Zero To A Full OpenClaw Setup In 26 Minutes

Key Takeaways

Here is the complete open claw setup guide that will save you hundreds of dollars, hundreds of hours, and maybe your sanity, too. This is not just like a quick start setup tutorial, okay? I'm giving you my actual complete open claw setup with multi-agent systems, a custom mission control, sustainable security and memory systems that took me weeks to figure out. I'll also explain how I combine open claw with cloud code and other agents, too. And I will be dropping exact prompts and instructions, so you can have your own autonomous, secure, and sustainable open claw doing whatever you want it to do. Mine, for example, curates news, manages my investment portfolio, and autonomously builds software throughout my entire business. Like, I literally tell it to build me something delightful that I can wake up to every morning. Open claw is honestly really magical if you use it properly, which, like, most people don't. So, without further ado, I'm going to show you how. Let's go. A portion of this video is sponsored by Bolt. >> [music] >> Okay, so the first thing you got to do is figure out what hardware your open claw is going to live on. For most people, there are three realistic options available. The first one is just an old laptop that you have lying around. This is free, and all you have to do is wipe everything on it. And no, you don't need to have super fancy hardware to get really good usage out of open claw. In fact, the setup that I'm using to show you guys today is on my old MacBook Pro that only has 16 gigs of RAM. The second option is a dedicated computer, like a Mac mini or a Mac Studio. This is an always-on machine that your open claw will be living in, and it's also the official recommendation from open claw. Mac mini will cost around $500 to $1,000, and Mac Studio around $2,000 to $7,000. So, it is a hefty price. The third option is called a VPS, or a virtual private server. This is like renting somebody else's hardware and you're running your open claw on it, and will cost you around like $5 to $20 a month. A lot of people do use this option, but personal opinion here, okay? I think you should only consider this option if you're really comfortable using the terminal and managing things yourself. It's not the most convenient. Anyways, not going to go into too much more detail about this. I'm actually going to put a summary slide on screen now that gives you more details about these different options. And I'm actually going to put a prompt in in description, which you can paste into your favorite AI chatbot, And if you answer these questions, it will help you figure out what's the best hardware for you. Final comment that I have on this, the only option that I highly, highly don't recommend that you do is installing Open Claw on your personal computer, the one that has all your data and you actually use on a daily basis. Cuz technically, Open Claw does have access to all of it, and that is where disasters could happen. Much better to have it isolated in its own computer. Cool. So, once you have your hardware sorted and booted up that computer, you want to go to openclaw.ai. And the easiest way to do this is to go to the quick start section, and there's a crow command. So, just copy this and [music] paste it into your terminal, and it should install Open Claw for you. Great. It's now installed. Yay! And once it's installed, it should kick off the onboarding wizard. If it doesn't, no worries. You can just type in open claw onboard. Click enter, click continue, and quick start. I already have an existing config detected because I already have Open Claw installed, but if you don't, then you can just follow through with the onboarding process. I'm actually going to be putting a prompt in the description, which you can paste it into your favorite AI chatbot, Claude, ChatGPT, whatever, that can help you figure out what are the best choices for you during the setup process. But, there are a few sections that I do want to go into a little bit more detail. The first of those is the model section. You will have a lot of different model providers to choose from. Now, the general rule of thumb is that if you want the best of the best and you don't want to think about cost is to go with Anthropic's Claude Opus model and the fallback model as Sonnet in case Opus fails. Be warned, this is expensive. The secondary option, if you do have a ChatGPT subscription, is to go with the GPT-5 model. This is really, really good as well. Now, say you don't want to use Opus and Sonnet, and you also don't have a ChatGPT subscription, do not worry. I have more options for you, depending on your hardware. Specifically, if you have less than 32 gigs of RAM, you can go for the Minimax M2.5 model through the API. So, it won't be like a local model that you download yourself because you don't have enough RAM for that, but you're using the hosted version to have access to this very powerful model. Mind you, it is not as good as Opus and Sonnet, but it is really, really good. Some say comparable to GPT-5, and it is a fraction of the cost. Now, if you do have a more powerful computer with more than 32 gigs of RAM, local self-hosted models become more doable for you. And I will actually put on screen some model combinations that are available depending on how much RAM that you have. And I will also put in a description a prompt you can put into your favorite AI chatbot to help