, we'll put a link to that in the description below as well. So you guys can go and just download that OpenClaw prime skill. I mean, it sounds to me, Nick, like you pretty much just upload that skill to Codex and then it can kind of walk through the OpenClaw Installation for you and also do it in a way that where it's I guess referencing all the official OpenClaw documentation so it's not just like hallucinating and making stuff up. Speaker 1: Yep, and I think that's the main takeaway, right, is there's going to be videos that tell you how to install things. There's going to be videos to tell you about new features and functionality. The goal or the outcome of this video is specifically to teach you how to fish. So build the skill that knows how to do it and then you can focus specifically on what's the outcome that you're looking to achieve. Speaker 2: Awesome. Perfect. Yeah, let's keep going. This is good stuff. Speaker 1: Okay. All right. So we'll go ahead here and put the share back up. All right, so that's a quick overview of how we can utilize the documentation. If we come back over, you can still see that it's going through its process at this point. So, it has a briefing of everything that's there for OpenClaw. I'm just going to move or close out the other windows so that way we can focus specifically on this one. Unknown Speaker: All right. Speaker 1: And we'll make this a little bit bigger so it's easier to see. So basically what it did here is it ran the skill and it searched through each of the different documents that was out on OpenClaw's website. It now gives us and shows us the core pages that it talked to and anything that it still had some follow-up questions that it wanted to know about. So right now it's picking up the fact that it knows that I have an OpenClaw instance running on this machine. So it's asking if it wants to do any work specifically on any of this infrastructure. We're not going to do that for today's exercise, but that's one of the things that the skill was already able to pick up on before we did any work that was associated with it. Speaker 2: So is that a way that you could, you're saying you could use that to basically troubleshoot your existing OpenClaw infrastructure? Speaker 1: Exactly. Speaker 2: Using that skill? Okay, I see. Speaker 1: So how we will use it is I have a Mac Mini, which is where my personal assistant is actually running, that we call Annie. So what we're going to do next as we're kind of going through some of these demos is right now Annie is currently, for this exercise, non-responsive. And we use her inside of what is known as Discord. You can set it up in Telegram. You can set it up in Discord, Slack. You basically choose the collaboration tool that works best for your business and your workflow. So what we'll do here is we'll just use WhisperFlow and natural language and see if the instance or this Codex instance is able to bring Annie back up online. All right, so a little bit of how we can set this up. We're gonna see, I might actually even use fast mode here. All right, and I'm gonna talk with WhisperFlow. So a little pro tip here, WhisperFlow has what they call a hotkey. It's just a key that you can press. So that way it'll take what you say and put it into text. I'd like you, now that you are properly trained on the OpenClaw infrastructure, We have an OpenClaw instance that's running remotely on my Mac mini. I'd like you to be able to SSH into the Mac mini and troubleshoot why Annie, our agent that is running on the OpenClaw setup, is currently non-responsive. Please be very thorough as you look through and put together a comprehensive implementation plan that we can review together. So a couple things as this is going through, just to kind of annotate what we were talking through here. So SSH is a technical term to just say, hey, we're going to log into the Mac Mini infrastructure, the Mac Mini that we were, where this OpenClaw is running. And it's going to go ahead and try to troubleshoot or figure out based on everything it learned, why that instance is currently down. The way that we prompted it here is we gave it the opportunity not just to troubleshoot, but it's going to provide a plan that we can review before it takes action. That's generally a pretty critical step when you're interacting with these AIs is you want to separate the planning functionality from the execution functionality and make sure that whatever it's going to do, it's done the way that you intend or expect it to occur. Speaker 2: Got it. That makes sense. Now, could you do the same thing using Claude Code? Like I know you're using Codex. I know there's some people out there that prefer Claude Code. I'm assuming you can do the same thing with Claude Code. Speaker 1: Yeah, exactly. So everything that we are doing right now with Codex, we can also do with Claude Code. So the nice thing about skills is skills are portable across each of the different providers that are there. So there may be slight implementation details between something that's in Claude Code versus Codex versus Gemini, but the beauty of it now in this world is all you need to do is say, hey, I have this skill that we wrote in Codex. I'd like it to be compatible with Claude Code. Please provide me an implementation plan to set it up in that type of way. Speaker 2: Got it. Okay. So yeah, they are portable across models, basically. Speaker 1: Portable across the models there. All right. So this