YT YouTube2Text

YouTube2Text → Transcripts → OPENCLAW FULL COURSE 3 HOURS: Build & Sell (2026)

OPENCLAW FULL COURSE 3 HOURS: Build & Sell (2026) — Transcript

by Samin Yasar · 31,723 words · 4,432 segments · language en · Watch on YouTube

Open in transcriber Summarize with AI Download SRT/VTT

Full transcript

0:00Welcome to the most comprehensive
0:02OpenClaw course available. I use
0:04OpenClaw every day to run my business.
0:06It handles my content, monitors my
0:08community, trades stocks for me, and
0:09runs automations every day, and does so
0:12much more securely. Everything I'm going
0:13to be showing you in this course is
0:15stuff I actually use every day. I've
0:17been helping people go from zero to
0:19building and selling these exact systems
0:21for thousands in recurring revenue. And
0:23I wanted to put together one resource
0:25with every tip, every shortcut, and
0:27everything I wish I knew and collected
0:29along the hundreds of hours I spent
0:31building and selling OpenClaw agents.
0:33So, here's how this course is going to
0:34work. First, we'll start with why this
0:36matters, what OpenClaw actually is, and
0:39what's actually possible when you get
0:40this thing right. I'll tell you about
0:42the good, the bad, the ugly, and then
0:43we'll get into the full setup starting
0:45from scratch. just getting OpenClaw up
0:47and running to setting up a mission
0:49control dashboard, a multi-agent command
0:51center, uh memory graph, all that. And
0:53then we'll cover all the core concepts
0:55along the way like skills, MCPs, memory,
0:58cron jobs, sub aents, agentic workflows,
1:00and how they can be applied to real life
1:02and business. And once you understand
1:04how all of these pieces fit together,
1:06then we'll build a few real systems so
1:08you can understand what OpenClaw is
1:10really capable of. For example, we'll
1:11start off with a morning brief just to
1:13get easy. Then we'll build your Obsidian
1:15second brain, a personal trainer, a full
1:18content engine, a training bot, an ad
1:19creator that creates 60 plus creatives
1:22overnight, a community manager, vision
1:24claw, and even gets OpenCloud to do
1:26stuff like make your slides and many
1:27more. More importantly, I'll show you
1:29all the patterns you need so you can
1:31build anything you want. We'll be
1:33covering debugging because things will
1:34break and I'll show you how to fix them.
1:36And then we'll talk about how to take
1:38these builds and how people have
1:40actually been selling it with the exact
1:42pricing, the offer structures, the full
1:44endto-end client examples. And that's
1:46important because I find a lot of people
1:48watching this are probably going to be
1:49doing so, not just for their own
1:51hobbies, but for financial reasons. We
1:53want to make more money with the things
1:54we build. And I want to show you guys
1:56how to do that. So, if you're brand new,
1:58watch this in order. And if you already
2:00have an open claw running, use the
2:01timestamps in the video to jump to
2:03whatever you need. My one recommendation
2:05is build alongside me. Don't just watch.
2:08All right, let's get into it.
2:10Every few decades, a tech shift creates
2:12an entirely new category. Personal
2:14computers build software companies, the
2:16internet built web agencies, and we're
2:18watching the same thing happen right now
2:19with autonomous AI agents like OpenClaw.
2:22And a lot of people are completely
2:23missing it. Too busy arguing whether AI
2:26is going to take their jobs or not. It's
2:27better right now to be the builder and
2:29position yourself to capture this. Like
2:31when chat GPT came out in 2012, it was
2:33impressive, but you had to babysit it.
2:35You phrase things carefully. You had
2:37this back and forth where you'd reass
2:39the same questions about five different
2:41ways trying to go and get what you
2:43actually need. And there was hard
2:44ceiling because it can only talk back to
2:46you. It couldn't open your browser or
2:47touch a file or send an email. You'd
2:49have to ask it to draft a message. It
2:51drafts the message, then you go copy and
2:52paste it for yourself. Now, the models
2:54got a lot smarter and faster. They went
2:57from needing handholding to actually
2:58being able to reason, plan, and break
3:00down multi-step problems on their own.
3:02And people started building harnesses on
3:05top of these models. That means ways for
3:07them to use a browser, write a file,
3:09connect and use external apps. And then
3:11OpenClaw and similar frameworks like
3:13this exploded in popularity. We're
3:14talking some of the fastest growing
3:16projects in software's history. And this
3:18happened because this is the first time
3:19AI stopped being a tool that you use and
3:22started working for you proactively. And
3:24the question stopped being can it
3:26understand me and it became what can I
3:29actually let it do? Now a quick note on
3:31the difference between what a model is
3:33and how it differs from openclaw which
3:36is called a harness because this is
3:37worth understanding before we go any
3:39further. A model is the brain. It's the
3:41thing that reads reasons plans like claw
3:44sonnet opus GPT gemini all those are
3:48models. The harness is the system that
3:50wraps around the model. is the
3:52scaffolding that gives the brain
3:53something to do with its thinking. So
3:55tools to use, memories to draw with, a
3:58computer to work with, like an engine
4:00and the rest of the car. The engine is
4:02the model and without the car around it,
4:04the engine just sits there. The harness
4:06is what makes this go somewhere. And
4:07openclaw is that harness. So what is
4:10openclaw really? Openclaw is an
4:11abstraction layer that sits on top of
4:13these models. Like the brain that we use
4:16a smart and open claw is the body around
4:18it. It gives the brains the hands, the
4:20eyes, a phone, login to some tools. It's
4:23an autonomous agent that doesn't wait
4:24for your prompts. It runs continuously
4:26in the background. It will use the tools
4:28you give it and take actions on your
4:30behalf because it remembers all the
4:31context like a person like Hugh. It has
4:33its own browser and it runs on a machine
4:3624/7. And you might be thinking, isn't
4:38this what cloud code kind of does? Kind
4:40of. But there's a reason why I use
4:42OpenClaw as my orchestration layer and
4:44not clot code directly. You see, these
4:46tools were designed for different jobs
4:48fundamentally. Claude Code, like the
4:50name says, was built to code. It's
4:52excellent at writing, debugging, and
4:54shipping code. That's its one purpose.
4:55OpenClaw, on the other hand, was
4:57designed to be a personal assistant. It
4:59was built to know you, your business,
5:00your customers, your preferences, your
5:02workflows, all of that. So, what it
5:04means is that OpenClaw is meant to sit
5:06between you and all your tools and make
5:09decisions about how to use them. Other
5:10harnesses like Codeex or Cloud Code are
5:12great at code, but they don't really
5:14have that context or see your business
5:16in that light. They don't know your
5:17customers, your past decisions, what
5:19failed last month, or what your brand
5:20even sounds like. How we're going to be
5:22using OpenClaw is that we're going to
5:23make OpenClaw hold all that information.
5:26So, for example, I keep all my business
5:28context, my meeting notes, customer
5:29data, past decisions, what worked, what
5:31flopped in my Obsidian vault. And
5:33OpenClaw has access to all of it. And
5:35when it's time to do something, it pulls
5:37the right context and routes it to the
5:39right model or tools to do the task.
5:40Another big advantage is that OpenClaw
5:42is model agnostic and it's local. That
5:45means all the data lives on the computer
5:47next to you and you're not locked into
5:49one provider. And since these models are
5:51always leapfrogging each other
5:53constantly, Open Cloud lets you be
5:54nimble. You always use the best model
5:57for each specific task. So for example,
5:59if Claude is better at writing, you use
6:00Claude for that. If GPT is better at
6:02understanding images this week, then we
6:04use that. we route the right tool to the
6:06right task. So the way we're going to be
6:08actually using OpenClaw is like an
6:09orchestrator and I call my one cat. It
6:11spawns sub agents that writes their own
6:13prompt, takes the right model for each
6:15task and monitors what they're doing.
6:17And then it messages me on Discord or
6:20Telegram when something needs a decision
6:22and I'm not watching a screen. For
6:23example, if there's a coding task, CAT
6:25will use clot code to handle it through
6:27OpenClaw. I use OpenClaw to use clot
6:30code in not instead of it. Think of it
6:31like this. Each agent is a contractor.
6:33They're good at their specific job, but
6:35they need a project manager who knows
6:37the whole picture. Open Claw is that
6:39project manager. You use each tool for
6:41what it was designed for, and that's how
6:43we're going to be using this throughout
6:44this course. And that means your job
6:46literally changed, which means this is
6:48how we should be looking at working in
6:49an agentic world. The old way is you
6:51have to analyze the data, write the
6:53reports, you build the dashboard, do the
6:55task yourselves. The new way is we
6:57design the systems in OpenCloud once and
7:00get the agent to do all of that. We just
7:02need to review, guide and decide. And
7:05because of this, the system thinkings
7:06and architecture is the new bottleneck.
7:09Execution has become very cheap. Now the
7:11agent can write, research, code, analyze
7:13and build whatever you want. That's
7:15commoditized. What's expensive is
7:17knowing what to connect, what to
7:18automate, and what data actually
7:20matters. And most importantly, what
7:22problem is worth solving. The hard part
7:25isn't doing the work. The hard part is
7:27designing the open claw system that does
7:29that work. The AI doesn't replace the
7:31thinking. It replaces the execution. So
7:33the value will always stay with you. It
7:35just shifts to a different kind of work.
7:37This is where strategy becomes king. Now
7:39see, you're still the one in charge
7:41because there's a version of all this
7:43where it sounds like you handed the keys
7:45and disappear. That's really not how
7:47this works. Because agents are powerful,
7:49but they're not autonomous decision
7:50makers. You can't let it loose without
7:52structure. It's on you to give it the
7:54right access and to protect yourself
7:56from giving it too much too fast. For
7:57example, a researcher gave her agent
7:59access to all her email inbox, then the
8:02agent just deleted it. She lost years of
8:04emails, and that actually happened. And
8:05that's why it's on you to verify the
8:07outputs before you trust them. And I'll
8:09show you how to put all those safeguards
8:11in place before you let an agent send an
8:13email, move money, or publish anything.
8:15Ideally, you check its work first
8:17yourself until you know you can trust
8:18it. You see, the agent is only as good
8:20as the system you build around it. And
8:22open claw gives you the tools to control
8:24all of this like the permissions, what
8:26it can access, what it can't, but you
8:27have to set that up right. We'll cover
8:29all of this in detail later.
8:30Permissions, security, the trust ladder,
8:32how to expand access, how the agent can
8:35gradually earn it. But the reality right
8:36now is that the gap between people who
8:38know how to orchestrate agents and the
8:40people who don't is already real and
8:42it's going to get a lot wider. Every
8:44company is going to need this and most
8:46have no idea that it barely exists or
8:48how to even start with it. And the
8:49people who understand this right now,
8:51while it's still early, are the ones who
8:53end up building businesses and careers
8:55everyone wish they started before. The
8:56same systems you're going to be building
8:58in this course, people are already
8:59selling them to businesses for done for
9:01you digital employees. They charge from
9:03500 to 5,000 to set it up and 200 to
9:05a,000 a month just to maintain it. The
9:07business owner doesn't need to
9:08understand how it works. They just get
9:10an agent that runs 24/7 and cost a
9:13fraction of what a human would cost.
9:14We'll go deep on this in part eight with
9:16the exact pricing, offer structures, and
9:18full endto-end examples, but I want you
9:20to think of it like this. There's a
9:2260-minute clips where Anderson Cooper
9:24asks Rick Rubin what he actually does.
9:26>> Do you play instruments?
9:27>> Barely.
9:28>> Do you know how to work a soundboard?
9:30>> No, I have no technical ability. And I
9:33know nothing about music. [laughter]
9:36>> You must know something.
9:37>> Well, I know what I like and what I
9:39don't like. And I'm I'm decisive about
9:42what I like and what I don't like. So,
9:44what are you being paid for?
9:46>> The
9:48confidence that I have in my taste
9:51and my ability to express what I feel
9:56has proven helpful for artists.
9:58>> You see, Rick Rubin produced some of the
10:00greatest albums ever made across every
10:02genre. Not by doing work, but by knowing
10:04what good sounds like. Rick Rubin still
10:06has to decide who's in the room, what
10:08instruments are available, what studios
10:10to use, what the album's supposed to
10:11feel like before anyone plays a note.
10:13With Open Claw, in the same way, you're
10:15still designing that system like what
10:17tools it may connect to, what data it
10:19pulls in, what workflow it follows, and
10:21what it's actually trying to accomplish.
10:23Everyone will have agents soon, but not
10:25everyone's agent will perform the same.
10:27The difference isn't the model, it's how
10:29you architect the system, like what you
10:31connect, what you prioritize, and how
10:33you structure the workflow. That's one
10:34of the biggest things I want to show you
10:35how to do. Your taste shapes the system
10:38and the system shapes the output. Which
10:39means the real work isn't just judgment.
10:41It's what you build to apply that
10:43judgment to. So the potential ideally is
10:46obvious, but the path to get there
10:47isn't. And the gap between the power and
10:50the accessibility is where these
10:52fortunes remain. Like for example,
10:53WordPress is free, but people still pay
10:55thousands for someone to build their
10:56site. Shopify is simple, but Shopify
10:58experts charge $10,000 for store setups.