you figure out what is the best model setup for you. The next section of the setup that I want to draw attention to is the communication channel setup. The communication channel is the way in which you communicate with your Open Claw, so that you can actually talk to it, and it will respond back to you. There are also a lot of options for this. The easiest option by far is Telegram. Recommend that you pick that one to set up, and after you finish the setup wizard, you can just follow the instructions to finish the configuration on Telegram, which I will show you in just a bit. But first, let's just get through the setup wizard. The next section where you have a lot of different options is the skill section. It's going to ask you what you want to include. Don't be overwhelmed by this. I put into the prompt link below that you can ask your favorite AI agent to determine which tools that you want to install first. Besides, this is just the initial installation. You can always remove skills and add skills later as well. The final question of the wizard is how do you want to hatch your bot? You can hatch in a 2i and just talk to it on terminal, but I actually prefer the web UI because it is going to launch this dashboard, which is basically your control panel for Open Claw. It has a lot of different tabs, which you'll become a lot more familiar with as you're using Open Claw more. Sort of like the central hub. I personally prefer this visual UI compared to just writing commands on terminal. Yay! You have now completed your initial setup of Open Claw. Amazing. Next up, let's actually get Telegram up and running. So, you do need to follow the instructions. You need to go to Telegram and go to BotFather and create a new bot with {slash} new bot. Got to choose a name that is not taken and then {slash} start to start it up. Copy the open claw terminal command with your Telegram user ID, paste into terminal, click enter, approve it. You can say hello to it and great, it responds now. It says I just came online. Yay! Congratulations, you can now talk to your open claw and it will respond to you. First thing it's going to do is ask you some questions about itself and you. You could type this of course, but if you are an avid user of AI chatbots, what you can do is just paste the prompt which I will put in the description below, you get the drill, into your favorite AI chatbot and ask it to come up with a very descriptive answer for this. The more descriptive, the more context, the better it's going to be. Because these answers are going to be seating the text files called markdown files. That is the essence of your open claw. This is actually really interesting and really important for you to understand how open claw actually works, which will make you a lot better at using it. Like look, this is my finder window and under open claw workspace, you can actually see these plain text called markdown files like agents, heartbeat, identity, memory, soul and tools, etc. If you open up soul.md, for example, it says it's who you are. I'm Inky an octopus. Core truth, a collaborator not assistant. Tina doesn't need a yes machine. Think with her, push back when she's wrong, challenge ideas during brainstorms. When she's heads down, keep it crisp, etc. etc. etc. Soul.md is who the agent actually is. Like it's personality, its voice, its temperament. It's like a character sheet. Then there's agents.md. It says this folder is home, treat it that way. Then it tells your engine what's the standard operating procedure. Like what it's supposed to be doing in its first run, which we already did, and for every single session start up, before doing anything else, it's supposed to read soul.md to remember who it is, user.md, which is information about the user, also in this plain text markdown file. Then there's user.md, which is everything that open claw, your agent, knows about you. There's memory.md, so this is long-term memory, important things that it needs to remember, like facts, preferences, how it should be making decisions, and etc. I'm actually going to put on screen a brief description of each of these markdown files. As you continue to work with your agents, there's going to be more and more things that go into text files. And this combination of text files is the essence, the core essence of your agent. So, if you ever want to migrate your agent to some other type of machine, if you can retain these markdown files, these text files, you can actually recreate the agent that you have. Pretty cool, right? Cool. Anyway, so on Telegram, you can paste this information about you. I got this from Claude to get started. And there you go, Inky, my Open Claw, has put into its markdown files. All right, so next thing we're going to do, I'm actually really excited to show you, because it's going to be the first thing that you actually built that is completely custom to you and such a game changer, which is the mission control. The mission control is like your central hub. It's where you can monitor what your agents are doing, assign a task, and just overall have a good understanding of what's happening. Think of it like a central monitoring system of everything that is happening with your Open Claw. So, on my mission control, what we're looking at here is the agent office. Um, you can see that these are like my little agents actually doing stuff here. I have a multi-agent system that is set up here, which I'll actually show you guys