will take a little bit of time to go through, but it does look like what we've already seen here is it's able to identify the plan. So if we take a look, let's review what the plan looks like at this point, and we'll make this slightly bigger. So we can read through it. So you can see at the top of this screen, it's basically showing the same type of experience that you would have if you were in a regular chat interface. We're just on a terminal now at this point. So it's able to go through the README file in ANI and understand a little bit about its configuration. It sees that it's running locally on the Mac Mini. There were two different areas that are up to note. So the LLM provider that, so the large language model provider that we're using is called Minimax. So that could be you have your Claude infrastructure, OpenAI. This is a model that is written by one of the Chinese providers known as Minimax, and it's utilized as a cheaper equivalent to be able to have as your OpenClaw infrastructure. So it looks like for this one specifically, it was hitting an edge case Related to some of the Discord pieces. So it's gonna give us a plan. It's gonna inspect the configuration. It's looking through just each of the different parts and then it ranks what it thinks are the root causes. All right. So we'll go ahead here. Excellent. Let's go ahead and approve this plan for implementation. All right, so it's going to continue with its triage. So it's giving me an update of what could be the potential issues that are here. And generally, this is what the experience will look like for any time you're doing troubleshooting as you're going to use, you're going to iterate back and forth with it. Speaker 2: Now, how often are you not approving the plan? Like if you ask it to put together a plan and then it comes back and gives you a plan, how often are you saying like, oh, actually, that's not what I want you to do. I want you to do X, Y, Z different. Or does it, How do you get the plan right 99% of the time? Speaker 1: It's going to depend on what I'm doing or how complex it is. For tasks that are slightly easier, I'm pretty much just reviewing it and then approving it for things that are there. For items where it has more assumptions it needs to make, I do like to iterate back and forth with it and have a conversation. And the good news with that is you can use, a pro tip, on your keyboard, you can use the shift key and tab, which allows you to enter into plan mode, both in Codex and in Claude Code. So that's a way for it to go into a mode where it has an interview experience with you and allows you to kind of remove some of those assumptions that are in play. Speaker 2: Got it. OK, so yeah, for more complex tasks, it sounds like there's more iteration back and forth during the planning phase. And is this one of the a case where. If you can get the plan right, even if it takes a few iterations, that the execution of the plan usually gets the job done. It sounds like this is one of those situations where like, what's the quote? It's like an ounce of prevention equals a pound of cure or like whatever that quote is, right? Where it's just like getting the plan right is like 90% of the battle. Is that the case here? Speaker 1: That's it. Yeah, so you'll do probably 90% of the time is in plan mode. Got it. And then just having it check its work after it gets done. There's a little tactic you can use like, hey, you are the expert on this code base or you're the expert on this tool. You just did some implementation work. What would you change differently? Speaker 2: Interesting. Okay. So asking it to almost review what it just did and suggest other ways for improvement. Speaker 1: And that gives you a way to usually just to find some considerations that you didn't even think about. Speaker 2: So while this is running, Nick, let's kind of shift gears a little bit. As soon as it's done, we'll get back to it. But what are some of the things that you're using OpenClaw for? I feel like you see a lot of content on X and just all over the internet of people talking about how big it is and how crazy the technology is. You know, it's kind of taking it down to ground level. What are some real use cases that people can get benefit out of? Speaker 1: Yep. So I put an article out at this point in time of recording. It was back in mid-March. So it's basically going over what are three different things that could apply to a business owner today. And the use case of that was really around the AI receptionist. And the idea of that is, let's take it down to like a barbershop, a roofing company as the two main items. What are the two things that they're looking through? They're trying to bring more leads in and what leads are they unable to get to? So in that scenario, having a setup for a WhatsApp, so OpenClaw allows you to Configure, we talked about Telegram, we talked about Discord as it, one of the mechanisms is WhatsApp. So if you're a business owner and you have a WhatsApp account or a meta account for business, now you could set up OpenClaw. So that way it knows based off of your business documents, what are the services that you provide and who are the type of ideal customer profile that you would like to go to. So in that roofing company example, If I'm a roofer and I'm responsible for fulfillment, how many leads am I missing when I'm on a job site? Speaker 2: Right. That's like the biggest pain point. I had somebody comment on one of