OpenClaw Full Course | Set Up and Deploy Your Own AI Agent

Hey, build a landing page for the new course and deploy it to Vercel. That

just happened while I was making coffee. What's in my inbox that actually matters? I didn't open a browser. I didn't write a prompt. I texted my agent

matters? I didn't open a browser. I didn't write a prompt. I texted my agent the same way I text someone on my team. And it handled it. This is

OpenClaw, and it changes what it means to be a developer. OpenClaw is not an AI model. It's not a chatbot. It's not just another chat GPT wrapper.

It's an orchestration layer. You pick the brain, Claude, GPT, Gemini, or a local model through Olamma, and then OpenClaw gives that brain a body that can open up a browser, write and run code, send emails, read your Google Drive, manage your calendar, deploy apps to production, all through the messaging app you already use. Telegram, WhatsApp,

Signal, whatever. A record-breaking two 200,000 developers start this repo on GitHub in a matter of weeks. Not because it's cool, but because it's useful in a way that nothing else has been so far. And it

all comes down to three things. One, full computer control. Your browser, your terminal, your email, your docs, anything you can do on a computer, it can do. Not suggest what to type, actually do it. Two, persistent memory. It

can do. Not suggest what to type, actually do it. Two, persistent memory. It

remembers last week's conversation. It knows your projects, your preferences, your decisions.

Start on Telegram in this morning, pick it up on your laptop at night, the context carries over. And three, you text it like a coworker. No app, no dashboard, you open up Telegram, you send it a message, and it responds. You

can even give it a name and a profile photo. You're not using a tool, you're delegating to an employee. Every other AI tool works like this. You go to it, you give it a task, It gives you text back, you copy-paste, you close the tab, and tomorrow, you start from zero. But OpenClaw never closes. At 8am,

before you've opened your laptop, it's already sent you a briefing. And now, the obvious question. You're giving an AI full access to your computer? Are you crazy?

question. You're giving an AI full access to your computer? Are you crazy?

Yeah, that scares people. And every other tutorial spends half the video on it. But

here's the thing: you don't run this on your personal machine. You

spin up a $5 VPS, deploy there, and now it's sandboxed. Its own isolated computer that isn't yours. And if something goes sideways, you destroy the server and start fresh. Problem solved. That setup takes about 15 minutes and it's genuinely fast. So, here's what happens in this video. First, you deploy your

own AI agent on a VPS, one click through Hostinger. Connected to Telegram, wired up to an AI model of your choice, fully running and fully yours in under 20 minutes. Then, and this is what no other video does, you actually use it. Not toy demos, but five real developer workflows. Deploying code

it. Not toy demos, but five real developer workflows. Deploying code

from your phone, commanding your inbox, running deep research, building a persistent second brain, and even creating a custom skill from scratch so your agent does something nobody else's can. By the end of this video, you'll have a 24/7 AI agent that's already

working. Not set up and ready to explore.

working. Not set up and ready to explore.

Actually doing things. Today, I already have mine running.

So grab a coffee and let me teach you how you can set up yours. Alright, before we touch anything, let

up yours. Alright, before we touch anything, let me save you some money. If you've been looking into OpenClaw, you've probably seen people buying Mac minis just to run it. Dedicated machines just sitting under their desk, running 24-7. You don't need that. OpenClaw runs on any

Linux machine. Any. A $5 a month VPS does the exact same

Linux machine. Any. A $5 a month VPS does the exact same job as a $600 Mac Mini, except it's already online. It's already

backed up, and if something breaks, you click a button and get a fresh one. That's the other thing people don't talk about: running it on a

one. That's the other thing people don't talk about: running it on a VPS isn't just cheaper, it's safer because your agent gets its own isolated machine. It can touch your personal files, your passwords, your browser history. It's

machine. It can touch your personal files, your passwords, your browser history. It's

sandboxed by default. That security concern everyone spends 40 minutes on, a VPS solves it architecturally. Now, you can do this on any VPS provider, but I'm using Hostinger because they have a one-click open-claw template, meaning you skip all the manual Docker setup, the SSH config, and the dependency

installs. You click a button, it deploys, and you're configuring your agents in minutes instead

installs. You click a button, it deploys, and you're configuring your agents in minutes instead of hours. So, let's do it. First, head over to Hostinger by clicking a

of hours. So, let's do it. First, head over to Hostinger by clicking a special link down in the description that'll give you a big discount.