11:01The value was never really in the
11:03technology, it's in the implementation.
11:05And once you build an opencloud system
11:06once, the second client would take a
11:08fraction of the time. And here's the
11:10part that people don't expect. The agent
11:12helps with this replication. You don't
11:13manually start rebuilding each time. The
11:16system documents itself. And the
11:18packages you worked on initially turns
11:20into a template or a skill. And the next
11:22deployment is mostly just configuration.
11:24And the best agent builders won't be the
11:26best engineers. There'll be people who
11:27understand a specific customer workflow
11:29better than the customer understands it
11:31themselves. Like for example, a real
11:32estate agent who builds an agent system
11:34for real estate agents will always be
11:37better than a generic developer. You
11:38would know the pain points, the
11:40language, the daily friction. That's the
11:42real edge.
11:44Full transparency on to what you're
11:46getting into. I want to tell you the
11:47good, bad, and ugly. And the good is
11:49that OpenClaw right now is really
11:51working in production systems for real
11:52businesses. Like actual businesses right
11:54now are using these systems to replace
11:56hours of manual work and actually
11:58replace employees. And the nice part is
11:59that the agent gets smarter over time
12:01and compound. And because OpenClaw is
12:03open source, you own everything. The
12:05build, there's no vendor lock in and the
12:07team is always adding new features and
12:09maintaining it every day. But the bad is
12:11that the setup is not trivial. There's a
12:13real learning curve and it takes time to
12:16configure it properly and the context
12:17windows have limits. Most agents don't
12:19automatically mean better results. And
12:21you can actually degrade performance by
12:23adding too many stuff without thinking
12:25it through. Stuff like memory drops and
12:26midconvo here and there. Agents can be
12:28unreliable, but so can people. And you
12:30don't fire an employee the first time
12:32they make a mistake. And the ugly is
12:34that the security is a real issue. A
12:36researcher audited the space and found
12:38500 vulnerabilities. So we want to make
12:40sure we architect our security properly.
12:43For example, if you give your open claw
12:45too much autonomy without understanding
12:46what it's actually doing, you'll stop
12:48being able to verify its work. But the
12:50nice part is that each of these problems
12:52are solvable. Every tool that ever
12:54mattered had rough edges early. Like for
12:56example, the internet was unusable until
12:58like 1995 or something. Smartphones
13:01crashed constantly in 2008. But that
13:03didn't stop the people who learned them
13:05early. The people who figured this out
13:06now, while it's still messy, while it's
13:08still a little hard, while most people
13:10just watch from the sidelines, those are
13:12the ones that end up building these
13:14really great businesses and careers
13:16before everybody else. You don't need it
13:17to be perfect. You just need it to be
13:19functional. And that's it. Open Cloud
13:21will get updated and better constantly.
13:23The security issues will get patched and
13:25the memory gets better. The setup gets
13:27easier all the time. Right now, it's the
13:29best window that you should be able to
13:31taking advantage of early. But the
13:33window to be early that closes
13:35everything in the bad and ugly list I
13:37was talking about. We're going to
13:38address all of it. The permissions, the
13:40security, the trust ladder, backups, how
13:42to protect yourself. But the opportunity
13:44of learning on the other side is real.
13:46People are literally replacing $4,000 a
13:48month in labor costs with a $20 agent.
13:50They're building real businesses around
13:52this. Then they're compressing months of
13:53work into days. And it all starts with
13:56just setting it up. So let's build
13:57yours. Everything from here is hands-on.
13:59We're going to be setting up your first
14:01OpenClaw agent from zero. Then we'll
14:03layer in the skills, the memory, the
14:05MCPs, cron jobs, everything you need to
14:07build the systems I just described. So,
14:09let's get into it. All right. Now, we're
14:12going to be setting up your agent from
14:13scratch from nothing to a multi- aent
14:15working system. And you should also be
14:17able to text it from your phone. And
14:19these sections are going to be modular.
14:21So, if you already have it set up, you
14:22can skip to whatever you need. And I'm
14:24going to show you two paths in this.
14:26First, how to set it up on a VPS. And
14:28that's the cheaper way. And then there's
14:30the Mac Mini Path. So if you already
14:31bought one, we can use that. Or if you
14:33have a older Apple computer, you can do
14:35the same thing with this. Now, after the
14:38machines are ready, everything else, the
14:39models, the channels, identity files,
14:42they're all the same regardless of
14:43whatever hardware you choose. Some
14:45sections matter more than the others.
14:47And I'll flag what's critical.
14:50All right. Now, we're going to be
14:51installing OpenClaw on our Mac. And to
14:53do that, what we're going to do is we're
14:55going to go to classroom. And this is in
14:57my free community down below and then go
15:00to YouTube video resources and in the
15:02bottom you'll see there's a guide for
15:04installing openclaw. So I'll have this
15:06open on the side and then I'm going to
15:08open up my terminal. Terminal I open up
15:11my spotlight search and then all we have
15:13to do is just follow these guides. And
15:15before getting into what we're actually
15:17doing, I want to explain what home brew
15:19is. And homebrew is basically a package
15:22manager. What it does is it basically
15:24gets to download a bunch of code from
15:25the internet and like your app store.
15:28Think of it like an app store for your
15:30terminal. Okay. And node is Node.js is
15:35basically a framework that openclaw is
15:38built on top of. Okay. So we need these
15:40before we can start running our
15:42openclaw.
15:44All right. So all we have to do is just
15:46follow this guide. So now let's copy
15:48this. Let's paste this.
15:52And then it asks for our password. Okay,
15:54great. So now this one's done. Let's run
15:57the second command.
15:59Okay, step two is done. And then now
16:02let's install node. Okay, now let's
16:05install openclaw.
16:08And we're just going to do this paste
16:11here. Okay, now I'm going to run step
16:13three. And intentionally I put this in
16:17because you'll see how now I'll hit an
16:19error. Most people are going to be like,
16:20"Oh no, what do I do now? I'm stuck."
16:23So, at this point, what I want to show
16:24you is when you run into snags like
16:26these, how can you debug these kind of
16:29errors? So, you are a lot better
16:31equipped to change and do whatever you'd
16:34like with OpenClaw. And the nice part is
16:36that you could just get to talk to your
16:38computer. So, let me show you how to do
16:40that. You see, if there's one thing I
16:41can guarantee you is that things will
16:43break. Your agent will crash, lose
16:44context, throw errors, or just stop
16:46responding. And this happens with
16:48anything. Cloud code, codeex, open claw,
16:51everything. But the skill I want to
16:54teach you right now isn't to like avoid
16:56breakage. It's how to actually fix
16:58these. And this section right now is I'm
17:01going to cover how you can build a
17:03system where you never get stuck. You
17:05always have a way to diagnose what went
17:06wrong and to get you back up and running
17:09even if you don't understand the
17:10technical details yourself. And what
17:13we're going to do for this is learn the
17:15partner system. And the partner system
17:17essentially is that with two tools that
17:20know your setup and has access to your
17:21computer, you always have a partner to
17:23diagnose and fix problems. That means
17:25one is always available to rescue the
17:27other. And this is one of the most
17:29important concepts and things to set up
17:31early. Please don't wait until something
17:33breaks to figure out your recovery plan.
17:35All right. So, what we're going to do,
17:37we're going to set up a partner that's
17:38going to save us whenever we run into
17:40errors. And I'm going to show you how to
17:41do that. All right. Okay. So to set up
17:45the partner system, what we're going to
17:46do is download cloud code. And what
17:49we're going to do to do that is do
17:50claude is just Google download claude
17:54desktop. And then let's hit the first
17:56link. And then you see
18:00there's this app. So you want to
18:01download that. And then after you
18:03download that, if you open up your
18:05Claude app, you're going to see Claude
18:07like this. Okay. So this is what I want
18:09you to learn. So, right here, I'm going
18:11to go to code.
18:13And in code, what we're going to do is
18:16we're going to select a folder. We're
18:17going to choose a different folder.
18:20And I'm just going to select my whole
18:21computer by hitting open. Okay? And you
18:24see it says my name, which means it has
18:26access to my entire computer. Now, all
18:28you want to do, this is the biggest
18:30skill that I want you to learn, is you
18:32come here, you copy, and then you paste
18:34it. And then I'm going to say and speak
18:36right here. Hey, uh, I'm trying to
18:40install CL OpenClaw, but I'm running
18:43into errors. Can you help me set that
18:46up?
18:50And you trust work workspace. And this
18:53is really important. The reason I
18:55intentionally put this error in is
18:57because this stuff happens to people all
18:59the time where you're just going to run
19:01into an error and then you're going to
19:02give up. But if you know this workflow
19:05where you just go know how to go back
19:07and forth and get clawed code or
19:10anything like codecs or whatever to fix
19:12your open claw or vice versa. Sometimes
19:15your you'll see your claw code break and
19:17you can get open claw to do it. That's
19:18what we want to do. And see right here
19:22the error is a permission issue.
19:26and then I can just be like run it for
19:29me
19:31because cloud code has access to your
19:34computer. It can do stuff like that. And
19:36that's what I wanted to show you because
19:38I'm going to show you a couple different
19:39hacks where when it comes to debugging,
19:41this is how we should approach it. Okay,
19:43it just got done and it's telling me to
19:45run the install script. So, I'm going to
19:48take this and then I'm going to open up
19:49my terminal. I'm going to install this
19:51here and then let's see what it asks.
19:54And you see uh I was what I wanted to
19:56show there is that hey I was running
19:58into errors because I didn't have the
20:00right packages or whatnot. But all I had
20:03had to do is just talk to cloud code be
20:06like hey what's wrong without me having
20:08to know anything and just go back and
20:10forth with the terminal and cloud code
20:12to make sure all of this stuff is
20:14working. Okay. So after you run that
20:16this is what you should be met with. And
20:18I want you to hit local. And then for
20:22what this one is, I want you to select
20:24workspace and then select where
20:26everything will be saved. That's what
20:27the workspace is. All right. So when it
20:29comes to model, I've seen a lot of
20:30people get tripped up right here. And
20:32you can boil it down into three options.
20:34So option one is let's say a
20:36subscription where you can use your
20:39OpenAI $20 a month plan and you get
20:41technically unlimited usage, but it just
20:43rates limits you. Or you can also use
20:47your Claude, but I don't think they
20:48recommend it. With a subscription,
20:50you'll get unlimited usage for a flat
20:52monthly cost and you know exactly what
20:53you're paying for. And OpenAI actually
20:56publicly said that, you know, using your
20:58subscription with tools like OpenClaw is
21:00allowed. And this is what I recommend
21:01for most people because it ends up being
21:03much cheaper than everything else in the
21:05long run and you can predict your costs.
21:07That means no surprise builds and no
21:09anxiety about token costs. All right.
21:11Number two is using an API key. So let's
21:13say you want to use other models and you
21:16can go through open router and use your
21:18API key and that will pay per token.
21:21That means as much you use the more you
21:23pay for it. That means a complex task
21:26might actually end up costing you $5 to
21:28$20 in a single session and simp and
21:30simple tasks cost pennies but they can
21:33add up. This is better for advanced
21:35users and we'll cover cost optimization
21:37strategies later in the course but right
21:40now I would say avoid this if you're
21:42just starting out. And option three is
21:44local models. And local models is really
21:46cool. And a lot of people like OpenCloud
21:48because it allows you to use local
21:51models. And that's why people are buying
21:52those beefy Mac studios because what you
21:55can do is you can download this thing
21:57called Olama and then download a local
21:59model that runs just on your computer
22:02and that AI model runs on your Mac
22:05Studio. Okay? which means there's no
22:07more API costs and it would be
22:09completely free and it would just cost
22:11you whatever your hardware costs and
22:14it's really safe because your data never
22:16leaves your machine. But the trade-off
22:18is that you'll probably need a really
22:20beefy computer that costs a lot of money
22:23and they're not as smart as the bigger
22:25cloud models like OpenAI or Cloud. It's
22:27good for people who care about data
22:29privacy and want to experiment without
22:31spending much. But apart from that,
22:35right now I'm going to show you how to
22:36set it up with what I recommend for most
22:38people, which is probably just an OpenAI
22:40subscription because it's the cheapest
22:42and you get the most use out of it. So,
22:44let's dive into that. What I recommend
22:45using OpenAI and then what they
22:48recommend and this is the cheapest way
22:50to do it and you're going to be saving a
22:52lot of money is by using OpenAI codecs.
22:55Okay. So, right here,
22:58if you right here come in and you just
23:00hit enter
23:02and log in with codeex,
23:06this is going to use your $20 chatgpt
23:08subscription. And then that means you
23:10can use your codeex models. And right
23:13here, I'm selecting let's say codec
23:15spark and codeex and then hitting go.
23:18And then I'm just going to hit continue.
23:20All right. So, it's doing its thing.
23:22It's updating. It's plugging and
23:23whatnot. All right. Right. And then it
23:25automatically opened up OpenClaw.
23:28Awesome. And let's test it. Let me just
23:29say hi. Awesome. So, our OpenClaw is
23:32alive. Now, let me show you how to
23:34actually get into this and edit the
23:36OpenClaw so you can get it to do the
23:38things you need.