my tweets last week who said she is, I think, like a masonry contractor. And that was her exact use case. She's like, I miss a lot of calls while I'm on a job. Can I set up my OpenClaw to where it essentially fields these inquiries while I'm on a job? And essentially acts as like a speed to lead function. Like, can it just auto respond on my behalf, but not just like a generic, hey, we've received your call, we'll get back to you, but auto respond in a sense that, hey, I understand you need X, Y, Z. These are the types of services that we can provide. I think we're a good fit for you. What's the time for us to get back to you from a real human? So it's kind of taking that Generic auto response and turning it into something more personalized that translates to real revenue, right? Is that kind of what I'm hearing? Speaker 1: Exactly. And think of it in this lens, right? How much does a roofing job cost? That might be a $10,000 job that you do. And now if you're the AI guy or you're the solutions guy for those type of businesses, let's say you charge for that install $2,000 as your initial setup. That's a no-brainer for a business owner if they're able to get one additional lead that turns into a job that they're able to do. Speaker 2: Yeah, and even from there, I mean, you could charge a setup fee of, what, $2,000 to $5,000 probably, and then you could, from there, charge $1,000 a month. Most people, like you mentioned, where they're selling a higher ticket service, their average ticket price is $10,000, and you help them get one extra lead per month. Even if they're paying you $1,000 a month retainer to maintain that OpenClaw instance, that's a 10x ROI if they're getting just one more customer per month. And chances are they're probably going to get more than that, especially if up to that point, they've had no system in place for getting back to leads, right? So I think that's something you can apply to your own business or you could turn around and sell as a service to other business. We've done that before. We've sold speed to lead agents. But I think OpenClaw kind of takes it to the entire next level. Speaker 1: And you can apply that to anything, right? So going back to the examples that we were learning for today. So we use the skill for OpenClaw to understand its capabilities. Now I can use it in a planning session to put it in a business lens. So at that point, what is the business that I have? What am I looking to, how am I looking to monetize it? And just ask it, hey, what are some of the features and capabilities that you have that I could take advantage of today for that roofing company example? Speaker 2: Right. Is it still working there in the terminal? Speaker 1: It is. Yeah, it's still cranking away at trying to figure out how to bring it back up. And that does take some time on occasion as we kind of go through those. Speaker 2: But I think the point to mention here is that it's the concept of anytime your OpenClaw agent goes down, you can just point your coding agent of choice, whether it's Codex or Claude Code, at your OpenClaw files, and it will fix it 100% of the time. Like, that's been my experience, too. And Nick, I know you have a technical background. You are a developer by trade. I, on the other hand, am not, have never written a line of code in my life, and I've had a 100% success rate Pointing Claude Code at my OpenClaw files and telling it to fix whatever issue I'm facing, whether anything from, hey, my agent isn't responding, to my agent is responding too slowly, to any number of other problems that I've had, I've been able to point Claude Code at my OpenClaw and it fixes it 100% of the time. Sometimes it takes five minutes, sometimes it takes an hour, but the bottom line is it always gets fixed. Has that kind of been your experience too? Speaker 1: Yeah, exactly. So that's generally what I'll do for both troubleshooting items and then anything where we're doing new features for business enablement is I'll give it a set of tasks. We saw how in the terminal you can put different windows. Generally, there's like four or five different ones that I'll have going at any point in time. Speaker 2: Awesome, that's great. Do we want to maybe in the meantime jump into your Excalidraw diagram and kind of walk through some of those steps just from a high level while it's still working there in the background? Speaker 1: All right, let's pull this back up. All right, so a couple of things that we have here. Let's talk through how we got to this point. So we started prompting it in Cowork, and then basically all I did is I just asked it like, hey, we're going to be doing a YouTube video on the topic of OpenClaw. Help build me out based off of the documentation. So that skill that we mentioned earlier, I actually used that as the initial prompt. So the initial prompt, I gave it all of the docs From OpenClaw and said, I want you to become an expert in it. Read all of these articles to become that expert. And that was the priming exercise that we went through. And I just said, all right, now at this point, now that you understand it, I had a conversation back and forth. And got to a point where it created this diagram here. So now we have, what did we cover so far today? We talked through how you can use a diagramming tool known as Excalidraw in this example and co-work to be able to help you ideate on any particular topic. We talked about how to use a terminal experience to interact with an