You'll want a VPS plan. For OpenClaw, I would recommend the KVM2 plan.

It has two CPU cores, 8GB of RAM, 100GB of disk space, and 8TB bandwidth.

This will allow our OpenClaw instance to do whatever it wants securely within its own computer. But instead of choosing a plan right here, we'll head over to the

computer. But instead of choosing a plan right here, we'll head over to the one-click OpenClaw deployment. So, go ahead and click Choose Plan, and now you already know which one we want to go for. Once you're here, choose the 24-month plan as it gives you the biggest discount. And with it, you're also getting a free domain name for one year. Then if you scroll

down, you'll be able to choose a location of your server. Go with the region that is closest to you for lowest latency. And there are some upgrades such as daily auto backup, as well as ready to use AI, which is going to give you some instant credits so you can use whatever AI agent you want out of the box. If you want the simplest possible setup ever, you can

go ahead and get the credits right here. But since I know that most of you will want to connect your own favorite LLM, such as ChatGPT, Clod, or Gemini, I'm going to leave this option unticked for now. But in that case, remember that you'll have to have your own API keys for those services.

and then let's go ahead and enter a special coupon code javascript mastery that's going to give you an even bigger discount then click continue and after making a payment you'll be redirected to a quick open clock configuration right before that one click deploy button if you haven't purchased those additional hostinger ai tokens you'll have to

configure one of the ai models you want to use There's Entropic for their cloud models, OpenAI for GPT models, or Google for Gemini models. And there's also Grok. You

most likely already use one of these platforms, but maybe you use it as a chat. Now you'll need to get the API key. The process for getting a key for each one of these platforms differs a bit, but more or less, it's very similar. First, you can choose which one you want to use, and then you can head over to the link below. Once you sign in, you'll

have to head over to API keys, and create a new secret key.

Then you'll be able to copy it and paste it right here. In

this case, I'll be using Claude, so I'll head over to their console, create my account, and then head over to API Keys where you can create a new key.

You can call it something like open claw and click add. Then you'll be given your secret key, which you can copy and paste right in. You don't have to enter your WhatsApp number right now. We'll be able to do that at a later step, but what you need to do is to copy this open claw gateway token and store it somewhere safe because you'll need it later on when reentering your

open claw instance. So write it down somewhere. And then let's let's click deploy and then you'll see the process of it getting set up if for some reason you leave the setup process at any time don't worry you can just head over to your hosting your dashboard head over to vps find yours and then head over to manage here you'll be able to see that your

project is being deployed but even if you lost the one-click setup deployment don't worry the only thing you have to do is head over to docker manager head over to projects here you can top up your ai credits if you want to And here, you'll be able to select from a catalog of different Docker containers. Specifically, we can go with OpenClaw if that deployment

before didn't work for you or if you ever decide to reset your VPS. Now,

while our VPS is being deployed, I want to give you a quick note on API usage. This usage is separate from any ChatGPT or Cloud Pro subscription you might have. It's pay-per-use. You get some free credits to start, which is enough to test everything. But if you hit rate limits, especially on the bigger

models like Clod Opus or GPT-4.0, add a payment method and load 10-20 bucks in credits. And honestly, consider using a lighter model like Clod Sonnet for day-to-day

credits. And honestly, consider using a lighter model like Clod Sonnet for day-to-day tasks. It's fast, cheap, and it'll be able to handle most of the

tasks. It's fast, cheap, and it'll be able to handle most of the tasks. So save Opus and other heavy models for heavy lifting, like deep research

tasks. So save Opus and other heavy models for heavy lifting, like deep research or complex code. You'll be able to switch these models anytime in the dashboard. And

our server is up and running. Your agent has its own computer now, so let's go configure it. Click the OpenClaw button, and here you'll have to enter that OpenClaw gateway token that I told you to save at the beginning, and then click login. This is the OpenClaw dashboard. This is where you

manage absolutely everything from your connected apps, your AI models, the channels you connect so you can speak with it, different instances of it running, your usage, how your agents behave, and special skills that they have. There's a lot of stuff happening right here. But here's what most people miss. You don't have

to manually go through all of these settings and files and modify your agent or its soul or its identity. You don't have to manually toggle on or off any of these things or figure out what these skills do and whether you even need them. See this chat button at the top left. This

is your direct conversation with your agent. You can configure almost anything just by talking to it. We'll use this a lot. So now, let's give your agent a way to talk to you. We're starting with Telegram because it's the fastest setup. Your bot will get its own identity, its own name, its profile

photo, all for free. No second phone number needed. WhatsApp? works too. But the difference is that it requires linking to your own number, which means the bot runs on your account and it doesn't show up as a separate contact. So with WhatsApp, for the best experience where the bot feels like a real team member you're

texting, you need to grab a cheap eSIM, register a new WhatsApp account on that number, and then link it. Then you message that number like any other contact.