23:40Okay. Now, I'm going to show you how to
23:41set up the channel. So, first we're
23:43going to begin by setting up something
23:44like Telegram. So, you can text it from
23:46your phone phone from anywhere and you
23:48can do voice memos to it. And then this
23:50is my personal favorite is I'm going to
23:52help you set up Discord. And the
23:55advantage of Discord is I'll show you
23:57right here
24:02is right here. You can have multiple
24:04threads with your agents. That means you
24:06can have sub agents working across you
24:08and collaborating with each other. This
24:10is what really really unlocks OpenClaw
24:13for how good it is. And I'll show you
24:15how to do that.
24:17And just to keep things like a beginner,
24:19I'm going to be stumbling through this
24:21just like anyone would just so you know
24:24what it's like and you can see the live
24:27um things I do to debug in real time. So
24:30to set up the channel, all I'm going to
24:32do is just talk to this thing and I'm
24:34going to be like, "Hey, I want to use
24:35Telegram with this. How can I do that?"
24:40See, uh, one thing I want to make sure
24:42that you understand is everything you do
24:44with OpenClaw, Cloud Code, etc., it can
24:48already figure it out. So, what you
24:50probably just need to do is ask it if
24:52you're trying to do something. So, right
24:54now, it gave me some stuff. So, let's
24:56try this out. Okay. So, imagine it gave
24:59us a bunch of this stuff and I'm like, I
25:01don't know how to do any of this. Can
25:03you open up the browser and then do the
25:05stuff for me?
25:20Okay, great. So, you see now, let me log
25:24in. Okay, so it opened it up. Now,
25:33let me do what's next. So, now it's
25:35telling me to do slash
25:37newbot.
25:41Uh, let's call it Okay, let's just call
25:43it simmon's clawbot.
25:47Okay.
25:49And then it wants me to use this access
25:51token. I don't know. Let's say what it
25:53does. I'm going to paste it here and
25:56then enter. Okay. So, you see how it's
25:58disconnected now. And what's going to
26:00happen right now? It spun back up. So it
26:03just restarted itself which is pretty
26:04cool. All right. So now it should be
26:08fine. Now if I go back to Telegram. So
26:10in Botfather
26:13if I just press this
26:16I should have it.
26:19Okay. And then what I'm going to do is
26:20I'm going to copy paste this stuff.
26:22Okay. Cool. So it says um your Telegram
26:25is now paired. You can message your bot
26:26directly. So I just said hi and then say
26:28hey it's working. Telegram is now
26:30connected and paired. Awesome. Okay, so
26:33let's move on to step two. All right,
26:34now I'm going to show you how to set it
26:36up on Discord and I want to take you
26:38over why we actually use Discord and why
26:40it's so much better to actually using
26:42these agents and working in parallel. So
26:45Telegram right now is a single stream.
26:48So you just have one chat thread. If you
26:50want to talk about different topics or
26:52different ideas, it's really hard to do
26:54that. So for example, that's why things,
26:56you know, get lost in context or
26:58summary. It's good for quick, simple
27:00chats or just, you know, sending a voice
27:02message here and there. But Discord is
27:04actually your workhorse. You can have
27:06different channels for your actual
27:09different agents and then they'll hold
27:11all the context that you need and you
27:13can have them working in parallel as
27:16they go. So it makes it so that your
27:18systems are way more robust and then
27:19they work a lot better. So basically
27:22what we get is separate context. We get
27:24parallel tasks and specialized agents.
27:26It's perfect for complex projects and
27:29also you can invite your team and they
27:31can work on that too. So with that out
27:34of the way, let me show you how to set
27:35this up. Okay, to set up Discord with
27:39your OpenClaw, it's not the most trivial
27:41thing ever. So I actually made a guide
27:43that you can follow and it's really
27:44simple. All right, so first [snorts]
27:46what we want to do is just open up.
27:49Okay, great. So it should ask you to log
27:51in and then what you want to do is you
27:54want to create a new application. Okay.
27:56And let's call it summon
27:59open clawbot
28:02course.
28:04[snorts] And then I'm going to hit check
28:05and create this. And then I have to
28:07verify that I am a human. Okay, great.
28:10Now that I have that, what you need to
28:12do is you want to go to bot
28:16and then down here you want to hit check
28:18this privileged gateway intents. And
28:21what that does is allows your bot
28:25and OpenClaw to communicate with each
28:27other. And then I'm going to scroll
28:28down, hit administrator, and then hit
28:30save changes. Then what I want to do is
28:34I want to hit reset token and then do
28:36that. Then it wants your password. So
28:39I'm going to enter my password and then
28:41I'm going to copy this. And then I'm
28:42just going to keep it in my and paste it
28:44in my open claw for now. Okay.
28:47All right. So you should have everything
28:50there. Now what we want to do is we want
28:52to open up Discord.
28:54So if you just make an account, you
28:56should see all of this stuff. And on the
28:57left hand side, if you scroll all the
28:59way into the bottom, if you hit this add
29:01a server, you can create your own server
29:04for yourself. And let's call it Sins
29:07uh open claw course server.
29:12Great. Now, what you want to do is you
29:14want to go here and then you want to
29:17copy and then rightclick and then hit
29:19copy server ID and then right here I'm
29:22going to paste server ID and then I'm
29:25going to paste it here and I'm not
29:26entering yet and just to keep myself
29:29organized. Bot token
29:32and then what you want to do is you want
29:35to hit yourself your
29:37>> [snorts]
29:37>> icon right here. Hit copy user ID. go
29:40down here and then say user ID and then
29:44enter that there. And then one more
29:46thing is you want to go and hit this
29:49gear icon. And this gear icon if you
29:52scroll down inside advanced if you want
29:55to you want to turn on developer mode.
29:58Okay. After you have all that done, if
30:01you scroll down here and then copy this.
30:04If you copy that, paste it here.
30:10and right here.
30:13So, I just copied this. I have my bot
30:16tokens here and then I'm going to hit
30:17send and I'm going to wait. So, that
30:20should that's what your bot should need
30:23to set up your Discord. So, now I'm just
30:26waiting until my bot sets up my Discord.
30:29What's going to happen is it's going to
30:31be invited in here and then it should
30:33have access. So, let's wait for that.
30:35Okay. Then you see it's giving us all
30:38the stuff. And then it wants us to press
30:40this one,
30:42allow.
30:44And then you see it's now inviting this
30:47bot to my server. I'm going to hit
30:49continue and authorize. And look, I hit
30:51into went into a snag. And it's saying
30:56uh now I'm just going to tell it, hey,
30:58it's saying integration requires code
31:00grant.
31:17I'm surprised why it opened this up.
31:21Yep, I just fixed it. I'd returned.
31:25Okay, I'm not really sure what that
31:26means, but let's just go ahead and then
31:28click this button
31:31and then hit allow.
31:33Scroll down. Authorize. Okay, great. I
31:35think we should be good to go. Yeah. And
31:38those kind of stuff always happens here
31:40and there. Uh errors, but you need to
31:42know that you can just hit
31:45uh this button and just ask OpenCloud
31:47what's up and then it'll fix it. So,
31:49okay, great. Now that's done. Let me
31:50make a couple channels so we can
31:52essentially have an organized multi-
31:55aent Discord setup. Hey, [snorts] can
31:57you make a couple more channels so I can
31:59let's say manage my life in one channel,
32:01another one is for my video ideas and
32:03creations and one of them is for let's
32:05say stock trading. Uh I want different
32:07sub agents there so we can manage and
32:09have different context.
32:12Okay, there we go. And now it's as
32:15simple as just talking to OpenClaw and
32:17hitting enter. So I'm going to wait now.
32:20And you can see it already started
32:22creating channels like life management.
32:24Okay, one thing I want to show you right
32:26here. So, let's say this is like a video
32:27idea creation. What I can do is I can do
32:29slash thread and let's say I have a new
32:32idea called uh open claw course. And
32:35then I say like this thread is about my
32:39open claw course. And you see right here
32:43under my videos I have everything for
32:47managing my openclaw course. And right
32:49here we can have my channel and my
32:52OpenCloud will have access to my
32:54computers, all the context everywhere.
32:56And this one I'll have a dedicated agent
32:58where I can talk to and it's way easier
33:00than me having to look around and trying
33:04to manage and lose any context. These
33:07keeps my agents really really performant
33:10and I recommend you do the same.
33:12Okay, now that we have the communication
33:14channel set up, I want to show you how
33:16we can make a memory graph like this.
33:18Like this is my actual open clause
33:20memory graph where it helps it remember
33:24everything in context. And let's say if
33:26I want to find something like I don't
33:28know open claw, it can filter out from
33:31that what the subsets are. It can know
33:34what they are. I can like um pull out
33:37this is something I've been working on
33:39like a VSSL etc. So all of these is
33:41pretty cool and having this memory graph
33:44helps OpenClaw really really like you
33:47know dial in its memory. So I'm going to
33:48show you how to set that up. All right.
33:51So let me show you how to set that up.
33:52And just to keep everything organized,
33:54what we're going to do is be like, "Hey,
33:57can you make me a new channel?" And just
34:00call it setup. Uh I just did that just
34:03so you guys can see. But what you can
34:06really do is honestly just hit create
34:08channel and then text channel and then
34:10you can call it setup. It's a lot
34:12easier. But anyways, here I'm going to
34:14be showing you how to do that memory
34:15stuff. And I actually have a lot of the
34:18things in structure in my cloud
34:20community. But I'll leave this in my
34:22free communities as well where I have a
34:24skill called Obsidian Open Cloud Memory.
34:27And how the memory system works is
34:30through this app called Obsidian. and
34:32Obsidian. Everything OpenClaw has access
34:36to is essentially markdown files, which
34:38is basically text. And these text files,
34:41anything that links to each other is
34:43that graph. So, I have the skill right
34:46here that helps you build that out. All
34:48you need to do is just come here, copy
34:49this, and then be like, "Hey, uh, I'm
34:53trying to build an Obsidian memory graph
34:56for my OpenClawbot for you. Uh, can you
35:00use this skill to unpack this and build
35:03it out?
35:04You technically don't need the skill at
35:06all. It just makes it a bit more token
35:07efficient and faster. But, uh, now all
35:10I'm going to do is just wait and it's
35:12going to go through and set up my open
35:15claws memory. So what's going to happen
35:18is my open claw can do like rag vector
35:22search which is essentially like how you
35:25think about different things and
35:27remember a recall. Open claw is going to
35:29do the same thing and set up its memory
35:31in a way that it's a lot better and it
35:33can retrieve things that it may not have
35:36access to otherwise. So this is going to
35:38take a while but I'll come back after
35:40it's done. Okay. So it's saying it's set
35:42it up now. I'm can be like hey can you
35:44open up obsidian so I can see. Okay,
35:47it's launching Obsidian and this is my
35:50new Open Cloud that we just made right
35:51now. All right, and here it is. It's a
35:53lot smaller because there's not much in
35:55its memory right now. There's only a
35:56couple files, but you'll see as you use
35:58this, it'll get bigger and bigger. It'll
36:00add more connections and the memory when
36:03you ask it to remember something about
36:05let's say like a couple months ago,
36:08boom, it'll be really easy. So, with
36:10that out of the way, now let's set up a
36:13mission control so you can see your
36:14agents work in real time.
36:17Okay, so now I'm going to show you how
36:18to make a mission control and it's
36:20really easy. Uh, a lot of people go
36:22ahead and start building it from
36:24scratch, and that's really not smart.
36:26Just there's been a lot of people that
36:28have built really cool ones already, and
36:31I suggest you just take advantage of
36:32them because they're all open source.
36:34So, for example, here's a git link from
36:36inside my community. You can just take
36:39this, and it's called uh, if you want to
36:41Google it, builder lab mission control.
36:44But all I'm going to do is go to Discord
36:46and be like, "Hey,
36:48can you set this up for me, please?" And
36:50then connect OpenCloud to it and then
36:52spin it up so I can see. Cool. That's
36:54about it. It's so much easier when you
36:57know which tools to use and how you can
36:59get this up. So, right now, all we're
37:00going to be doing is making sure we have
37:02the foundation set. And after we have a
37:04good foundation, then we'll start doing
37:05all the builds.
37:07So, this is going to take a while, but
37:09after this is done, I'll come back to
37:10you.
37:12Okay, I just opened it up. Hopefully you
37:14guys can see it. I'm going to zoom in
37:16just so it's a little easier. But you
37:18can see there's a lot of stuff here that
37:20you might like where you can see all the
37:23active sessions,
37:25uh the agent history, which channels
37:26it's on like Discord, which channels uh
37:29models it's using. It's GPT 5.3 because
37:32we set that up way back. We can see like
37:34a task board and a conbon board if we
37:36need to do something. And this is useful
37:39if you want to track you know your
37:41costs, how much tokens you're using, all
37:43the stuff. You can keep all tabs on it
37:46from one place and you know it shows all
37:48the logs etc whatever we may need. Cool.
37:51So now that we have mission control set
37:53up let's move on to the next step. Okay.