OpenClaw instance and to be able to leverage it in a way that's using natural language through WhisperFlow. We have the OpenClaw skill itself, which makes it an expert by trade that you can then ask it specific questions around business or technical use cases. And what it's doing under the covers is reviewing the documentation programmatically, so you don't have to. And what we're currently, we're in the mix of working on is, hey, Annie's down right now. How do we bring her back up? And how do we do it in a way where we're still able to have this conversation and it's working in the background to resolve it? Speaker 2: Right. And I love how you've got each step kind of diagrammed out there too. Do you want to kind of talk through those? I think you have the example here of setting it up in Telegram just because that's the easiest platform to set it up on. Is that right? Speaker 1: Yep. So connecting it to Telegram is the easiest way to do that. And essentially how this works is there is what's known as the bot father. On Telegram, so you will message the bot father and that will give you the information to be able to plug right back into your OpenClaw setup. So that's a good way that you can go back to that terminal you had. You say, hey, I'm trying to set up Telegram right now. Walk me through exactly what I need to do to get that achieved. And it'll give you the steps and the token information that you would need to be able to accomplish that. Speaker 2: And in order for it to give you those steps, does it need to have that OpenClaw skill that you mentioned earlier? I assume so. Speaker 1: It helps to have it, but you could do it as simply as here's the URL to the docs, go read these docs, and then once it has that context, then ask the question. Speaker 2: I guess, but by giving it that skill first, you pretty much guarantee that it's going to pull from the latest information and just essentially guarantee that it's going to be accurate. So yeah, just a reminder to people, guys, if you want that skill, Nick's giving it away for free, so we'll put it In the description or in the show notes, if you're listening on a podcast platform, just click the link and you'll be able to download that skill and give it right to your Codex or your Claude Code inside your IDE of choice. I mean, that's a no-brainer there. Unknown Speaker: All right. Speaker 1: Yeah, and that's pretty much the overall setup of OpenClaw. So a couple of things that we covered on here that we haven't discussed at this point is just setting up the rest of your agent identity files, right? So a couple of things that OpenClaw will have is it's going to have a sole file and basically that's its personality. What do you want your OpenClaw infrastructure to how to act and feel and what type of skills or tools should be as a part of that? So you can give all of that information into your sole file and what is called the agent MD. You can think of that as the brain or the personality for OpenClaw and then your skills are the different hands or actions that it can take. Speaker 2: And that soul file, is that something that it comes with out of the box or is that something that you have to give to it? Like, how does that work? Speaker 1: Yep. So the soul file does come out of the box and then you have the ability to personalize it to you. So you can have the conversation with this and say, hey, this is exactly what I want you, who I want you to be and how I want you to operate. Speaker 2: Got it. Okay. Yeah, and that's something I know that I edited early on and made it just more friendly and also made it more concise and made sure that it took a side when giving me an opinion, made sure that it didn't hedge by saying like, oh, well, you could do this or you could do that. I made it to where it takes a stance and it's firm with that. So there's a lot of things you can do to further customize it for sure. Awesome. Well, Nick, this has been fantastic. Any other key points or any other data pieces you want to mention before we wrap up here? Speaker 1: No, I think that's it. I think that's the biggest thing is some of these items do take a little longer to go through. So if you have the capabilities in front of you, you can set them, set them up and let them run. And then when you're away from keyboard, go back and check on their status when they're finished. Speaker 2: Awesome. And so, Nick, before we tell people where they can find you, just a reminder, guys, if you want that XCALIDRAW diagram, which is kind of just like a step-by-step of everything that we talked about today during the video, grab that in the description along with the OpenClaw prime skill that Nick has made, where you can give that skill to Claude Code or to Codex so that it's automatically Briefed on all the OpenClaw documentation. And then from there, you can just let it run and fix your agent pretty much autonomously. That'll be in the description and the show notes as well. So Nick, where do you want to send people if they want to connect with you or learn from you or find out more about you? Speaker 1: Yep, primarily on X. So right here in the video, Nick Spisak underscore is the location to find me. And I'm usually putting out articles and different topics about AI as we're learning more on the journey. Speaker 2: Awesome. Well, thank you so much. For everybody in the audience, we will be back next week as always, and we'll talk to you soon. Thanks, Nick. Speaker 1: Thank you.