But for this video, I'm using Telegram because the setup is instant and it just works. First, you'll have to head over to your Google Play Store or Apple

just works. First, you'll have to head over to your Google Play Store or Apple App Store and download Telegram. Once you download it, simply open it up.

and set it up for your device by entering your phone number. Then, press the search icon at the bottom right corner and search for the Bot Father. That's

Telegram's built-in bot creator. Start the chat, and here you can see a list of different commands that you can send to it. The most important command is the new bot command. So, simply click it right here or type it as a message, and then it'll ask you for your bot's name. Call it whatever you

want. I'm going to call mine Dev because that's what it is. So simply

want. I'm going to call mine Dev because that's what it is. So simply

send that message and then you'll have to choose its username ending in underscore bot such as Adrian's Open Claw Bot. Of course, you can use your own name and click send. And congrats, you have your own Telegram bot. The bot father will give you a token. You can see it right here. So you

can copy it. and then go back to your OpenClaw dashboard. You can paste the entire message, but from it, you just need

OpenClaw dashboard. You can paste the entire message, but from it, you just need to extract the token. So the only thing you need to do is tell it what to do. And in this case, that is to set up a Telegram channel for communication with my new bot token. And you can then paste

this token you got right here and press Enter. It's going to check how to configure Telegram in OpenClaw. It says that it's already enabled, but it needs the bot token that you just provided it. So it's going to fix it for you right away. And there we go. Telegram is now configured and the gateway is restarting. Here's what's set up. Bot token applied, DM policy set

to pairing, which means that you'll need to send the start command to the bot on Telegram, which will then give you a pairing code to approve it. So let's first find their bot by heading over here to the search.

it. So let's first find their bot by heading over here to the search.

and then searching for your bot name. If you search for your exact username that you gave it, you'll be able to find it right here. Mine is called Dev. And while we're here, we can also modify its settings by heading over to

Dev. And while we're here, we can also modify its settings by heading over to its profile and clicking Edit. At any point in time, you can give it a description, change its name, or modify its profile photo. I'll go with the Open Claw logo, but of course, you can go with any kind of a persona or a profile photo of an imaginary dev. It can even be like a

human developer that has claws. I'm sure you can generate that and then use it. So I'll go ahead and set it up and then go back to

use it. So I'll go ahead and set it up and then go back to the chat and click Start to start having a conversation. As you can see, it says OpenClaw access not configured. It's going to give us the Telegram ID and the pairing code. And this is important for security. You are specifically allowing it to be able to access your OpenClaw dashboard. So go ahead and copy it. And

then back in your OpenClaw chat, you can simply paste the pairing code we just copied, which is going to be this one right here, and press enter.

There we go. It's recognized the start command and it's checking out the pairing request and pairing is approved. Your Telegram account is now linked so you can message OpenClaw directly through Telegram from now on. This is what we wanted access to this complex dashboard that has access to all LMS that we want to give it

access to directly from our phone. But the key thing right here is that you're not just talking to an LLM from your phone. We were

able to do that for years now through tools like ChatGPT. But this is different.

Through your phone, you can now talk to an AI agent that has access to its own computer and can perform tasks when your computer is offline or when you're on the go. You just send it a message and it does things.

Now, before you move on, let's give your agent a personality. This is the single most important thing you'll configure. In the dashboard chat, send a message like this, and I'll provide this to you in the video kit link down in the description. Update your soul.md file with the following: dev, senior engineer, and chief of staff. And then give it some core truths to

believe in, the communication style, as well as the boundaries. These are the main parts of a soul.md document. And if you want inspiration, you can also check out souls.directory. It's a community library of personality templates from mindful companions and code reviewers, all the way to pirates,

teachers, security auditors, and also Groot. This right here is just my setup. You can make yours whatever you want. The OpenClaw creator's whole

setup. You can make yours whatever you want. The OpenClaw creator's whole philosophy is that AI assistance shouldn't be boring. Give it opinions, give it a voice, and make it someone you'd actually want to send the text to at 7:00 AM. So let's go over mine real quick. I tell

it to be genuinely helpful, have opinions, if the code is bad, say so, if the approach is wrong, just push back, never sugarcoat, be resourceful before asking, read the files, If it can be handled without me, handle it.