how to do that in just a little bit. But yes, you can monitor what your agents are doing. It also has the task tab, uh, which it shows all the tasks that I'm supposed to be doing, who is the owner of the task, and a continuous activity log. So, I know what Inky is doing at all times. I have a content tab, which is one of my central workflows, cuz I care about content. For you, it's going to be different. A calendar that shows me what are the scheduled task, different projects that I'm working on. Again, for you, this is going to look different. I have a memory system. So, this is where I can see the long-term memory from memory.md, and I can see the daily logs, like what my agents are writing down every single day. I have a docs tab where I have documents for everything that my Open Claw writes about, so it doesn't just like lose these files later on, cuz I like to see them, so they're real. I have a team, and this is my team of agents, which I will go into more detail later in the video. So, yes, you should definitely build your own mission control that is custom towards you. And I will put a prompt in in description that will help you build out a mission control that is like this, but custom towards you. And all you have to do is copy the results of that prompt and put it to Telegram to tell your open claw to build this out for you. And it will do this and you will have your own little mission control. I implore you to please spend some time building this out because it's going to be central to how you're going to be working with your open claw and you'll be building on top of this over time as well as you include more workflows and more projects that you're working on. All right. So next up we have a decision point. Are you ready? The first option is that you can stick with Telegram. Honestly, Telegram is fine. You're just talking to it, right? But if you really want to level up your experience, I recommend going with Discord. Okay. So here is my Discord, the Inky Claw Discord. The reason why Discord is so powerful is that you can have multiple channels. Like for me, I have different channels, my general channel, a daily digest channel where I get a brief for things that I care about, research channel, content ideas, YouTube scripts, and other projects that I have as well. Like like student success, micro SaaS, brand growth, lonely octopus, business growth, personal finance, etc. etc. And because I have a bunch of different agents, they can also alert me in the individual channels, too. So I hope you can see why this is a lot more powerful than just Telegram where it's just like one channel, right? It's not very organized and it's hard to kick off like multiple things at the same time. I highly recommend that you set up Discord if you intend to be very serious about using open claw. However, it is a pain in the ass. I'm sorry. [laughter] It is a little bit of a pain in the ass to get it working. But if you want to, what you can do is just to go on Telegram and just type in "Help me set up Discord as communication channel." Then it will give you instructions for how to do that. I'm going to put on screen the instructions. Please be a little bit patient. It might be a little finicky, but at some point I promise you it will work out. Then you'll have a Discord channel, you can go on general and it will show that Inky or whatever your opening claw name is is online and you can type "Hi." And Inky will say, "Hey." Wow. Amazing. Then you can ask Inky to set up your collection of different channels. So, all the different channels I have here, Inky actually set this up for me. I will put the prompt for this in the description. You know the drill. Note that you do need to give some permissions and fiddle around with it a little bit, but when in doubt, you can actually just ask Inky directly for help to fix the problem. Like you can paste a screenshot or just copy paste with the error messages and ask Inky to fix it for you. Okay, next up is projects. These are the things that you actually want your open clock to do. During the mission control setup, part of the prompt is also to set up the different initial projects that you want to work on. So, you can just start off by implementing one of these projects. For me, my central pipeline, my central project, is my personal digest to content pipeline. Specifically, what it looks like is that Inky gathers from a lot of different sources about the recent news [music] from Hacker News, X, Reddit, YouTube, LinkedIn, etc., etc. And it compiles together the stories that are most relevant to content that I'm creating, which for me is AI news. So, it tell me the top stories, like Anthropic launches Claude design plus Claude Opus 4.7, blah blah blah, all of these things. Then it has a topic watchlist, which I made in reference to my content that tells me about content that I may want to make a video about. So, it tracks these different subjects, like AI coding agents, local AI, AI policy and geopolitics, and also stuff that it's watching closely. This is very, very specific to my channel because I don't make content about news. I make topics in which I deep dive into specific topics. And then if there is a topic that crosses a certain threshold, Inky would then go into the content ideas channel and give me a brief concept and format about that topic. All of this is also reflected on my mission control under the content tab. So, here are my topic watchlist and here's my