[0:00]OpenClaw is the most powerful AI agent framework in the world right now and it's about to replace your entire workflow. While everyone else is still prompting Chat GPT one message at a time, you could be running a specialized team of AI agents that work for you 24/7. Remember everything and cost significantly less than a full-time employee to operate. The problem is, there are a lot of tutorials out there showing you how to set up OpenClaw on a Mac Mini or VPS. But nobody is showing you how to actually get it to do anything. I spent over $200 in the last 48 hours stress testing the system so you don't have to. In this video, I'm going to show you how to set up the chat model so you don't break the bank, how to deploy a self-healing team of AI agents that communicate with each other, and how to set up the infrastructure that lets your AI run on autopilot. And if we haven't met yet, my name is Duncan Rogoff. I'm a former art director for brands like Apple, PlayStation, and Nissan, and I now run a six-figure AI agency. In under 12 months, I've grown a following of over 110,000 people and I lead a community called The Build Room where more than 2,000 are already using AI to grow their audience and generate leads for their business in under three hours per week. So, focus in, close all your tabs and let's build. So, here we are inside of OpenClaw and it's not always the sexiest to look at, but today I really want to cover like the most impactful things that I've set up that actually helped me run my business. These are a couple things like communicating with it using Telegram, setting up different LLM models to handle different tasks so I can save on cost, like using cron job to just tackle tasks like a daily morning briefing or saving this to GitHub or things like that. And then I break down of how I've set up my AI agent or structure so that basically we have the orchestrator agent that communicates with all the other agents, which really allows us to be as efficient as we possibly can be. I'll probably bounce back and forth between OpenClaw and some slides I have prepared because like I said, OpenClaw is not the prettiest to look at, but I do want you to understand where everything lives. Now, when you first get into OpenClaw, it is essentially a baby. It knows nothing, so it is really up to you to configure this the right way and there's not a lot of information out there on how to do it. So there are a couple things that I've learned that have actually been really, really impactful for me. So, the very first thing that is really helpful is actually setting up different LLM models for different tasks. So, let me just show you what I mean. Can you tell me which LLM models we are using and for what tasks?
[2:22]So now, I can just communicate with OpenClaw directly here in the chat in the browser and basically it's showing me that now I've actually switched between a couple of different models depending on the complexity, right? So, for here we can say the default for everyday tasks, we're using Gemini 3 or Flash preview, which like gives me a 90% cost reduction. So when I'm just chatting back and forth with the agent itself, it's using Gemini, which is super, super cheap to use. Then from there, for any of the content sub-agents, we're using like a step up model Cloud Sonnet 4.5, which is like a really great model we're using for lead magnet and writing and our LinkedIn posts because Cloud is super, super good at writing. And then for the most expert layer, we're switching dynamically to Claude Opus 4.6, which is the latest anthropic model. It's the most expensive to use, but it's the best at complex reasoning or complex task, like coding or automation builds or any sort of multi-step reasoning. And so, basically, you can just talk to OpenClaw like it's a smart assistant and tell it to figure it out. Like you can just say to it things like, hey, I want you to set up different LLM models for different tasks, use Gemini for this, Sonnet for that, and Claude Opus for this. And then that way you can actually save on cost instead of relying on heavier and expensive models for everything. One other super valuable setup pack is switching between your Claude Max plan and the API, like dynamically and intelligently. Can you tell me how we are intelligently switching between the Claude Max and the API? So, basically there has been like a lot of chatter right now about how Claude or Anthropic is banning people who are using the Claude Max subscription in OpenClaw because it's just kind of like racking up the credits so to speak. But we basically are able to integrate a system that will dynamically switch between the two. So as you start to hit rate limits on your Claude Max plan, it'll dynamically switch over to the API when needed. So, this is essentially how it's doing it. Basically, it just uses the Claude Max plan first. Right now, I'm just paying $100 a month because I'm not ramping this up like crazy, and then once you start to kind of like hit those rate limits, it'll switch over to the API because the API does end up actually being really expensive if you're only using that. If you need to know how to get access to your Max plan on OpenClaw, literally just ask, it walked me through the setup and how to get my Cloud Max API key in like two seconds. The next is a directive that I actually got from a friend. Can you tell me about the Figure It Out directive? So, basically this is just like an instruction to OpenClaw to basically just like be smart, figure it out. If I ask you to do something, just like handle it so I don't have to have this constant like back and forth with OpenClaw to actually execute on something. You can see here, it's the core operator philosophy. Basically, I can't is not vocabulary. If I don't know something, learn it now. So this way, I'm not constantly having to like baby OpenClaw. It's just going to go out if it doesn't know how to solve a problem. It's going to research the internet and figure out how to solve something.