If it needs my attention, flag it with Y. When the answer fits in one sentence, one sentence is what I get. Be direct, no corporate drone energy, use code when code is clear, and be careful with external actions, such as emails, tweets, and anything public facing. Be bold with internal actions. And if you ever decide to change your own source code, tell me. But don't be afraid because

you can always change your sole document by simply chatting with your agent.

Go ahead and click send. And after some thinking, it says done. SolMD has been updated with the new identity and guidelines. I kept the continuity section since that's operationally important. Everything else is your spec verbatim. See what just happened? I didn't navigate any menus and I didn't edit any config files. I

happened? I didn't navigate any menus and I didn't edit any config files. I

told my agent to update itself and And it did. The system prompt shapes every response. And a good one means you stop fighting the AI's defaults and start

response. And a good one means you stop fighting the AI's defaults and start getting actually useful output from message one. And you're not limited just to this example. You control it completely. Wanting to be a research assistant? A project

this example. You control it completely. Wanting to be a research assistant? A project

manager? A content strategist? A coding mentor? Whatever your workflow needs, just describe it. Some ideas could be: Review my code before I push. Flag anything

it. Some ideas could be: Review my code before I push. Flag anything

that looks off. When a client emails, summarize what they need before I respond. Track

my active projects and send me a weekly status update. Or quiz me on system design concepts every morning. your agent becomes whatever you tell it to be.

And here's the moment of truth. Pull out your phone and ask it, "What can you do?" Immediately, back within our OpenClaw dashboard, within the chat, you'll be able to see that the message was recognized and that the assistant is doing something. At the same time, on your phone, you'll be able to see

something. At the same time, on your phone, you'll be able to see a typing indicator. And if you ever get a message like this, that the AI service is overloaded, this isn't coming from OpenClaw, but rather the AI brain that we're using. In this case, it seems to be anthropic, so maybe there's a temporary

we're using. In this case, it seems to be anthropic, so maybe there's a temporary outage of the Opus model. So you can always switch back and change the model to something else or change the AI provider entirely. Then if you retry it, and I might have asked it the same question before, and as per our instructions, it looks like it's not going to give me the same answer again.

So I'll just tell it, you know, go ahead, tell me one more time. And

we got back a full response. And in short, it can code, it can research, it can manage servers, reminders and schedules, browse the web, keep track of the memory, messaging, and it can have proactive checks. Of course, this is specifically what our agent can do, but you can train it to do everything else as well.

This means that it works. You've got a deployed AI agent ready running on a VPS connected to your phone, powered by the AI model you choose. That

took, what, 15 minutes? And right now, it's basically a smart chatbot.

It can answer questions, it can have conversations, but that's not why you're here. You're here because this thing can do things, real things, things that actually change how you work as a developer. So

let's get into it. Okay, your agent is deployed, connected, and running. Right

now, it can answer questions and hold a conversation, but so can ChatGPT. The difference is what happens next. I'm going to show you five things I

ChatGPT. The difference is what happens next. I'm going to show you five things I actually use this for as a developer, live from my phone right now. No

slides, no hypotheticals. I message it, it does the thing, you see the result. So

let's start with the one that I couldn't believe was possible. This is maybe the most exciting use case for me, and that is the ability to build and deploy projects from your phone. Let me show you how it's done. First, you can head over to Vercel.com and create a free account. We'll use this as our deployment

platform. Once you're in, you can search across the settings and find tokens

platform. Once you're in, you can search across the settings and find tokens under account. We'll need to create a token so our AI agent can deploy the

under account. We'll need to create a token so our AI agent can deploy the websites for us. So go ahead and click create, and we have to enter a token name. We can do something like open claw dev agent.

token name. We can do something like open claw dev agent.

select a scope, in this case, I'll give it access to my JS Mastery Pro repo, and then select an expiration date. I'll go ahead and set it to one day, but you can totally go ahead and set it to no expiration, and then click Create. You'll be given a token that you can copy. So

with it, go back right here, paste it, and I'll call this a Vercel token, and I'll also give it my GitHub username. That's just my first and last name. Now, why am I providing it these things? So it

can develop a fun little project. I'll tell it to build a simple dashboard that shows my GitHub commit activity for the last 30 days. I'll tell it to use Next.js and deploy to Vercel, and then I'll provide it the necessary info.