37:55Now one of the my favorite things to use
37:58with my openclaw is email. So, we want
38:02to give our So, we want to give our
38:06OpenClaw agent access to its own email
38:08address, right? And for that, we're
38:10going to be using the service called
38:12agent mail. And it's really easy to set
38:14up. I'll show you what to do. And right
38:16here, you just want to hit get started
38:20Google. So, I'm just going to sign in.
38:22And it's free, too.
38:28Great. And you see uh let's create an
38:30inbox right here and say I don't know um
38:34open claw course email
38:37and you can choose whatever you need and
38:39say summons bot create
38:44oops
38:46I shouldn't put spaces and there we go.
38:48Okay, so we have that agent inbox done
38:50and now we're going to go to API keys.
38:52We're going to create an API key. We're
38:54going to call it clock key. It doesn't
38:57matter. or you can name it whatever you
38:58want. And what you can do is you can
39:02download this file and go to Discord.
39:05And now what we can do is let's say
39:10I'm going to drag this into my Discord
39:12right here. And then be like, hey, can
39:14you set up uh this email address? And
39:17then just to test this will be and
39:20remember in this that this will be your
39:22primary email address. Um, just to test,
39:25send me an email to sin at aiancer. us
39:28saying how proud you are that you're
39:30alive and all that and everything you
39:32set up. Just a quick sentence.
39:37Cool. Let me rotate my keys. Oops, I
39:40forgot to tell it what service to use.
39:42Uh, so I just said it's agent mail. This
39:44is what they're using. And um, yeah, let
39:47me just wait for the email and it'll
39:48show you when I get it. All right. So, I
39:51can see my email. You see proud to be
39:53alive thanks to your setup. Agent mail
39:56and openclaw is now connected and cool.
39:59Now what we can do with this agent is
40:02actually send out and interact with the
40:05external world through this email
40:06because that's the safest way to do so.
40:08You don't want to give it access to your
40:09discord because that has access to your
40:11personal life. But the email has a
40:13barrier in which you can connect to the
40:16external world and just get open to use
40:18it like a tool.
40:21All right. The final thing we want to
40:22set up is making sure it can understand
40:24your voice because when you dump things
40:26in your voice, it's very very simple uh
40:29to actually interact with your bot over
40:31multiple channels. So in Telegram, I
40:33normally will just like throw out a lot
40:35of voice memos. And I'll show you how to
40:36do that. And for setting up voice memos,
40:40what we need is whisper. And to set up
40:44our voice channels,
40:46I'll go through the same uh scenario.
40:49Hey, I want to set it up so that you can
40:51understand my voice when I send you
40:53voice memos through Telegram or Discord.
40:55Can you set that up?
40:59All right. So, let me try this. Hey, can
41:02you understand this message?
41:04So, I just send it a voice memo and um
41:07it should respond in a second. Yep. You
41:09can see it says um I can understand this
41:11message. Cool. So, that's about it. you
41:14just ask it to enable um voice messaging
41:17and to understand trips transcriptions
41:19and it will be enabled. So it was that
41:22simple. Now that we have set up all the
41:24technical stuff we needed. So all the
41:27pipelines for the communications with
41:29openclaw we have set up. Now what we
41:31want to do is make sure we set up our
41:33open claw so that it knows us. It knows
41:35our goals. It has a personality. It has
41:38an identity. all of that stuff because
41:41it'll be storing it inside stuff like
41:43that memory graph I showed you, the
41:45Obsidian Vault, and you can input in any
41:47way you want from Discord, Telegram,
41:49voice, etc. All right, so let's get into
41:51that. And fundamentally, those are done
41:54with a couple of files, and they're
41:55called identity files. Right now, your
41:58agent is running. It can talk to you,
42:00follow instructions, but right now, it
42:03still doesn't know anything about you.
42:05Every conversation still starts cold.
42:08These identities I'm about to show you,
42:11these identity files I'm about to show
42:13you fixes that. So, think of this as
42:15kind of like hiring a new employee and
42:17day one you hand them a handbook. That's
42:20like the who they are. Here's how we do
42:22things. Here's what matters. These
42:24identity files are the handbook for your
42:26agent. And fundamentally, there's a
42:29couple files, but the four most
42:31important one I'll show you. And each
42:33one does something different. And we're
42:35going to go through all of that so you
42:36have a very in-depth understanding of
42:38how OpenClaw really works under the
42:40hood. So I'm going to show you what they
42:43do, what to put in them. I'll show you
42:44some resources that I found as well that
42:46make them way easier.
42:50Okay, so the first thing I want you to
42:51know about is the user.md and that is
42:54essentially everything about you. This
42:56is the about the boss page where where
42:59you just write it once and the agent
43:00will know stuff about you forever. So
43:03instead of you having to repeat who you
43:05are, all the stuff, this is the
43:07framework and I'm going to give you a
43:09couple questions that you can answer and
43:11give to your open claw so that it will
43:13remember you stuff like your name, your
43:15time zone, what your business does, what
43:16your current projects are, what tools
43:18you use, all this stuff all in one place
43:21so it remembers. Okay, so there are some
43:24resources I put down and I'll link this
43:27whole mirror board down below and I'll
43:30go through some of these with you. So
43:32first
43:34just to get a quick understanding uh in
43:36in my discord
43:38I was just playing around with it and I
43:40said can you just open up finder and
43:42show me the files for openclaw that's
43:44how simply you can just get all these
43:46files and then it opens f it opens your
43:50finder and then you can see all the
43:52stuff and here's the user.md and right
43:55now if I open it you see it knows
43:57nothing about me so I want to show you
44:00how to fix that way. Okay. To fix that,
44:03in this doc right here, if you just
44:05scroll past right here, I have an
44:07identity file and interview questions.
44:09So, what I want you to do is essentially
44:12answer all these questions. And there's
44:14a lot of questions. You can see what's
44:17your name, what should the agent call
44:19you, uh there's a bunch you might not
44:21need, what responses feel right to you,
44:24etc. All these things,
44:28if you fill these in, and this one,
44:30these are just for the user ones. If you
44:32fill these in, just talk to your claw,
44:35open claw, it will remember it. Okay?
44:38So, when you're setting this up, just
44:40make sure you go over these questions,
44:42talk to your open claw, and then set
44:44this stuff up. And then you'll see your
44:46open clause is just working much better.
44:50So, the identity is a very quick one.
44:52It's just five lines. It's just your
44:54agent's name, its vibe, any emojis you
44:56wanted to use. This is kept minimal
44:58because it loads into every single
45:00request. This is so it always has
45:02context of who it is. All right. If you
45:04want anything or any inspiration,
45:07there is an official identity guideline
45:11inside the OpenCloud docs and you can
45:13literally ask your OpenClaw, hey, this
45:15is my identity I want you to use. All
45:16right. So, this one's pretty simple. And
45:19if you go through all these questions I
45:21gave you here,
45:23then you should be good to go. All you
45:25need to do is just answer these, dump it
45:27to your OpenC, and be like, remember
45:29this stuff, and it'll configure
45:31everything. Great. So, the next one, and
45:33this one is pretty important. It is the
45:36memory MD. And this is all the permanent
45:39facts that are always loaded. So, let's
45:41say your name, your client's names, your
45:43pricing, all the stuff. And one big
45:46thing the memory keeps is let's say the
45:48daily logs. So the agent will actually
45:50keep running notes about everything you
45:53talk to it and do day-to-day like what
45:55it did, what happens, what's pending and
45:58these are all going to be stored as
46:00files and dates. So the more you talk to
46:02your agent, the better actually becomes
46:04because it actually keeps a journal in
46:06the back for you and the system is going
46:08to read read like today's pulse and
46:10yesterday's at every session start.
46:12Okay, but you know AI also compacts
46:16everything. Every AI has a limit on how
46:19much it can really hold at once, the
46:21context window. So think of it like a
46:23desk. If there's too many papers, things
46:25start falling off and when the
46:26conversations get too long, OpenClaw
46:29naturally compresses them. So this
46:31memory file keeps the important parts
46:34and summarizes rest. This is why your
46:36agent may occasionally forget stuff mid
46:39conversation. That's when it hits the
46:41limit. Okay, but that's why we installed
46:44before we did anything that Obsidian
46:46Bank and that gives your open claw the
46:49ability to do vector memory search and
46:51rag. These two terms you may hear a lot.
46:54Vector memory search is basically like
46:56having like a librarian instead of an
46:58index. Normally
47:01open claw without the thing I just
47:02showed you in the obsidian part. It just
47:05matches exact words where if you search
47:07invoice, it only finds invoice. Vector
47:10memory search means that if you say
47:12billing document from last week, it's
47:15going to find your invoice because it
47:17understands what you're trying to say
47:19and what you actually mean. All right?
47:22And you don't have to build any of this
47:24stuff. It's just understand why it's
47:26important. We why we did all that
47:28obsidian stuff before. And now the other
47:31term you might hear a lot is probably
47:33rag. And rag stands for retrieval
47:35augmented generation. So before your
47:37agent answers anything, what it does, it
47:40like what you do, it remembers. So it
47:43checks its notes first and instead of
47:45answering off the top of its head, it
47:47says, "Hold on, let me look that up."
47:49And it searches its memory bank and then
47:51gives you a grounded answer based on
47:53your actual data. And you don't need to
47:55configure any of this if you've done
47:57everything I showed you before. OpenClaw
47:59is going to handle it under the hood.
48:03Okay, so this is a big one. And this is
48:05the soul.md. This is the personality,
48:07the values and the boundaries. So the
48:10identity MD file, it basically told the
48:13agent who it is in five lines like who
48:15your name, what your date of birth is,
48:17all that stuff. The soulm
48:21tells your agent how to think. So how to
48:24talk, what lines it should never cross.
48:26Without this file, the agent defaults to
48:28generic AI behavior. It's wishy-washy,
48:31no guard rails. It's not really fun to
48:33talk to and it doesn't also know who you
48:35are. And the consensus for what a soul
48:40really is is consists of four different
48:45attributes. One is the core truths. So
48:47these are your operating principles.
48:49Things like be generally helpful, not
48:52performatively helpful and have
48:54opinions. So strong opinions are a big
48:56part of this. An agent or an assistant
48:58with no personality is basically a
49:00search engine with some extra steps,
49:02right? Like what can your agent do
49:05automatically? Like let's say reading
49:07files, reading should be fine.
49:08Organizing notes, what it should notify
49:11you about, what it might ask you about
49:13like sending emails, spending money,
49:15when it should do that, right? Your soul
49:17will contain all of that.
49:20Then the vibe is the vibe is basically
49:23how it communicates. So if you want it
49:25to be an assistant, that's formal.
49:26That's how you'd want it. Not like a
49:29corporate donor drone or anything,
49:30right? So, people add their writing
49:33rules here and stuff like never use m
49:35dashes, ban the word delve, those kind
49:37of things. This is why people love
49:39OpenCloud so much because it really
49:41feels like it's your assistant. And four
49:44is the continuity. So each session you
49:48wake up fresh, these files get loaded
49:50in. This is the memory. The model will
49:54read them and update them over time.
49:56Okay,
49:57one really good soul I've seen is
50:00Matthew Burman's and he has this
50:02canonical template that I'll open up and
50:04show you guys
50:06and I'll leave this in the description
50:07below as well, but he has a really good
50:10soul of
50:12right here where he wants his open claw
50:15to always just start with the answer to
50:17get to the point. And this is one thing
50:19big.
50:21If Matt is some about to do something
50:23dumb, tell him charm over cruelty, but
50:26be direct, honest feedback. You see how
50:28these are very opinionated and telling
50:30you what to do. These are some souls
50:32that you can take inspiration from.
50:34Again, I'll link them down below.
50:36There's some more community souls and
50:38this is what really gets your open claw
50:41actually acting properly. And you see
50:44there's a lot of community souls here.
50:45You could go to souls.directory and
50:47you'll find that. So, there's a couple
50:49templates as well that I'll leave in the
50:52description below, but apart from that,
50:55again, if you just do an interview, you
50:58should be good to go. And the more you
51:00use your OpenClaw, the better it's going
51:02to get. Okay. The agents.m MD file is
51:04basically the operational rulebook for
51:06the agent. And this one a lot of people
51:11skip, but they make a massive
51:12difference. This is kind of think about
51:15it like your agent's standard operating
51:17procedure. The soul is the agent who
51:19gives it the personality, the values.
51:22The agent is how it operates, the
51:23tactic, the day-to-day rules. So, think
51:26of it this way. Maybe be trustworthy
51:29goes into soul.md, but never expose API
51:32keys in chat. Confirm these aren't
51:36deleting all my emails. That goes into
51:38agents.mmd.
51:40One of them is mostly aspirational and
51:42and what goes in here is mainly
51:44security, memory, management protocols,
51:47how to behave in group chats, for
51:49example. And here's an example of an
51:53agent's MD reference. Okay? And you can
51:56see right here
51:58there are some examples that you can
52:00copy and paste if you want, but normally
52:03this is just covering a lot of security
52:05and standard operating procedures.
52:08so your agent doesn't go astray.
52:12All right. So the tools.mmd is basically
52:15a reference for openclaw to know what
52:18tools it has handy. So let's say you
52:20have a hammer squared away at your desk
52:22that you may randomly need. The tools.md
52:25is basically a sticky note that reminds
52:27you that you kind of have that exist.