[5:36]It's going to search for docs or tutorials, it's going to reverse engineer stuff, it's going to look at APIs, right? It's going to try multiple approaches to actually solve the problem first before coming to me and asking me questions, that way I don't have to hold his hand through everything. So, if you want, just pause this, take a screenshot here, drop it into OpenClaw and say, hey, like this is how I want you to behave. This has actually saved me a lot of time. So, now the next thing to understand is basically the agent's infrastructure. This is kind of like having OpenClaw act as like a different team of sub-agents and this will allow you to save on costs and be as efficient as possible. Because what I was finding is that if you're just doing everything with a single agent in the main chat, it gets really bloated, really quickly. It tends to drift or hallucinate, you're constantly having to like restart the chat new. Oh, one thing that's also helpful in the chat, if you just type slash new, you can basically start a new session. It'll clear the memory or the cash and basically gets rid of all the previous context, so that way you don't have kind of that bloat and that drift happening. There is a really clear sort of orchestrator pattern that's been developed with all of these AI agents in general, where you have kind of like the orchestrator at the top level and then it speaks to all of the other agents beneath it and tells them how to communicate with each other. I put together this little guide about the ultimate guide to just multi-agent systems, like how to set these up, how to design like a sub-agent, especially for OpenClaw using the soul.md file. I'll leave a link in the description where you can just get that totally free, so you can have access to this. The most important thing inside of these agents to pay attention to is these files, so there's an agent, soul, tools file. I'm going to walk you through what each one of those is so you understand. And basically, I found that not every agent needs all of these files. Basically, you have me. I just have my Duncan agent as the main agent, and this has access to everything and then it just essentially provides all of the sub-agents with the context that they need and nothing else. So again, you're not sending extra tokens to the LLMs to process, you're only sending exactly what it needs. So, first is the soul.md file. This is the brain, this is the agent's identity and the core instructions. It defines who it is, how it behaves, and crucial behavior boundaries, so like what to do and what not to do. So, if we come in here to the Soul MD file here, this is just for my main agent, right? You are Duncan's orchestrator agent. You do not execute tasks yourself, you route them. So you're a task router and coordinator. When Duncan gives you a task, your job is to identify which sub-agent should handle the task, you should spawn that sub-agent with a clear, complete task description, and then report back the results when they arrive. So again, it's this two-way communication between the orchestrator and then the agents that live beneath it. These are the tasks that we expect it to delegate, and then these are the kind of tasks that we expect the agent to just to handle itself without having to call a sub-agent. And so, the next file to understand is this agents.md file. This is the directory of all the sub-agents that we create, and so it lists each of the sub-agents and the specific roles and their strength. So this is an example of the type of file that only your orchestrator agent needs. It basically goes ahead and it just describes each of the sub-agents. And so, basically the orchestrator needs to understand what each of the sub-agents are, like something that creates Instagram and TikTok carousels, something that writes Twitter threads, something that generates images or like basically creates lead magnets for me, right? So, it's really clear what each of the sub-agents do, so the orchestrator understands which agents to call for what tasks. But this is a really good example, like inside of like my carousel creator, there is no agents.md file because the carousel creator doesn't need to understand the agents around it because that's a job for the orchestrator. So we're not sending additional context that the sub-agent doesn't need. The next file you need to understand is this tools file, and this is just basically anything that your agent needs to execute on its task. So it doesn't need access to things like GitHub or N8N. For me, I gave the carousel creator access to like images of me so that it knows what I look like, so therefore it can pass my image to Nano Banana to generate images. And then any information about the process itself, right? Do we have a specific structure for in my case like this set of slides, right? And then here I gave it access to Notion so that it can connect to Notion and once it's done creating a carousel, it can basically save it inside of Notion for me so that I can have access to it and review it. There is a file in there called identity.md. I don't really use it too much, but if you want your agent to have like a specific identity or behave a certain way or have certain properties or characteristics, this is where you would create that information. And again, you can always tell OpenClaw to create it for itself. The user.md file is everything about you, the operator, the founder, whatever it is, right? So this is everything about you, so the agent will read this to understand like how it should communicate or how it should behave or how it should write or things like that. And the last one that is arguably one of the most important is memory.md. This is like the long-term knowledge, so this will actually get updated automatically by OpenClaw, the more you interact with the system and the more information it has about you and your processes and the way that you like to work. So we can see here that this has a ton and ton of information about me, like the technical stack that I like to use, some information about my philosophy and things like that. Who I collaborate with, all of my social proof, how I like to operate, all of these things. These are super, super detailed, but if we come down here like into the carousel, you can see I don't have the memory in here. Because I'm leaving it up to the orchestrator to decide which pieces of information should get passed to any of the sub-agents. So again, we're not over bloating the system. If you're trying to figure out how you can create all of that information easily, I have this great prompt, which is to create a data packet of information about yourself. If you've been talking with Claude or if you've been talking with Chat GPT, you would just come over to one of those LLMs and paste this in and you're just going to get basically a giant PDF full of information about your business, the projects you've worked on, your goals, your communication style and things like that. And you