But the key part is that you don't send it from your computer. You

could have coded that yourself. Rather, you send it from your phone.

Immediately, it gets to work, saying that it's a solid chunk of work and that it'll get started. And that's it. It's actually working on it right now as I speak. So, let's give it a minute and see how it does. And I mean, just think about what's happening here behind the scenes.

It's not giving me code to copy and paste, but it's actually building it.

It'll create a project, install the dependencies, test the application and make sure it doesn't have any bugs, connect to the GitHub API, build the UI, push it to our repo, and even deploy it. And even though it's only been a minute, it gave us a little update. And check this out. Our dashboard is now live. So let's check it out by clicking the live deployed link.

now live. So let's check it out by clicking the live deployed link.

It has a nice mobile-friendly design, but it's not really making me seem like a good developer. It seems like the number of commits is zero, and I'm pretty sure I committed a lot of stuff in the last few days.

So let's tell it that we can't see any commits and that it needs to fix it. I'm actually super glad this happened because it's going to show us that sometimes AI is going to make mistakes. But hopefully, just one more message is going to be necessary to get a fully functional project up up and running. Back on OpenClaw chat, we can see that it says that no

commits array and no size field either. The events API seems to be a bit too limited, so it'll switch to using the GitHub search API for commits. And there we go. It is fixed and redeployed. So

once again, I'll open up the URL. And this time, we have real data.

Both the chart as well as the heatmap are working as well. It looks

like I went crazy with the commits a few days ago. And we can also check out the dashboard on our computer. It has a nice loader, and here it looks entirely different. We can see a nice looking fully functional chart with the commits and even the commit heatmap. But it's not even the code that I'm most impressed with, but it's the fact that I was

able to develop and deploy this little project from my phone in literally a minute while I was sitting here talking to you. Of course, this is a toy project and it looks like one, but just imagine what else it can do.

I just wanted to show you the possibility for you to develop something from your phone. Now it's up to you to continue building further with it. But now

your phone. Now it's up to you to continue building further with it. But now

let's try something different. find one of your older GitHub repos and tell it to check your open PRs on it and ask if anything needs attention. And we ran into a little issue. With a free account, it's highly likely

attention. And we ran into a little issue. With a free account, it's highly likely that you're going to go over the free credits that your AI agent of choice provides you with. That means that we have to top up our credits if we want to continue testing or using OpenClaw. One thing that's worth noting though, is that initially, your agent will eat up more of the tokens because

most of the things that we're trying to do are still uncached. But

as you use it with time, your usage will go down significantly as most of the things will be cached. So you can see that it's going to pull the info from the cache, so it's going to use less tokens. But still, with using OpenClaw and these AI agents, usage is not something that

tokens. But still, with using OpenClaw and these AI agents, usage is not something that concerns me because I understand the kind of efficiency and productivity they give me.

So go ahead and top up your agent or switch to another AI provider of choice so we can continue using it. And then let's see what it has to say about our PRs. Super quickly, it got back to me with a response. There

are a couple that are worth merging, a couple that are potentially useful, and some that we should probably close without merging. 1.13, the security fix, seems to be the most critical. So let's tell it that that one seems serious to review it and then leave some comments if anything looks off.

and we immediately get a response. This PR is legit and clean. It bumps

up Next from version 14.2 to 14.2.35, touches Package JSON and Package Log JSON, but specifically, it fixes two related vulnerabilities affecting the app router. What I love about it is the verdict. It says that it's

app router. What I love about it is the verdict. It says that it's safe to merge. It's mostly just a version bump and also addresses some serious vulnerabilities.

So you can go ahead and just ask it to merge it. It'll

likely ask you your GitHub token, which you can provide to it, and that's it. Or we can give it one last check yourself and then merge it

that's it. Or we can give it one last check yourself and then merge it if it's a good one. So what you've just seen are Git workflows from Telegram, code review from your phone, and deployments without ever opening a terminal.

This alone would have sold me, but it gets better. So let's get on to the second use case. Now, let me show you a second use case. I'll tell

it to scan my inbox and tell me what actually needs my attention today. It'll be able to access my inbox by connecting to Gmail using an

today. It'll be able to access my inbox by connecting to Gmail using an app password. So send it over and let's see what it says. Perfect. It

app password. So send it over and let's see what it says. Perfect. It

gave us a link that we can go to to set up our app password.