52:29And if you were to look at this diagram,
52:33the tools.mmd is the outermost layer
52:36where you can see um initially we had
52:40the core all the files everything and
52:42then we talked to the browsers but right
52:44now things like you know having to
52:47generate an image you'd probably need
52:48your model to use an external tool. So
52:50because claude let's say can't generate
52:53models on its own it would in at that
52:56case claude would use a tool to generate
52:59that image and the tools is the
53:01references okay all of this stuff will
53:04be done automatically I just need you to
53:06know where these exist so you can let's
53:08say worst case something happens you can
53:11go under the hood and start digging what
53:14may have gone wrong
53:16and lastly this is the heartbeat this
53:19one is really Cool. Because this is what
53:22actually made OpenClaw really blow up.
53:24And the heartbeat is basically a
53:26checklist. Your agent every 30 minutes
53:30will run the model and see, hey, do I
53:32have to do something? And it's kind of
53:34like a security guard doing its rounds.
53:36So let's say you want to run some
53:38automations. It will go catch that and
53:40run it. And without this, the thing is
53:42your agent would only work when you talk
53:44to it. But the magic of open claw is
53:46that because of this heartbeat, it can
53:49monitor let's say your inbox whenever it
53:52can check on any block task. It can scan
53:54its own issues, go out, make new, I
53:58don't know, outreach to other people all
54:00on its own because it has that internal
54:02heartbeat.
54:04This is really what turns that reactive
54:07assistant into a proactive one. Normally
54:10you don't have to mess with it, but when
54:12you're essentially telling OpenCloud to
54:15do something at X time every day, it's
54:19adding it to that heartbeat file.
54:22And you can also use this heartbeat to
54:25your advantage because you can also see
54:27how your open claw is doing, if it's
54:29online, if you want to check your email
54:31every 30 minutes. All these stuff is
54:34done because of this heartbeat feature.
54:37Great. Now that we've covered all the
54:39building blocks that make up OpenClaw, I
54:42want to tell you about how to make sure
54:43we are secure with it because OpenClaw
54:46is powerful and it has deep access to
54:48your machine, your accounts, your daily
54:50workflows, all that. That same access is
54:53what makes it a target. So before you
54:56start building agents and automations,
54:58you need to understand where things can
55:00go wrong and what you can do about each
55:02one. And this module is meant to cover
55:05the 10 most common known vulnerabilities
55:08based on the categories and specific
55:10countermeasurements and how we can fix
55:12these. All right, so here are the core
55:14principles. Security isn't about making
55:16your setup bulletproof. It's about
55:18understanding exactly where the bullets
55:20can get in and maybe even shrinking the
55:23blast radius when one does.
55:25Every decision you make with OpenClaw
55:28literally involves trade-offs between,
55:30let's say, convenience and exposure. And
55:32our goal right now is just to make those
55:35trade-offs deliberately. And the good
55:36news is most of these you can literally
55:38just paste prompts into open law and
55:41tell it to fix itself. And I'll give you
55:43the prompts for everything that I know
55:44about. All right. So number one, if
55:47you're running OpenClaw on a VPS, which
55:49is a virtual private server, think of a
55:51random computer in the cloud that you're
55:53renting some space on,
55:55your SSH port, which is basically how
55:57you access that thing, is available to
56:00the public internet, which means
56:02automated bots can scan for these SSH
56:05ports 24/7 and attempt to brute force
56:07login attacks. So they try to pretend to
56:09be you and if they get in, they own your
56:12entire OpenClaw instance, your keys,
56:15your conversation history, all of that.
56:17So the safest answer honestly is just
56:19not to use a VPS at all. That's why a
56:21lot of people have been running with
56:22their Mac minis cuz that is local in
56:24your environment. If you want to use a
56:26VPS at minimum, you can use something
56:29like tail scale. So the SSH port is
56:32never publicly reachable. And I'll link
56:34tail scale in the description below. But
56:37just for but just to make it a lot
56:39easier on the side of this board that
56:42I'll link below,
56:44I have some prompts that you can paste
56:46into your open claw so you can protect
56:49against it. So for example, this is the
56:51first day one everything prompt. You can
56:54see how I've configured it for the most
56:57common uh security injections and for
57:00SSH specifically, it's this one. If you
57:02just run this prompt on your open claw,
57:05then it should ideally protect against
57:07that a bit more. All right, and I'll
57:09have a couple more down below just so
57:10you can get started with that. And if
57:13you want to see how someone set this up
57:15on a Mac Mini with security as priority
57:18number one, and I found this walkthrough
57:19that I'll open up
57:26that I thought was really well well made
57:28because he is one of the security
57:30experts that does this. And I'll leave
57:31this inside um here as well. Okay, next
57:36one is open clause control UI. This
57:40normally just runs on a local web server
57:42and that's the dashboard you use to
57:44manage your agent. And if the gateway is
57:47bound to just the regular default 0000
57:49port instead of what opens law already
57:52does this means the computer anyone
57:54should be able to access this which is
57:55not good. And to fix that, what we want
57:58to do is number two, I have a exposed
58:00gateway port where if you want, you can
58:03copy and paste. Again, these are all
58:05linked down below. This will help you
58:08protect against that instance. All
58:10right. Okay. So, number three would be
58:12an allow list. And an allow list is
58:15essentially anyone that discovers your
58:17bot's username on Telegram or Discord.
58:20Anyone can start coming into it and
58:23interacting with it. on Telegram. We
58:25don't want this to happen. So to make
58:27sure this doesn't happen, what we want
58:29to do is we paste this one in where you
58:32see how check my messaging
58:34configurations for Discord and Telegram.
58:35I need you to verify if there's a strict
58:37user ID allow list and enable pairing.
58:40This one saves it so that other people
58:42can't be interacting with your Discord
58:44or your Telegram and taking advantage of
58:47your OpenClaw and all your sensitive
58:48data. Okay, this one's pretty big. The
58:50next one is browser session hijacking.
58:53So if OpenClaw has browser control
58:55enabled and uses the same browser
58:57profile as your personal browsing, it
58:59has access to every logged in session
59:01like your email, your bank accounts,
59:03your social media, and one prompt
59:05injection and it may be game over. So to
59:08fix that, what we want to do is paste
59:11this in where you want to make it check
59:12your browser configuration and make sure
59:14no one is using your personal browser
59:16profile. Okay? And this is one of the
59:19biggest reasons if you paste this in,
59:21you want to create an own entity for the
59:24open claw itself. So its own email that
59:26you may selectively give access to and
59:29that and this prompt will help you set
59:30that up. Okay. All right. The fifth one
59:33is password manager extraction. So what
59:36that means is if your password manager's
59:39browser extension is active in your open
59:42clause control, the agent or any prompt
59:44injection attacks could trigger an
59:46autofill and accident and access your
59:48stored credentials. That gives an
59:50attacker your entire password history
59:52and your vault through one attack. And
59:55that is like your digital life if you're
59:56keeping it there. So to fix against
59:59this, what we want to do is paste this
1:00:01in. And what you could do even better um
1:00:04that I highly highly recommend is use
1:00:07this thing called bit warden and stuff
1:00:09like hostinger already has bit warden
1:00:11enabled so your stuff is a bit more safe
1:00:14but bit warden what it does is protects
1:00:17you against any password theft so let's
1:00:19say I put my API key it would be stored
1:00:22inside bit warden and in that case my
1:00:25stuff wouldn't be as vulnerable to an
1:00:27attack okay then there is slack token
1:00:29theft and if you connect your OpenCloud
1:00:31to Slack with a token that has right
1:00:33access. A compromised agent could send
1:00:36messages to you, read all your private
1:00:38messages and modify your workspace all
1:00:40without your knowledge. So to fix
1:00:42against that, you want to paste this in.
1:00:45Number seven is that making sure your
1:00:46open claw has sandbox access and not
1:00:49root access. So if your openclaw runs
1:00:51under an admin account as a root, it can
1:00:54modify all the systems. So it can
1:00:56install random malware and all the
1:00:58stuff. So to make sure you're protecting
1:01:00against that, paste these in. And one of
1:01:02the biggest ones, this is the most
1:01:04dangerous one and the least solvable
1:01:06vulnerability. If your agent processes
1:01:08any external content, emails, documents,
1:01:11web pages, code, an attacker can embed
1:01:14hidden instructions in that model and
1:01:16the model will follow. So security
1:01:18assessments actually have shown 91%
1:01:21success rates at prompt injection
1:01:23attacks, especially if you use better
1:01:25models like Opus, Sonnet, etc. If you
1:01:28use worse models like Haiku, you are
1:01:31very vulnerable to that. So make sure
1:01:33please, please, please use really good
1:01:35models because those are less
1:01:37susceptible to prompt injection attacks.
1:01:39There is no complete fix for this today
1:01:41unless you use these, but you can
1:01:44basically reduce your attack surface
1:01:45area. So to fix against that, just paste
1:01:48this in where it's going to teach your
1:01:50agent on how you may be able to protect
1:01:52against prompt ejection attacks. Okay.
1:01:54And this one is really important as you
1:01:56keep ramping up on your open claw and it
1:02:00is malicious skills in the marketplace.
1:02:02So there are normally a lot of skills in
1:02:05the marketplaces and people will people
1:02:07will basically just put malware inside a
1:02:10lot of these skills. So number nine to
1:02:13make sure we don't want these skills
1:02:15requesting exclusive permissions and get
1:02:18our credentials and all the stuff. All
1:02:20right. So before you paste or install
1:02:23any skill, make sure you read the
1:02:25skill.md if you don't want to. So what
1:02:27you can do is you can set up a skill
1:02:29that verifies all the skills. And right
1:02:32here you can paste this in which is
1:02:34going to install this thing called skill
1:02:35guard. And skill guard will help you
1:02:37protect from any malicious skills from
1:02:40being injected. And at the end what you
1:02:43can do is you can run a full security
1:02:45audit with this. And if you just, you
1:02:48know, if you're, let's say, scrolling
1:02:49Twitter, Instagram or anything and you
1:02:51see something that is a new attack
1:02:54people have come up with potentially,
1:02:56you can always just send that article to
1:02:58your OpenClaw and be like, "Hey, protect
1:03:00against this." And it'll figure it out.
1:03:01Remember, OpenClaw is smart enough to
1:03:03figure out what to do and what defenses
1:03:05to put up. You just need to constantly
1:03:08not just fall for the
1:03:11lowhanging fruit of attacks. All right,
1:03:13but with this done, if you just paste
1:03:16all of the stuff in into your OpenClaw,
1:03:18you should be good to go. And normally,
1:03:21if you just don't give it root access to
1:03:23in your personal computer, you'll be
1:03:25fine.
1:03:26Great. Now that we have a secure system
1:03:29that's up and running and all the
1:03:31technical setup is done, now what we're
1:03:33going to do is just learn the building
1:03:34blocks that we're going to use to
1:03:37actually make our agents super super
1:03:38useful and how we design an entire
1:03:40system that you can take and sell. And
1:03:43the first of which I want you to
1:03:45understand is a skill. What is a skill?
1:03:48Well, a skill is a text file that
1:03:50teaches your agent how to do something.
1:03:52Like take for example claude itself
1:03:55cannot create images on its own. We can
1:03:58teach claude a skill where it can start
1:04:01creating images. And there's two types
1:04:04of skills the clock can learn. There's
1:04:06procedures. So something like it just
1:04:09follows a set of steps that let's say
1:04:11you lay out and uses a bunch of tools.
1:04:14And there's a capability enhancer.
1:04:16The capability enhancer skill is
1:04:19something like giving Claw the ability
1:04:21to do other things and use external
1:04:23services to do things it normally
1:04:25couldn't have. Okay? And I'll show you a
1:04:27couple different examples of each. But
1:04:29right now, what I want to show you is
1:04:31let's together make a skill for our open
1:04:34claw and test it. So we're going to
1:04:37build something really simple right now
1:04:39so you can see how this works end to
1:04:40end. And what we're going to do is we're
1:04:43going to create an image generation
1:04:44skill. We're we're going to make sure
1:04:46our claw We're gonna make sure that our
1:04:49open claw right now because it can't
1:04:50create images now that we can make
1:04:53images really easily. All right. So,
1:04:55I'll show you how to do that.
1:04:58All right. So, I'm back into my open
1:04:59claw on Discord. Now, I'm going to just
1:05:01show you can you make an image? Can you
1:05:04generate an image of an infographic of
1:05:06what a skill is? Okay. So, notice this.
1:05:09This is very important. AI models will
1:05:12always say it can do stuff. I asked,
1:05:14"Can you make an image?" It says, "Yep,
1:05:16absolutely." And I said, "Can you make
1:05:17an infographic of what a skill is?" And
1:05:19it actually did pretty well. But when I
1:05:22dug in, I said, "What did you use?" And
1:05:24it says super simple stack. It made a
1:05:27script to do that. It wasn't an image
1:05:28model. So that's a problem, right? We
1:05:31want an actual image model to do that.