content ideas. So, say if I actually like an idea, I can actually click approve and it will get routed to YouTube long form or it would give me a video outline. For example, I have one called AI agent memory explained in 28 minutes and it gives me a brief outline of what I can put into that video. Of course, I actually don't just take this and make a video of it. I actually spend a ridiculous amount of time digging into this and actually creating the video out of it. But this is a really, really good start. It saves me, gosh, I don't know, like 40, 50% of the effort it usually takes me to make a video. And it's able to track these topics for me, which normally I spend hours and hours every week having to track them myself. I have other projects as well, like under my personal finance, I have an investment portfolio. So OpenClaw Inky is helping me keep track of my portfolio and recommending me stuff based upon my own investment thesis and a lot of other projects, too. So this is really hard for me to give you a starter prompt to create your own specific project because it is going to be hyper-personalized for you, right? But I'm going to give you a prompt where if you work with a chatbot, you'll be able to figure out what it is that you want to build and be able to have the correct format to put into your OpenClaw to build out the project. Another thing that you can do, what I actually do, is you can ask Inky, your OpenClaw, to build something every single day for one of your current projects. So that when you wake up in the morning, you are delighted by a new product that I built. For me, it's built resources for my boot camp students, a student success dashboard, finance dashboards, and a lot of other things as well. I will put the prompt on screen and in description that you can say to your OpenClaw to implement something like this. This is a really big game-changer for me because it makes Inky proactive as opposed to reactive. So I'm waking up every day seeing what it's building. Maybe I don't like it and I wouldn't use it, but oftentimes I actually do like it and I end up using it and I build on top of it as well. I don't have to ask it to do stuff for me. It takes initiative by itself. Makes a really big difference. I'm also going to put on screen now a bunch of different examples that you can ask your OpenClaw to start building for you if you need some ideas. As an engineer, this is a scenario that I've experienced firsthand. Someone on my team built something with an AI tool and it is amazing. The PM loves it, everybody loves it, and I just stare at it because the code is locked inside the tool. Don't get me wrong, I love it, too, but the design components don't match anything that we actually use. And to use it, I would have to rebuild everything from scratch. All that time that's potentially saved, unfortunately, I still got to pay for it with my time. That is the wall that most AI coding tools hit, which is why Bolt is the AI coding tool that I keep coming back to and telling my team to use. Because Bolt launched a design system agent that is specifically built for teams that actually need to ship. You can upload your real design system, your CSS files, your NPM packages, component docs, and Bolt builds every project using your actual components. It's not just a theme, not just some hex code and a logo. It is your real system, the same ones that your engineers are already building with. And now your whole team can build with it, too. So, these days when it's time to hand off project, you can pull the code from Bolt and it already speaks your language with the same components and the same structure. Don't need to rewrite or do some sort of design review from scratch. Saves us a lot of time and is so much smoother. If your team has been curious about AI coding tools, but keep hitting this handoff wall, this is worth looking into, seriously. You can now try Bolt for free, no credit card required. Link is in the description. Thank you so much Bolt for sponsoring this portion of the video. Now, back to the video. Great. At this point, things are set up and you also have your first project done. Amazing. Wouldn't it be sad if you lost all the progress that YOU MADE? >> NO. GOD, PLEASE, NO. NO. NO. >> YES, VERY SAD. That is why you need to now introduce version control. Specifically, you want to be tracking the changes that Open Claw is doing in its files. So, even if something does go wrong, you can always revert back to an older edition. And the easiest way to do this is to stick it on GitHub. All the files, all the code, everything onto GitHub. Simply by just asking Inky, "Hey, could you help me put everything onto GitHub?" And it will give you the instructions for how to do that. Please do this. Next up, I want to talk about multi-agent frameworks. As I mentioned earlier, the system that I have is a multi-agent framework. That's why you have Inky, who is the primary agent, the one that's talking to me on Discord. But, you also have other agents here as well. Specifically, there is Blinky, Pinky, Dinky, Linky, and Winky. Now, there are two primary reasons why you want to split into multi-agents and not just have a single agent, except for the fact that it looks cool to have multiple agents. Just kidding. The first one is that if you have a project that's getting bigger and bigger and bigger, it's better to split up that project and have multiple agents working on it. So, you can make sure that every