It's easier to do that on desktop, so you can open up that link there, and you should be redirected to your app passwords. But it's possible that it's going to say that app passwords are not set up for your account.

That would only be the case if your two-step verification isn't yet set up. So, for the app passwords to work, two-step verification on Google needs to

up. So, for the app passwords to work, two-step verification on Google needs to be turned on. This is for security reasons. And it's recommended that you have this anyways, so go ahead and set it up. The only thing you'll have to do is choose one of these second steps. Either a Google prompt, the authenticator, a phone number, and then click Setup. As soon as that's done,

you can open up this link. Then, give it a name. I'll go with something like Open Claw Mail, and then click Create. It'll give you a password, which you need to keep safe and secure. Don't worry, because I'm going to delete this one just before I publish the video, so you can't use it.

And then when you get it, you can paste it either in the chat or in your Telegram interface. So press enter. And now that we've hooked it up, let's ask it the same question once again. scan my inbox and tell me what's happening. And there we go. There are three emails from today. One is

what's happening. And there we go. There are three emails from today. One is

the auth bug on the banking repo, issue 247, mobile Safari auth callback failing. Okay,

this is a major one that needs attention. There's a potential sponsorship mail and then some kind of a job posting. Okay, this is interesting. I've actually just sent these three emails to myself just so I can check it out. I mean, this really sounds like an exciting opportunity, but as soon as you dive into the details,

you'll notice that it's talking about a junior HTML developer role with a competitive salary just for HTML. The email sounds valid, but OpenClaw automatically disregarded it.

This is great. It means that it knows how to differentiate spam from real emails. But that Safari auth issue does sound serious. So I'll tell it

emails. But that Safari auth issue does sound serious. So I'll tell it to read that GitHub issue and fix it. And check out the response. It

looked into the repo and found the issue. The problem was the same site said to strict. So the app write session cookie blocked it on Safari. But if

to strict. So the app write session cookie blocked it on Safari. But if

we change that same site to none in both sign in and sign up, It'll work. And the cookie already has secure set to true and HTTP

It'll work. And the cookie already has secure set to true and HTTP only set to true as well. So this is safe. Now it can either push the change for us or we can push this simple one liner. To enable this, you can head over to github.com/settings/tokens and then create a new open cloth token.

I'll give it read and write permissions and then generate. You'll be given this token, which you can then copy and then paste it back into the chat.

And that's it. Fixed and shipped. It cloned the repo and found the bug. It changed the same site policy. It committed it, pushed to a branch, and

bug. It changed the same site policy. It committed it, pushed to a branch, and opened up a PR. And it's now ready to merge. This means that our bot just went from reading my email to pushing a code fix from one message. And about that sponsorship email, let's ask OpenClaw to draft a

message. And about that sponsorship email, let's ask OpenClaw to draft a counterproposal in a professional tone and ask them to push the budget up and see whether they can shift it to the second quarter. I've just sent the message and that's a draft in my tone, referencing the actual conversation thread.

I didn't give it a template. It's pulling the email straight from the context and from what it knows about how I write. And if I approve it, it sends. If I want to tweak it, I tell it what to change.

it sends. If I want to tweak it, I tell it what to change.

Now, one more cool little trick is that you can tell it to set up a morning email briefing. so that every day at 8am, it'll scan my inbox and send me the summary of anything that needs my response. And that's it. It did it in a second. It'll categorize them

response. And that's it. It did it in a second. It'll categorize them and send them straight to my Telegram. It's even asking if I wanted to do a test run right now to make sure it works. So I'll just say yes.

And that's it. Test run triggered. It gave me my morning email briefing.