can just copy and paste that information into OpenClaw and say, hey, here's everything you need to know about me. What should we do with this? So setting up these agents the right way in the beginning is extremely impactful to the efficiency of your system. And it's going to save you so much headache in the long run. Before I show you the jobs that I have OpenClaw doing for me automatically, one of the first things you should do is set it up to communicate with you via Telegram or WhatsApp so that way you can take this on the go. So, I was actually out this morning and I was texting with OpenClaw and I was asking it to create that kind of AI Dream Team starter pack for all of you guys. So, basically you can see here, I just had it say, can you create one thorough master prompt that would instantly deploy and set up a team of powerful agents for anyone? Maybe like the orchestrator agent and sub agents. Don't use my agents specifically. They should be generic that can be more trained over time. What should those be? And then it said, hey, I love this idea. A starter team can deploy. Here's my recommendation for the universal agent squad. And then it basically what it went ahead and did is it created this entire document for me in Notion. I didn't write this, I didn't type this in. It figured it out, it organized this, it did all of this design, it created this prompt for me and for you guys. It's just a massive, massive time saver. So the other way I like to use this for my business is there are a couple of what are called cron jobs, and they're called cron because it has to do with, you know, chronological or time. These are basically just jobs that will run on a schedule. And this is generally like pretty ugly to look at, but you can just talk to OpenClaw and say, hey, I need something that does XYZ. One of the first things I had it build was this idea of a morning briefing. So every day when I wake up, I actually have a message inside of Telegram. It looks like this. It basically gives me yesterday's progress, like what we accomplished yesterday, what we have to prioritize today. Anything like any pending decisions, any feedback that it needs from me, and then it basically just sends us to Telegram. So you can see here that says, here's the top three today. I should post a new carousel to TikTok or Instagram. I should write a new Twitter thread. I should basically just test the setup guide end to end. Here are the tasks, these are OpenClaw's tasks. It's going to monitor Twitter for any auto-replies, and so basically if I have lead magnet posts on Twitter, it's going to recognize when somebody like comments the word AI and it's automatically going to DM them for me and send them whatever the lead magnet is. And so here, the quick decision section is basically feedback that it needs from me, like yes or no, do I want to add a hostinger affiliate link to my setup guide? Like, yeah, for sure, let's do that. Do I want to ship a Twitter thread today or do I want to wait for something else to be done, right? And then I also had it just like give me ideas, like based off the conversations we've been having, what we've been building, like here are some growth ideas to help the business grow, help the community grow, all of those good things. So, do I want to create a quality control mini-course for the build room? Do I want to turn the OpenClaw setup into a YouTube tutorial series, like we're doing right now. Do I want to cross-promote the comment content engine, which is basically like a little micro-ass app I built the other day. Pretty cool. So this will just run every day at 7:00 AM. So if there's like a structure that you want for this, just tell OpenClaw, hey, I want a daily morning briefing at 7:00 AM, send me these things. On the flip side, I had it actually create this midnight daily tracker. It's basically just going to review what we did for that day and it's just going to log everything in Notion for me. So that way I just have something to reflect back on. I can use this for content, I can use this for marketing. It's just for my kind of keeping to understand what we've done. So you can see here, we can see like here's what we accomplished like cost optimization, strategy for TikTok carousels, content briefs. Here are the key decisions that we made about like aesthetics in my brand, right? And here's what's kind of in progress or what's blocked, here all the skills and files that we created. And so again, this is just a way for me to keep track of everything that we've been building. This one's honestly a life saver. You can just create a repository on GitHub and you can just say, hey, OpenClaw, I need you to backup all of our code into GitHub every day. And so this way all of your configuration files, your memory scripts, everything is just pushed to GitHub so that way if anything breaks or you're just running into problems, you can instantly roll it back to the repository. So to a working state. If you want access to my AI Dream Team prompt library and autopilot stack, it's inside the build room. All you have to do is come to my lesson, open this up, and I give you ready to deploy, copy and paste prompts so you can set up all of these agents and cron jobs just like I have. All right, back to it. And arguably the most impactful cron job I have set to run every week, it is this weekly trend's analysis report. It's basically going to look at Reddit, YouTube and X to figure out what people are actually talking about based off of kind of some keywords, my audience, my ICP, their pain points and things like that. And so, it gives me like a top pick, like content idea number one. I gave OpenClaw full access to my consulting business. Basically, OpenClaw is the viral moment right now, which is why I'm talking about it in this video. It gives me a full execution strategy, it gives me more content ideas based off of kind of the findings from the different channels. It gives me some opportunities to create different lead magnets for my audience, so again, I can send this information to my lead magnet sub-agent to create them for me. And it's going to go ahead and rank all this content for me. It's going to break it down by different platforms like different platforms, LinkedIn posts, YouTube video, short form videos. Any like audience language that I can like kind of pull in, right? Any pain points and objections that I can really hit on so that I'm speaking directly to my audience's frustrations. And so this is probably one of the most impactful things that I have created. And then at the bottom, it just kind of links to all the sources that I pulled from. If you want to get access to any of the prompts that we use today or learn how to build a highly profitable personal brand using AI, just check the link in the description to join thousands of other people inside the build room. If you think OpenClaw is cool, just check out this video up here of 63 insane use cases that other people are doing. I'll see you over there.