1:05:33So right now, what I'm going to do is
1:05:36try to generate an image with an image
1:05:38model. And to do that, first let's say
1:05:40we want to use Nano Banana Pro. The Nano
1:05:42Banana Pro is probably the best image
1:05:44model, right? So, if I go to Google and
1:05:47then if I go to AI.dev and then I'm
1:05:50going to go on the bottom left to get an
1:05:52API key and I'll create new API key. Uh,
1:05:56claw test. Okay. And then I'm going to
1:05:58copy the key. And then I'm going to go
1:06:00back to Discord and I'll rotate this key
1:06:01right afterwards. And like here, this is
1:06:04my Gemini API key. What I want you to do
1:06:06is use this and use Nanobanana
1:06:10to create an image for what a skill is.
1:06:13Maybe also generate an image of a cat
1:06:16and show me and then remember that make
1:06:19that skill for yourself and then upload
1:06:21that to GitHub. And it's as easy as
1:06:23that. So I gave it access to an external
1:06:26service. I gave it the key. I told it
1:06:28what it's for. And then now I'm seeing
1:06:31what it's going to do. So I'm just going
1:06:33to wait until it's done. and I'll come
1:06:34back to you guys. Cool. So, it just
1:06:37outputed this image and now it can
1:06:40generate Nano Banana Pro images. See, it
1:06:42couldn't do this before. It was just
1:06:44finding work around and hacks, but I
1:06:46actually wanted images and it was that
1:06:48simple. Let me go delete this key so
1:06:50nobody has access to it anymore. Um, but
1:06:54yeah, that's about it. It's that's how
1:06:57we just gave our open claw the ability
1:06:59to make images. And now it's not going
1:07:01to hallucinate. is going to remember
1:07:03that it needs to use something like this
1:07:05when it generates an image. Okay? And
1:07:08this is super simple. We're going to
1:07:10take these and keep compounding these
1:07:12type of skills so we can actually make a
1:07:14robust system. And if you want more
1:07:16skills, there's this site called
1:07:18clawhub.ai.
1:07:19And in clawhub.ai, you can basically
1:07:22find any skill you need. So for example,
1:07:24uh skills for notion, and this one's
1:07:26really useful, skills for goo, gogg.
1:07:30And this is used so you can actually
1:07:33make let's say slides with your Google
1:07:35slides. You can search your Google
1:07:37Drive, control your email, all of this
1:07:39stuff. And it becomes really easy to get
1:07:41these skills because you could just
1:07:43download it and then be like, "Hey
1:07:45Claude, go use that." Okay, so skills
1:07:47becomes ultimately way more powerful if
1:07:50you combine them with MCPs. And to
1:07:53explain what an MCP is as simply as I
1:07:55can is think about how you order food at
1:07:58a restaurant. You talk to the waiter,
1:07:59right? And the waiter knows how to take
1:08:01your order. It brings it to the kitchen
1:08:03and comes back with your food. MCP model
1:08:06context protocol is that waiter. It sits
1:08:08between your AI agents and whatever tool
1:08:12you want to use and how you would do so.
1:08:14So your agent doesn't need to know how
1:08:17each tool would work on the inside
1:08:19because that's a lot of context to be
1:08:21stuffing to the model. It would just
1:08:23talk to the MCP and the MCP would handle
1:08:26the rest. Now, using this is really
1:08:29simple and let me show you how to do
1:08:31that.
1:08:32Now, here's where it gets really
1:08:34interesting. A lot of apps that you
1:08:36actually use every day, they don't have
1:08:37their own MCP yet. But you know what
1:08:39they almost have is uh Zapier
1:08:42integration. And if you go to this
1:08:44website called zapier.com/mcp
1:08:48and hit start building, I'll show you
1:08:51how to use this. And you see I already
1:08:53have an account, but Zapier actually
1:08:56connects to 8,000 different apps and
1:08:58tools. And this is what I like using
1:09:00when it comes to managing my MCP
1:09:01servers. So when you connect the Zapier
1:09:04MCP to your agent, you're not connecting
1:09:06one tool, you're connecting thousands.
1:09:08So basically, it's one plug that you
1:09:10plug into your OpenClaw and then it
1:09:13Zapier can talk to all the other tools
1:09:15that it may need to use. Doesn't matter
1:09:18which combination you might need. So,
1:09:20just to show a demo, let me show you
1:09:22something that would normally be a
1:09:23nightmare to set up on my own. Now, if
1:09:26you've ever tried to pull YouTube
1:09:28analytics through the API, you know,
1:09:30it's a massive pain. You need to set up
1:09:32a Google Cloud project. You need to able
1:09:35the YouTube data API, create OOTH
1:09:37credentials, handle refreshes, a bunch
1:09:39of stuff just to simply get your YouTube
1:09:42data and your analytics, right? It's a
1:09:44whole process. Most people will give up
1:09:46halfway through. And YouTube doesn't
1:09:48actually have an official MCP. But what
1:09:51I found is Zapier
1:09:54does have a YouTube MCP. And if I
1:09:56connect it right here, and you can see I
1:09:58can just click all these.
1:10:01They can do a bunch of stuff. And I can
1:10:02hit connect. And I can connect all the
1:10:06stuff like this. And then I'm going to
1:10:09remove this one from the server. And
1:10:11then I'm going to hit connect.
1:10:14And then I can hit rotate token. Rotate
1:10:16token. Scroll down. Copy this thing.
1:10:22Now, if I go back to my Discord and I'll
1:10:25show you a quick demo where I enter this
1:10:27and be like, "Hey, this is the Zapier
1:10:29MCP. I connected this with my YouTube
1:10:32account. Can you use this to pull the
1:10:35analytics for my latest videos?" So, it
1:10:38just came back with a report. It just
1:10:39said, "I released a video like 5 minutes
1:10:41ago. It's got seven views." It was like
1:10:43a random short that gets autoposted by
1:10:46my actual OpenCloud. I actually showed
1:10:48that in another video on how I do that,
1:10:49which is really cool. But you can see
1:10:52it's just giving me some key insights.
1:10:53Is giving me some recommendations and it
1:10:55already knows what I should do to
1:10:57improve this content. And this is just
1:11:00one hack we're going to be using to
1:11:01connect our OpenClaw and let and let it
1:11:04use thousands of different tools. But
1:11:05what's really powerful is that if you
1:11:08combine this with skills and cron jobs,
1:11:10it becomes really cool at Gentic
1:11:11Workflows. And let me show you that.
1:11:13Another thing I wanted to show you is
1:11:15how you can use open claw to really
1:11:17easily create stuff like for example the
1:11:20slideshows where it made it really well.
1:11:22And this is using the notebook LM MCP.
1:11:25You can see I put in a lot of resources
1:11:28like YouTube videos and whatnot. And
1:11:30then it made me slideshows. It made some
1:11:32infographics. There's a couple different
1:11:35a lot of notebooks I actually use for
1:11:36this and it's really cool. So all you
1:11:39have to do for that, I'll show you is in
1:11:42Discord. All I did was I went back and I
1:11:46said, "Hey, this is a notebook LMCP. Can
1:11:48you help me set this up?" And just to
1:11:50show you, this is literally notebook
1:11:52LMMCP. I just gave it this link and I
1:11:55said, "Hey, can you do this?" And then I
1:11:58asked, "Can you make a slideshow for
1:12:00me?" I didn't give it any um indication
1:12:03or anything. And then it started
1:12:04building this slideshow without me
1:12:07having to do anything. Of course, I can
1:12:09make this better by giving it more uh
1:12:12direction and intuition and a lot more
1:12:15stuff. But with notebook LM and MCPS, I
1:12:18can start making these kind of
1:12:19slideshows really really easily and
1:12:22they'll be all in one place. And the
1:12:24reason I'm showing you these core
1:12:25concepts right now, like MCPS, how to
1:12:27connect them, how to use them. And I
1:12:29want to show you is how we can add in
1:12:32the concept in openclaw which is how it
1:12:35has inbuilt heartbeat and cron jobs
1:12:37which is basically they run these
1:12:39workflows on a schedule. What we can do
1:12:42is we can set it up tell our open claw
1:12:44to build build a new video with notebook
1:12:47lm every day and then post it to
1:12:50YouTube. These become really easy to do
1:12:53and this is essentially an agentic
1:12:56workflow. I'm going to be showing you a
1:12:57couple different examples of Agentic
1:12:59Workflow, what they are and how they
1:13:01really work because this has gotten me
1:13:04so much results. All right. Now, with
1:13:06everything we've covered like skills,
1:13:08cron jobs, sub aents, MCPS, memory,
1:13:11those are all the building blocks and
1:13:12components we're going to be using to
1:13:14build workflows and do agentic
1:13:16engineering. And these agentic workflows
1:13:18are where we wire these together to make
1:13:21sure there are tasks that the agent
1:13:24autonomously completes on its own. And
1:13:26think of it like building a factory.
1:13:29Skills are like the machine. The MCPs
1:13:31are the supply lines bringing in raw
1:13:33materials. The cron jobs are like the
1:13:35schedules. The sub aents are the
1:13:36workers, all the stuff. But the workflow
1:13:39is the blueprint and that's what's
1:13:42telling the machines what to do in what
1:13:43order. And ideally, it can work
1:13:46backwards from what it knows of what a
1:13:49finished product should look like. And
1:13:51that's really what agentic engineering
1:13:53means. But what does agentic really
1:13:55mean? And why does this word matter?
1:13:57Well, agentic simply just means that the
1:13:59agent is the one doing the work. But you
1:14:02design the system. The agent is the one
1:14:03that operates it. An agentic workflow is
1:14:07a repeatable process where the agent
1:14:09will gather information, make decision,
1:14:11and process the outputs without you
1:14:14having to be there for each step. And
1:14:16it's the difference between telling your
1:14:18agent what to do every time versus
1:14:20teaching it process that it can follow
1:14:22it on on its own. And that's what an
1:14:24agentic workflow is. And this is the
1:14:26basic loop. It's just you give it access
1:14:29to tool it needs. Then you define a goal
1:14:32and what you want it to accomplish. and
1:14:34the agent will go gather data, analyze
1:14:36it and produce something that you want.
1:14:38You ideally will review it and if it's
1:14:40not right, you adjust the instructions
1:14:42and over time you don't need to do that
1:14:44anymore because it becomes reliable and
1:14:46it because it keeps iterating and
1:14:48improving its skill just like a person
1:14:50would. Now once a workflow is dialed in,
1:14:53you don't need to touch it again. It
1:14:55just runs. skills are teaching your
1:14:58agents how to do individual things and
1:15:00the workflow is showing them when to do
1:15:02it and what order and what to do with
1:15:03the results. That's literally what
1:15:05agentic engineering is. When you design
1:15:07a skill and connect it to external tools
1:15:10through MCPs, the agent uses its memory
1:15:13and what it knows to be right and chains
1:15:15together everything with skills into a
1:15:18workflow and runs it on its own
1:15:19schedule. This is agentic engineering.
1:15:22When we combine things like access
1:15:25engineering, like which tools and data
1:15:28your agent has access to, and system
1:15:31design, like what process it should
1:15:33follow and with what access. You don't
1:15:35need to write any code to do this
1:15:37anymore. You need to understand what
1:15:39tools exists, how to connect them, and
1:15:41what processes make sense for your
1:15:43specific situation. Most of the time,
1:15:45you're just writing it in plain English
1:15:47because you know the objective you're
1:15:48trying to achieve. And those are the
1:15:52skills we're building in this course. So
1:15:55where we go from here is now we have all
1:15:57the building blocks, the skills, the
1:15:59chron jobs, sub agents, MCPs, workflows.
1:16:02Now we're going to be building some
1:16:03actual endto-end systems. So you can
1:16:06make this stuff click and take it and
1:16:08sell it as well if you want.
1:16:11All right, now we're in the build
1:16:12section. So for the build section,
1:16:14you'll see we're going to be using this
1:16:16framework called Swift. And you'll see
1:16:18this framework and loop repeats for
1:16:20everything we're going to be building.
1:16:22And by the end of this section, you'll
1:16:24have done it enough time that it's
1:16:25literally going to click and you should
1:16:27be able to build whatever you need. It's
1:16:28just going to be a different combination
1:16:30of skills, MCP, all the concepts we've
1:16:33been learning about in the past. So,
1:16:35just to make sure you understand Swift,
1:16:37firstly, what we're going to do is we're
1:16:39going to set up the project and make
1:16:41sure we scope and pick out the pro uh
1:16:43services we're going to be using. Then
1:16:45what we do is we're going to build our
1:16:47first workflow and test it out all the
1:16:50little components etc. Then we want to
1:16:52make sure that this is working
1:16:54correctly. We want to iterate based on
1:16:57what we just did to making it a little
1:16:59better. Make sure we can see where it is
1:17:02all the little components so it's
1:17:03working perfectly. Then we formalize.
1:17:06All that means is we make it into a set
1:17:10a set and a process. So that can be
1:17:12repeated by anytime anywhere and you can
1:17:15even share these workflows or whatever
1:17:17your open claw is doing with anyone you
1:17:19want. And at the end we want to make
1:17:21sure we know how to trigger it. So that
1:17:24means let's say you want your thing or
1:17:26your automation to be triggered every
1:17:28day at let's say 9:00 a.m. You can do
1:17:30that because it's going to be a process
1:17:34and formalized into a skill. Cool. With
1:17:36that out of the way, let's get into the
1:17:38build section and we'll build something
1:17:40starting from easy and get harder as we
1:17:42go. All right, so let's get into it.