So that's it. Tomorrow morning at 8am, before I open up my laptop, I'll have a message waiting with everything I need to know. And every morning after that, it'll happen automatically. This is the thing that replaces the first 20 minutes of your day where you're just moving emails around. Now, let's move to use case number three, and that is research. Research the top five AI agent

frameworks right now. Compare features, GitHub stars, community activity, and pricing, and then send me a report. On it, subagent is researching now, it'll pull real GitHub data, check pricing and compare features, and then send you a full report to Telegram when it's ready. So, let me go make a cup of coffee, and let's see if it can do it in time. And there we

go, the report is in. It first gave me a bit of info of what it researched and asked me where should it send the full report. So I'll

just send a message saying here. And check this out. It provided a full research right here and it took it just a couple of minutes to do that. It browsed the web, pulled the data from GitHub, cross-referenced documentation,

that. It browsed the web, pulled the data from GitHub, cross-referenced documentation, and organized it into something that I can actually use, not a generic summary, Specific numbers, pros and cons, and a recommendation at the bottom. Now think about how long it would take you to do all of this manually. Opening tens of

different tabs, scanning docs, comparing feature lists, and writing it up. An hour?

Three minutes from my phone. And here's where it gets even more useful. I can

tell it to do this on a schedule. Every Monday morning, research what happened with AI and send me a digest. Now, I have a weekly research assistant that never misses a week and never sends me fluff. So if you're running a business, building a product, or creating content, this

alone is worth the $5 a month for the VPS. But we're not done here, so let's dive into the use case number four. This is the one that's hard to demo in a flashy way, but honestly, it's the most powerful one over time because every conversation you have with your agent, it remembers. Not

just in that specific chat session, but permanently. It builds context about you, your projects, and your decisions. Now, for this to work, you need to have been talking to your agent about your projects, I've been feeding mine context about what I'm working on for weeks. So let me show you what that looks like. I'll ask it to tell me what it knows about the

agentic development course that I'm working on. And here it is. I

never sent it a document called course plan, but this is context that it's built up from conversations. Me asking it to research topics, draft outlines, or compare tools. It assembled a picture over time. This is one of the most useful

compare tools. It assembled a picture over time. This is one of the most useful course that I will have released on jsmastery.com ever. It's all about teaching you those AI-assisted coding workflows and the fact that you're not just here to code, but rather to architect. Specifically, we'll dive deep into cloud code, AI agent frameworks, and

more. I'll teach you how to think in agents, prompt effectively, and structure workflows that

more. I'll teach you how to think in agents, prompt effectively, and structure workflows that compound over time. I'll leave the link to the waitlist down in the description below so you can sign up for early access and be the first one to know when new lessons are out. If you like what you're seeing in this video, you'll love just how deep you can go if you truly master agentic development.

Now, we can also ask it if there are any other courses in the pipeline.

And immediately, it replies, "Yep, backend course is coming up." I'll leave the waitlist for that one in the description as well, so you can check it out. But this

right here is the difference between a chatbot and a second brain. ChatGPT can do this, at least not on this scale right now. You close the tab and it's gone. But this thing accumulates knowledge about you and your work. And it's

it's gone. But this thing accumulates knowledge about you and your work. And it's

available from any device at any time. You can also feed it directly by dropping a PDF, a Google Doc link, or a Notion page, and tell it to read those documents and add them to the context about your project info. Your

agent gets smarter every time that you interact with it. So over weeks and months, this becomes genuinely hard to live without because you're becoming that much more efficient. And this is the part that no other Open Claw video covers

efficient. And this is the part that no other Open Claw video covers properly. And it's the part that turns this from a cool toy into something that's

properly. And it's the part that turns this from a cool toy into something that's genuinely yours. Right now, your agent can do a lot out of the box.

genuinely yours. Right now, your agent can do a lot out of the box.

But OpenClaw has a skills system, basically plugins that can extend what it can do. If you head over to skills right here on your OpenClaw dashboard and expand built-in skills, you'll be able to see that there are tons of different community build skills. Here we have some for the most popular connections, everything from the coding agent to Gemini, GitHub, and so on. But

with a bit of Googling, you'll be able to find some skills that other people have created. But here's the real unlock. You can build your own skill,

have created. But here's the real unlock. You can build your own skill, and you do it in the same way you do everything else, by talking to your agent. So simply tell it to create a new skill that monitors. And now you can choose a repo of your own. I gave it the

monitors. And now you can choose a repo of your own. I gave it the next 15 repo for new issues every evening at 6 p.m. Categorize them

by priority based on labels and content and send me a summary on Telegram. And even though I entered the wrong repo name, the agent will now build