part of the project is executed better. It's kind of like if you have only one employee and tell her to do everything in the entire company, it's going to start getting confused. There's like a lot going on, right? >> It's okay, Sponge. >> That's why we have multiple people doing different things and they all work together to have a better product. In my case, this daily digest to content workflow started getting bigger and bigger. That's why I split it up. So, other than Inky, my primary agent, there are three agents who do this pipeline. There's Blinky, who is the morning scout, who scans Reddit, Hacker News, and GitHub for the AI news each morning, scores the topics, and services the video ideas. There's Pinky, who's the research analyst, who each evening will then pick up one of the concepts and do a deep dive on it. This is what it looks like, by the way. Something I did a deep dive on, like the hidden economics of AI agents. And then there is Dinky, who's the content producer. So, when I decide on my mission control on the content, like I want it to actually be a content idea and I want a draft of the script, it is a Dinky that gives me the video outline here and the script. Now, the second reason is actually related to cost. You see, I was spending a massive amount of tokens on Inky, who is primarily using Opus. Like literally hundreds of dollars. And I said to myself, "My wallet is hurting. What to do?" So, that's why I set up multiple agents, with different agents having different types of models to save me some money. I talked about models a little bit earlier, you know, open source models, different types of models, things like that. So, I'm not going to go into too much detail about that again. But basically, in my setup, because I only have 16 gigs of RAM, right? So, I can't really run like really big open source models on here to replace Opus or Sonnet. But what I can do is for my builder agent called Linky, for example, it uses Claude Opus 4.6 for the planning part when it has to build something, cuz that part is really important. But it actually switches to Code Llama coder, which is an open source model. I have the 8 billion parameter one downloaded locally, and that is good enough to write the majority of the code. Since using this model is free, it's saving me a lot of money. One key, for example, my system monitor, it runs a health check twice a day, and this is using Minstrel 3B, which is also a very tiny local model, and it works really well. So, if you want to set up your own multi-agent system as well, I will put the prompt, as you know, into the description and paste it into your open claw and build out a multi-agent system that is suitable for you. >> [music] >> Okay, so the next two steps I'm going to talk about are going to be preventative steps, meaning that's going to make your open claw more secure and sustainable. I'm sure you've all heard stories of horror stories of people losing all their emails, freaking out, you know, leaking their passwords and stuff like that. All the things that could potentially go wrong. Or people who just like end up wasting a lot of money. Or they complain that as they keep using open claw, it starts getting worse and worse. So, please trust me, if you do these steps, you will have a much better experience and not have to deal with those things. The first one is having a security where health check. The good thing is that open claw actually has an entire page dedicated to how to secure your open claw. You can read it, of course, but what you can do is actually just copy-paste this link and put it into your open claw and ask it to run this security check twice a day. Now, in my case, I have an alert channel where it would tell me what it finds twice a day. For example, it's telling me, "Oh, like there's an open claw update that I should do, and also three security critical findings. Oh, no." And it's telling me I should run open claw security audit for more details. So, I will certainly go do that in a bit. I will put the prompt for this on screen and also link to description. And then there is memory. The biggest complaint that people have as they use open claw is that the memory starts degrading over time. Like your open claw starts forgetting things, and then the experience becomes less magical and more annoying as you keep having to remind it about things that it's supposed to know. Caveat here is that this is still not a solved problem like for the entire open claw community, but there are things that you can do to make the memory much, much better. Number one is to actually tell your open claw to write into its agents.md file and it's sold.md file that it is an aggressive note-taker. Literally part of its personality is to aggressively take notes. And it'll be taking these notes across his daily logs. For example, you know, mine is saying like all the things that it's doing every single day. And it also writes documents for everything it's doing as well. So all this is written down. So your open claw can read about it every time it forgets anything. I will put the prompt on screen and also in the description. Next up, you want to enable dreaming. To do this you can go to your open claw dashboard, remember that. Click on the dreaming tab and then toggle dreaming to be on. Dreaming is a beta feature that helps consolidate the memory that it logs in its daily logs and its different sessions. It will put them into dreams.md and promote the