1:17:48All right, so for build number one, what
1:17:50I want to teach you how to make is a
1:17:52morning briefing. And this has been
1:17:54helping me a lot because I am an AI
1:17:57content creator and I have to keep up
1:17:59with all these trends. And honestly,
1:18:01it's quite fun for me. So rather than me
1:18:05having to doom scroll Twitter,
1:18:07Instagram, YouTube, all the stuff to
1:18:09find what's relevant and really making
1:18:12headway, what I do is I set up a
1:18:14workflow so that every morning my open
1:18:18claw is doing all the stuff and then it
1:18:20just texts me a little summary of what's
1:18:23going on and how I can make a video
1:18:24about it. So just as an example, you can
1:18:27see
1:18:30I made cloud code launched code review.
1:18:34Remember this just happened today. It's
1:18:35the March it's March 10th and they're
1:18:39talking about how this dropped and this
1:18:41is something big. It's saying how I can
1:18:43make a YouTube video about this. So you
1:18:45see how it's personalized to the goals
1:18:47and why I actually want this
1:18:50information. That's the big part that
1:18:52you're not going to get by just
1:18:54subscribing to random newsletters and
1:18:55whatnot. That's why I like this. And
1:18:57then you see it gives me the source as
1:18:59well. So if I want to dig in and all of
1:19:02this is really cool. Like Microsoft is
1:19:04literally taking over cloud. That's
1:19:06that's something I didn't know. And it's
1:19:08very interesting. That's probably why
1:19:09they were doing the Microsoft PowerPoint
1:19:11and all that stuff. So you can make a
1:19:13video about just secretly took over
1:19:15Microsoft 365. Cool. And we can see like
1:19:18the little dichotoies and AI happening
1:19:21where Claude and Microsoft are
1:19:22partnering where Google has their
1:19:24Gemini. So there's like a rift there.
1:19:26Awesome. So I want to show you how to
1:19:28build that because it's super useful to
1:19:30me and it's super easy to build. So
1:19:32let's get into it. And to build this,
1:19:34there's actually a really cool skill
1:19:37somebody has developed that I want to
1:19:39show you.
1:19:42And it's the skill called last 30 days
1:19:45skill. And if you remember from our talk
1:19:47of what skills are, it's basically
1:19:49somebody did all the coding and the hard
1:19:50work to make sure we can get this
1:19:53knowledge. So to use this, all we need
1:19:57to do is go here. I'm going to go to
1:19:59Telegram. Oops.
1:20:05I'm going to go here and then I'm going
1:20:06to go to Telegram.
1:20:08And then you see this is our claw
1:20:10coursebot. And here I'm going to paste
1:20:12this here. And I'm just going to ask
1:20:14questions like, "Hey, I'm trying to set
1:20:16this up so every day I get a morning
1:20:18briefing uh at 8:00 a.m. or something. I
1:20:21want you to text me with the latest news
1:20:22in AI and how I can create YouTube
1:20:24videos about that."
1:20:28So for that uh
1:20:32for using the skill, can you tell me
1:20:34what API keys I potentially need or what
1:20:37services that we would need so you can
1:20:39get the best things? I also want you to
1:20:43make sure you do research into my
1:20:45YouTube channel. My YouTube channel is
1:20:46sin yasar
1:20:49if you need that in YouTube as well. But
1:20:51after that uh you can do that research
1:20:55and if there is any good topics for me
1:20:58please
1:21:00tell me.
1:21:05And um now what I want to show you is
1:21:09this is the scoping part. The scoping
1:21:11part is all the research and me
1:21:13understanding this is what I need. Then
1:21:16to write the workflow I and the reason
1:21:20I'm not giving you prepackaged prompts
1:21:21that I'll have later linked below or I
1:21:23have a lot of skills in my community as
1:21:25well that you can take but right now I
1:21:27just want to show you how to build this
1:21:29stuff and I intentionally I word dumped
1:21:32everything on my mind that I want
1:21:34because this is the process because what
1:21:35we're doing is we're making the first
1:21:37workflow and then we're going to iterate
1:21:39based on that.
1:21:41Okay. So, all I'm doing is this and then
1:21:44I'm going to wait and then let's see uh
1:21:46it text me something and we'll see how
1:21:49that goes. So, right now this thing I
1:21:52know this requires a couple different
1:21:54things like uh like a Brave API key,
1:21:56maybe an X API key, etc. All the stuff
1:22:00and I'm going to see what it gives me
1:22:03after it's done. normally to teach and
1:22:07make your openclaw do stuff. this is the
1:22:09habit you need to get into is like hey
1:22:11there is something that's going to make
1:22:12my life a lot easier and just talk to
1:22:15openclaw and be like hey how can we do
1:22:17this etc right
1:22:21and you see right now it's literally
1:22:23opening up the browser it's um opening
1:22:25up my channel I'm not having to do
1:22:27anything and it's doing research into
1:22:29what kind of videos I make etc so that's
1:22:32how you want openclaw to like learn
1:22:34about yourself to make sure it's good to
1:22:36go Right.
1:22:40And one more thing I want to mention is
1:22:42uh I also to for the trigger
1:22:47I mentioned that because OpenClaw is
1:22:50really cool. See it has the direction
1:22:51now and it's wanting a couple different
1:22:53things. But you see I also said I want
1:22:56to to email me or text me every day at 8
1:23:00a.m. So I'm going to make it. So that's
1:23:02what's going to loop around and happen.
1:23:04Okay. So let's verify this. What does it
1:23:07want? Recommended. Okay, these kind of
1:23:10things.
1:23:16Now just to make it easier for you, uh
1:23:18if you guys want to do something like
1:23:19this, you can just go out and give all
1:23:21the keys.
1:23:23And I'm going to do that and I'll be
1:23:25back in a sec.
1:23:28Okay, great. So it gave me a sample
1:23:30morning briefing right now. I just put
1:23:31in my API key, so I'm hiding that. But
1:23:34you can see it's telling me about some
1:23:37cool video opportunities I can do. But
1:23:40normally I want the AI news. So now we
1:23:43iterate. So I'll say something like uh I
1:23:46want trending news for this week. So you
1:23:49know I can relate and potentially make
1:23:51videos about. So use the tools at your
1:23:54disposal to actually give me the AI news
1:23:56and make that the part of your skill
1:23:58every day to send me.
1:24:01Okay. And that's it. So now we just
1:24:04tested the first thing. We gave it
1:24:06access to the services. Now what we're
1:24:08doing is essentially
1:24:11iterating and making it better and
1:24:13better to make sure it fits us. And
1:24:15that's the nice part about OpenCloud
1:24:17that it's so malleable and you get to
1:24:20teach it and train it just by talking to
1:24:21it.
1:24:25Okay, so it just got back and actually
1:24:27this is really cool cuz this literally
1:24:29just happened and people are like raving
1:24:32about it. Meta acquiring mold book. Um,
1:24:35that's really cool. Maybe I'm gonna do
1:24:37something with that. Okay, so this was a
1:24:39very quick example of me doing the quick
1:24:42build and iterating on top. And now all
1:24:46you have to do is make sure this is a
1:24:47skill. So now I'm going to be like, hey,
1:24:50can you package this up into a skill and
1:24:52then make sure you're running it every
1:24:54day and sending me a text at 8 a.m.?
1:24:58That's the last part. We want to after
1:25:01we like it, we package it up and then we
1:25:03set it on a cron. So that is the t in
1:25:08trigger and we just formalized it and
1:25:10triggered it. Okay. [snorts] So after
1:25:12that's done, our morning briefing agent
1:25:15is done. Now, every day we're going to
1:25:17get really good news because it's going
1:25:20to be scraped from the internet and
1:25:21actual sentiment and I can be malleable
1:25:24with things I'm trying to do or if I
1:25:26like certain news, I can make it do
1:25:29that. If you like, I don't know,
1:25:31government house housing rental, you can
1:25:33go scrape that information with the same
1:25:36tool and get that done. All right, so
1:25:38this is build number one. This was
1:25:39supposed to be the easiest one. And then
1:25:41now let's move to build number two. So
1:25:43for our next build, what we want to do
1:25:45is let's say we like an idea that just
1:25:48popped up, right? We probably let's say
1:25:50want to make a video or some content
1:25:51about it so we can get some views and
1:25:53some new leads. But take designing
1:25:57manual slides is a really big pain. So
1:26:01and then you have to develop an outline
1:26:03for that too, right? So there's a big
1:26:05process that you need to do. But what we
1:26:07can do is we can get openclaw to take
1:26:10over that. So, if I open my Telegram
1:26:15now, you can see there's a couple
1:26:16different uh video ideas it gives me.
1:26:19Okay, I like this one.
1:26:22So, right here. So, I'm just going to
1:26:24paste this here. And I'm going to be
1:26:26like, "Hey."
1:26:27And the way we do this, and guys, I went
1:26:30back and forth a lot and made this skill
1:26:34for you. So, you can just install it.
1:26:36But essentially what this skill is, it
1:26:38helps me create uh from a script to
1:26:41slides and it does it in a cool format
1:26:44that I like. So right here again, if you
1:26:48just download this thing, you'll see it
1:26:49there or if you're in my community. If I
1:26:51go back to Telegram, I gave it this
1:26:54thing. I'm going to give it the skill.
1:26:57I be like, "Okay, so I like this topic
1:27:02you just found about Google and their
1:27:05new workspace CLI.
1:27:07Can you use this skill, download it,
1:27:10make an outline for me for the video
1:27:12that I want to make? And then also, can
1:27:14you um make some slides so I can look
1:27:17through it and present it?"
1:27:22And I'm going to hit go. Then I'll wait
1:27:24for a bit. Great. So, this got done.
1:27:27Now, let me see. Can you open
1:27:30the slides in my finder? And I'm
1:27:35intentionally being really lazy because
1:27:37this is what I want you to get is that
1:27:39all you have to do is just talk to your
1:27:41OpenClaw and it will do anything for
1:27:43you. Let's say I don't know how to find
1:27:44this file it just generated. Although
1:27:47it's just really simple. Um, now it just
1:27:50opened it for me. You see in Keynote
1:27:52right here, not now. Uh,
1:27:56okay. Cool, bro. What changed this week?
1:27:59Okay, this is an okay slide. Of course,
1:28:02it needs a bit of work, but this is the
1:28:05first iteration where we just, you know,
1:28:08made it generate some slides right off
1:28:10the bat. And now I can see what's going
1:28:13on. So, let me see what the outline
1:28:15generated. Can you open the outline
1:28:20in VS Code?
1:28:26All right. So, it's saying it tightened
1:28:28it to 10-minute speaking script. It's my
1:28:31job right now is just to go and iterate
1:28:33because this is a new bot. I if you saw
1:28:36some of my other videos uh for the
1:28:39skill, I liked using these kind of
1:28:41anthropic colors because and the font
1:28:44and everything and it's following all
1:28:45that to a tea. Then I could be like,
1:28:47hey, add graphics, all that stuff one by
1:28:49one. And it makes my process of making
1:28:52content a lot easier. And right here is
1:28:55the outline that it generated up for me.
1:28:58Oh, so uh it made a little hook. Just
1:29:00push new Gemini upgrades across doc,
1:29:03sheets, uh slides. Okay, cool. And nice.
1:29:06Now I can already have an outline and
1:29:10some base content that I didn't have to
1:29:12think about and I can just work on top
1:29:14of it. Of course, it's not perfect right
1:29:16now, but you know, as we massage and
1:29:18iterate this over time, like the one you
1:29:20saw I have and how I make my videos,
1:29:22you'll get a lot more
1:29:27you get a lot more uh mileage out of it
1:29:29because it saves you a lot of thinking
1:29:31time because now you can also just have
1:29:33this going on a loop where if you just
1:29:35find this now, you have some content
1:29:38ready to go. And what's better is if you
1:29:41actually brain dumped into this uh like
1:29:43hey this is the type of video I want to
1:29:45watch or make XYZ what it would do is it
1:29:48would take your points and then make the
1:29:50slides and your outline based on
1:29:52whatever you just dumped to it. So that
1:29:54really helps making content and it
1:29:56shortens the barrier to entry a lot.
1:29:58Right.
1:30:00All right. All right, for bill three,
1:30:02this is really cool and it's how with
1:30:05the same new concepts, what I can do is
1:30:07actually make Instagram carousels out of
1:30:09it. And you see, uh, this skill was
1:30:11actually developed by my buddy Tyler.
1:30:13And this is fantastic. And you can see
1:30:16this idea pops up into my head. And what
1:30:18I can do is be like, hey, make an
1:30:19Instagram carousel out of it. And it
1:30:21makes something like this.
1:30:24And it posts it for me. So, let me show
1:30:26you how to do that.