Telegram. And even though I entered the wrong repo name, the agent will now build out the skill to work with a new repo so we can easily point it out at the right one. And check this out. The agent created the skills. It did it through file creation, and it even showed me how it

skills. It did it through file creation, and it even showed me how it works. I didn't have to write a config file, I didn't have to read documentation,

works. I didn't have to write a config file, I didn't have to read documentation, I described what I wanted, and my agent built it. So every evening at 6pm, my agent scans the new issues, categorizes them, and sends me a summary I can act on from my phone. Nobody else's agent does this

because I built it for my specific workflow. And once you understand how skills work, you start seeing automations everywhere. A daily stand up generated from your GitHub commits and a calendar, a weekly digest of what your competitors published on YouTube, auto responses on common questions on your repos, sponsor deadline reminders,

invoice tracking, whatever your workflow needs. Your agent becomes less of a general assistant and more of a system designed specifically for how you work. And that's the shift. That's what OpenClaw actually is when you set it

work. And that's the shift. That's what OpenClaw actually is when you set it up properly. Not a chatbot you ask questions, but a system that runs your

up properly. Not a chatbot you ask questions, but a system that runs your workflows, remembers your context, and gets better the longer you use it.

Now, there's a few more things you should know to get most out of this, a couple of advanced tips, memory optimization, and where all this is heading. So, let's wrap it up. Now, before you go, here's a few things

heading. So, let's wrap it up. Now, before you go, here's a few things that'll save you time down the road. First is memory management. Keep in mind that your agent accumulates context fast. So every conversation, every document, every task, it all gets stored. And that's one of the amazing things about

OpenClaw. But over time, this gets bloated and responses slow down. So periodically,

OpenClaw. But over time, this gets bloated and responses slow down. So periodically,

go into the dashboard, review what it's remembering, and flush anything outdated. Do this before it hits compaction. because once it auto-compacts,

anything outdated. Do this before it hits compaction. because once it auto-compacts, you lose control over what gets kept and what gets dropped. Another point

we could get into are multiple models. You're not locked into one AI. I

personally run Clot for most things because it's the best at long context reasoning and code. But some people swap in GPT for specific tasks or even run a

code. But some people swap in GPT for specific tasks or even run a local model through Ollama for things that they don't want hitting an external API.

You can switch models per conversation or set a default. Experiment. And remember,

lighter models like Sonnet are cheaper and faster for everyday tasks. So, save the heavy models for when you need them. Also, voice messages. This is super underrated. OpenClaw supports out-of-the-box, or you can teach it to, support,

underrated. OpenClaw supports out-of-the-box, or you can teach it to, support, speech-to-text. That means that you can send a voice note on Telegram, and your agent

speech-to-text. That means that you can send a voice note on Telegram, and your agent transcribes and responds to it. Sounds like a gimmick until you're driving and need to delegate something. Then it's the most natural interface in the world. Also,

delegate something. Then it's the most natural interface in the world. Also,

multi-agent setups. Once you're comfortable, you can spin up a second agent, one for work, one for personal. Different system prompts, different skills, different memory, same VPS. I haven't done this yet, but it's where this is heading for most

VPS. I haven't done this yet, but it's where this is heading for most people. And finally, learning to work with these agents. That's the thing that nobody's

people. And finally, learning to work with these agents. That's the thing that nobody's talking about yet. Setting up an agent is the easy part. I mean, you just did it. But getting genuinely good at directing agents, knowing how to prompt

did it. But getting genuinely good at directing agents, knowing how to prompt them, how to structure your workflows around them, and how to build systems that compound over time, that's a skill. And it's the skill that's going to separate the developers who use AI from developers who can 10x their skills because of AI.

I'm building a full course on exactly this, agentic development. Not just the tools and the setup, but how to think in agents, how to architect your work around them, and how to go from someone who uses Claude to someone who runs a team of AI agents that handle your entire pipeline. Links in the

description if you want early access. So finally, let's zoom out for a second. An hour ago, you didn't have an AI agent. And now, you've got one

second. An hour ago, you didn't have an AI agent. And now, you've got one deployed on a VPS, connected to your phone, managing email, deploying code, running research, building memory, and executing custom skills you built yourself for $5 a month. That's not a side project. That's a shift in how

you operate as a developer. And honestly, we're still early. The skills

ecosystem is growing every week, the models are getting better every month, and what your agent can do six months from now will make today look like a demo. But

you're not waiting six months. You're set up now. You're already ahead of most developers who are still watching overview videos and thinking about it.

So, if this was useful, you know what to do. Like, subscribe, and drop a comment telling me what you built with your agent and what you want me to build next. I'll actually read those. Now, go build something. I'll

see you in the next one.