strongest ones into memory.md. So it becomes permanent long-term memory for your agent. The third thing you can do is what is called Karpathy memory, which is named after Andrej Karpathy, who came up with this memory system. That will basically turn all your memories into a wiki, which your open claw can go and search through to get the information that it needs. Not going to go into too much more detail about this. I am personally experimenting with this myself, too. But if you're interested, you should check it out and ask your open claw to build it out, too. I will put the prompt in the description. Yay, amazing. Great. We have pretty much covered the entire open claw system. The last section that I have is going to be a much more advanced section. So only if you're like super committed and really into open claw, like me for example. If you're just starting out, you don't need to implement these things. Maybe as you become more into it, you can come back and try out some of these tips. The first workflow that I started doing as I started doing more stuff with open claw is combining it together with Claude Code. I find that Claude Code is oftentimes better specifically for two different things. The first one it's is debugging. With open claw, it is pretty good at debugging itself when errors do occur, but I do find with Claude Code it's even better at debugging open claw. So I would just tell Claude Code to debug whatever is wrong with open claw and I usually get a better result. The second one is building major projects using Claude Code and then just asking Open Claw to write down into its logs the changes that have been made. The reason why I do this is primarily because of cost. Although I do have Linky set up so that it uses the Opus model, but it also switches to the open-source when a coder model, it still costs me substantial money when it's using that Opus model. Especially since I'm making projects like literally every single day. Now with Cloud Code, it's still using the Opus model, but since I personally have a max subscription, it's not going to be consuming additional credits and I can actually use my subscription and save those Open Claw tokens. Of course, if you don't have a Cloud Code subscription, then don't use Cloud Code. If you have an Open AI subscription, use Codex or whatever else that you have access to. Anything to save some tokens. Another thing that I started noticing as I'm using Open Claw more is that it becomes increasingly more unstable as the projects get bigger. Because a lot of these Open Claw pipelines and projects are built using text prompts that you're giving it to the agent. And you're kind of just like chaining together the results of different agents, right? But seeing as LLMs are inherently unpredictable, you end up getting different points of failure that accumulate over time. So the solution to this is after you get it working, getting your like really cool magical workflow working, you want to change as much of it as possible into code. Because code is very stable. It's stupid, but it's stable. Most of the time this really just looks like a cron job, which is a scheduled job using code. There are of course there going to be parts of your workflow that are based upon inputs and outputs for large language models across your different agents. But a lot of that workflow can be translated and stabilized using code. What I do is actually have a scheduled job every single day that translates as much of the prompt and text into code. And finally, another thing that I've been experimenting with is combining together Open Claw with other agents. You can have one Open Claw with another Open Claw or in combination with another type of agent like Cloud Co-work or Hermes. I personally really like the combination of Cloud Co-work with Open Claw. Because Cloud Co-work produced by Anthropic is also a local AI agent, but it is a lot more stable and safe. Like I would never allow Open Claw to have access to my primary emails. Cuz who knows, it might like go bananas one day and delete it, right? But I do feel more safe allowing Claude co-worker have access to my emails because I know Anthropic has much stricter guardrails. Yes, of course, you know, Anthropic is able to see my information and things like that, but it's a compromise. I'm okay with it personally. So basically, Open Claw and Claude co-work have different types of workflows that they're working on separately. And then at the end of the day, they would both give me reports combined together so I'm able to see what has been accomplished overall. I know a lot of people also like to combine Open Claw with Hermes. I don't have a verdict on that yet I'm still testing that out myself. But if you are interested, do let me know in the comments and maybe I will make a video about that, too. All right, amazing. We're at the end of this video. Congratulations. If you followed along and implemented everything I just talked about, you will now have a very complete, sustainable, and secure Open Claw system. It is not perfect, okay? But it will be very, very good and produce some pretty magical things. For those of you who haven't tried it or those of you who have only tried it a little bit, I hope this video is helpful for you to set up a much more robust system where you can see really cool results. So good luck and have fun. And let me know in the comments what you guys think. I will see you guys in the next video or live stream.