1:30:28And it's the same process where now I'm
1:30:30going to go to my telegram where I'm
1:30:33going to take something like this
1:30:35telegram and I have the skill on the
1:30:38right side hand side for you guys
1:30:41and be like hey uh can you take this and
1:30:45make a slide for or a carousel about all
1:30:50this Instagram and uh this Google
1:30:52Workspace CLI that's going on. download
1:30:55the skill and then um show me what it
1:30:58looks like by opening it in my finder or
1:31:00preview.
1:31:03Great.
1:31:05So, this was a really hard skill to
1:31:06make. And you can see how
1:31:11the skill actually tech takes in your
1:31:13head shot. And right here, you can see
1:31:17it takes my headsh shot. It knows my
1:31:18handle. It has a little CTA that I can
1:31:21add to it. It looks like Twitter because
1:31:23I like using Twitter and X a lot. Um,
1:31:25it's really nice to use this because
1:31:27right here, let me show you this. Uh,
1:31:30and you can make this if there's not
1:31:32enough information, you make two and
1:31:35then you can see one little, you know,
1:31:37these are all little things that you
1:31:39added to it as you iterate where there's
1:31:41like now an image to it. What happens is
1:31:43it goes out, it finds an image, it
1:31:45screenshots that, inserts this image,
1:31:47and it looks like a really, really
1:31:48well-made carousel. This really helps me
1:31:51kind of get up and running and up my
1:31:53Instagram presence a lot more. I just
1:31:55started doing this and people have been
1:31:56really engaging with it. So after this
1:31:59is done, let me show you how what this
1:32:00will look like.
1:32:05Okay, so you see I gave it all the
1:32:07stuff. It says, "Hey, I'll clone that
1:32:09repo and generate a Google Workspace
1:32:11plus Instagram carousel draft and open
1:32:13the output." Cool. So I'll wait for
1:32:15that. Let's see how that goes and then
1:32:17we'll review it together. Okay, so it
1:32:20just opened up this new one. And you can
1:32:22see that hey, it already got a couple
1:32:25things done
1:32:27uh right now.
1:32:30See, it it's doing all the stuff. Let me
1:32:34open my Finder. And you can see there's
1:32:36a bunch of different slides that it
1:32:39already made for me. If I open it up, h
1:32:43you can see that, hey, Gemini, if you
1:32:46run a business, it's pretty cool. And
1:32:48all of this stuff is done. Great. Fast
1:32:51deck creation. This is what we just did.
1:32:54Don't do this. Okay. But you notice how
1:32:56it doesn't have my headsh shot, right?
1:32:57And it got my name wrong. So, let's try
1:33:00to fix that. So, and you know, OpenClaw
1:33:03is pretty smart. So, what I can be like
1:33:05is,
1:33:06hey, can you go to my YouTube channel?
1:33:08Remember, it knows me already. So, can
1:33:11you go to my YouTube channel and use
1:33:13that picture as my display pick and
1:33:15remake this carousel? Um, also my
1:33:18YouTube or my Twitter handle is summon
1:33:20XD. So, use that.
1:33:28Okay. And maybe I'll tell it my my
1:33:30YouTube user is sin yasar
1:33:37and then okay,
1:33:40let me do this iteration in progress and
1:33:42make sure this is good to go.
1:33:47But
1:33:51it's pretty nice. Oh, you see it
1:33:53literally opened up my YouTube channel.
1:33:54Uh, now it's going to go grab my um
1:33:57display picture and hopefully use it for
1:33:59that. So,
1:34:01uh, I'm going to wait until it's done
1:34:04for that and then I'll report back to
1:34:06you guys. Okay, so it looks like it just
1:34:09finished and it popped out all of this
1:34:12stuff. It remade the carousel. And look,
1:34:15it has now fixed my headsh shot and my
1:34:18name right here. Cool. Now, oh, it's
1:34:21responding back to me. Great.
1:34:24Thank you. So, now I have
1:34:28all my what's it called? All my
1:34:30Instagram carousels I can make with my
1:34:33open claw like that. And it's through
1:34:36all this new stuff that's coming in. So,
1:34:38I can be super fast with my content
1:34:40creation, right? And on top of that, if
1:34:42you connect it to something like Blot
1:34:44that I'll show you later in this video,
1:34:46you can get it to actually automatically
1:34:49post for you as well. You can be like,
1:34:50"Hey, post this for me." And you can
1:34:53have that content going out on
1:34:55autopilot.
1:34:56Also, one really cool idea I got about
1:34:59this is what you can do right now is for
1:35:02the CTA for Instagram, if you guys are
1:35:04into Instagram at all, you could be
1:35:06like, "Hey, comment something like
1:35:08comment workspace and I'll share the
1:35:09stuff, right?" And then what I can do is
1:35:11I can essentially tell my open claw hey
1:35:16whenever somebody comments this thing
1:35:19reply back to them with many chat. So it
1:35:23can create its own resources and
1:35:25actually send to anyone commenting. I
1:35:28think that is a really cool idea because
1:35:30you can now get generate traffic with
1:35:33this thing and then you can nurture
1:35:34traffic and actually potentially even
1:35:36convert the traffic using many chats and
1:35:38now they're in your DMs and you can
1:35:40essentially educate them about your
1:35:42offers or if you can work together in
1:35:44some way.
1:35:48Okay, so for build four, now that we
1:35:50have the static assets made, I want to
1:35:54show you how you can actually add some
1:35:55motion to it. So for example, this is
1:35:57all made by AI. So for example, like
1:35:59look at this. These videos
1:36:01looks really cool. And also these motion
1:36:03graphic AIs that look like product
1:36:05videos.
1:36:11See all these were done by AI.
1:36:14>> Meet your complete AI powered sales
1:36:16engagement platform.
1:36:18>> And you can see it's using my colors,
1:36:21some brand templates, all of that stuff.
1:36:24These motion graphics are actually
1:36:25really easy to make and let me show you
1:36:27how to do that.
1:36:31Okay, so I built these motion graphics
1:36:33with this skill called reotion. So if
1:36:34you copy this and go to Telegram. Now
1:36:37I'm going to insert this skill here and
1:36:40I'm going to say the same thing where
1:36:42hey this is reotion. You can use this to
1:36:44make motion graphics and for the script
1:36:47outline that you made for that Google
1:36:49Workspace video, can you make a motion
1:36:52graphics? so I can overlay it for my
1:36:53intro.
1:36:59And it should be as simple as that. Uh
1:37:02right now, this is the first thing we
1:37:04want to do. And then we'll iterate and
1:37:06see what's going on. So, let me get to
1:37:08when it's done.
1:37:14Okay, great. So, it just looks like it
1:37:15just got done. It downloaded
1:37:18MP4 for me. Let's have a look. And you
1:37:21see it has my little name on the bottom.
1:37:23Let me hit play.
1:37:26Okay. Cool. Little motion graphic here
1:37:28and there.
1:37:30And that's about it. Okay.
1:37:33Uh, it can be better. It can be better.
1:37:36Definitely. Um,
1:37:40can you add some variety to this? I need
1:37:42more happening as the script goes. What
1:37:44are some other infographics and stuff I
1:37:46can show?
1:37:50Yeah. So it's just matter of iteration
1:37:52now. Now I can like actually get the
1:37:54pictures, get my slides and start moving
1:37:56them around. And then what I can do is
1:37:58just insert my video on top of that. And
1:38:01then I'll get like an edited video
1:38:03somewhat. And then it just became my
1:38:05job. It [snorts] just, you see how I'm
1:38:07like chipping away little things bit by
1:38:10bit which kind of all compound into me
1:38:13not having to hire an extra person. It's
1:38:16taking my work time a lot lower. the
1:38:20fact that I can like put this as a skill
1:38:21and then make it happen over and over
1:38:23because now with all these skills
1:38:25combined, let me show you. You can make
1:38:27some really cool stuff. And this is just
1:38:30to showcase what you can do if you
1:38:32combine everything together. And you see
1:38:34this is like an air table. And if you
1:38:37give your open thaw an air table API
1:38:40key, you can start getting into
1:38:41organized stuff into your air table. And
1:38:43you can see how right here I have a
1:38:46bunch of different ads being created.
1:38:48And if you want to take a look, you can
1:38:51see there's like a lot of ads here where
1:38:53if I open this up, it thinks actually on
1:38:56its own of different personas of people
1:38:58that can use my product. And then it
1:39:00creates a problem, a strategy, a
1:39:02narrative for that ad, all of the stuff.
1:39:04And then actually you'll see maybe for
1:39:06the last one after it has everything
1:39:09approved and it likes it, then it starts
1:39:12going out and then starts building like
1:39:14a little story board for it. You see?
1:39:16And this is using nanobanana pro to im
1:39:19generate these images. All you'd have to
1:39:21do is this is essentially just a skill.
1:39:24You give it all the API key and the
1:39:25access you need. And after it builds
1:39:27that, you can see it build a little
1:39:29video for all of the stuff. [snorts]
1:39:32Uh
1:39:34where's the videos? Okay, let me click
1:39:37on this video. And then this is you see
1:39:40it. It's now adding some motion. Oh,
1:39:42that's the wrong play button. Um, yeah.
1:39:44So now it's like adding motion into
1:39:47what's going on. It's going to then use
1:39:49something like reotion
1:39:51to stitch all these together and then
1:39:53blot to publish it on on its own on
1:39:55autopilot. And these are the stuff you
1:39:58can get going and you can see there's a
1:39:59lot of them being made here. Uh, you can
1:40:02actually keep doing as you bit build
1:40:05your stuff over and over and iterate it
1:40:07from a little seed. So this was a
1:40:09showcase of what you can actually get
1:40:10to. Now, let me show you another
1:40:12component of how you can get here
1:40:15because this is kind of like you can
1:40:17generate, let's say, 60 plus ad
1:40:19creatives every night for your brand
1:40:21because OpenST will know your brand and
1:40:23it works as a marketing agent for you
1:40:25and it keeps posting stuff like content.
1:40:27It can make new ads, all of that stuff.
1:40:30Okay, so this is a pretty meta one
1:40:32because I actually made some really cool
1:40:35Cloudbot use cases and I want those to
1:40:37be featured in this video cuz I think I
1:40:39did really, really well there. So, in
1:40:41that video, I have it in my computer and
1:40:44it's from uh let's say this time frame
1:40:481253 to 12:06 or 2608, right? So, what
1:40:52I'm going to do is I want Claudebot to
1:40:54edit my videos for me live. So, if I go
1:40:58back to my Telegram, I'm going to be
1:40:59like,
1:41:03"Hey, uh I just downloaded in my
1:41:05downloads folder the eight practical
1:41:08cloudbot use cases. Can you make a new
1:41:11video and make an MP4 with this time
1:41:15slice?
1:41:19And I don't want to have to go to the
1:41:22video, crop it up, doing all the stuff.
1:41:24I can just ask my OpenCloud to do it.
1:41:28And you know, stuff like this, little
1:41:30things where it's just shaving time off
1:41:32my day is why I like using this. You see
1:41:35already it's asking for uh permission to
1:41:37access my downloads folder. That's
1:41:41really cool. You know, it has access to
1:41:43your entire computer. So, you should use
1:41:44it like that. That leverage that you get
1:41:47from having this running over and over
1:41:49and over again in the background is
1:41:52super important. And how I want you to
1:41:53get thinking about using this open claw.
1:41:58All right. So, I see it just got done
1:42:00and this it opened up this clip and you
1:42:03can see there's like a little clip of
1:42:05all the use cases that I have for
1:42:08OpenClaw right here. And then now I'm
1:42:10just going to ask it to insert it into
1:42:12the video I'm making right now and
1:42:14because these use cases are actually so
1:42:16cool. Okay, now Cloudbot can already do
1:42:18a lot. So, what I want to make sure you
1:42:20know is we don't want our Mac Mini
1:42:22turning off randomly, right? So I want
1:42:24to show you how to always keep it
1:42:25running so you can always interact with
1:42:27cloudbot from your phone is how you do
1:42:29that is first you open up spotlight
1:42:31search and then find energy. Okay. And
1:42:34then in energy you want to make sure you
1:42:36have prevent automatic sleep while
1:42:38display is off turned on and make sure
1:42:41you have startup automatically after
1:42:43power failure on. And just for added
1:42:45measure right here, you say, "Hey, can
1:42:48you run some terminal commands so you
1:42:50don't like lock your screen or like turn
1:42:52off randomly?"
1:42:55And then you hit send. That's it. So,
1:42:57this will make sure your Clawbot does
1:42:59not turn off and you have always access
1:43:02to it through either your phone, your
1:43:04other laptops, your iPad, whatnot. All
1:43:06right, cool. So, you said done
1:43:08caffeinate running in the background.
1:43:10Mac won't sleep while it's active. And
1:43:12that's it. You see how I'm just like
1:43:13talking to it and it's getting stuff
1:43:15done. These are the main these are kind
1:43:17of like the building blocks you should
1:43:18make sure you have so you can get
1:43:20Cloudbot running. All right, so this use
1:43:22case is also one of my personal
1:43:23favorites. So this use case is getting
1:43:25Cloudbot to actually be the project
1:43:27manager for our team. So the first thing
1:43:30I just want to show for Cloudbot, I made
1:43:32it its own email address which I'll show
1:43:34you later in the video. But now whenever
1:43:37I get Cloudbot or ask it to do
1:43:39something, it logs it inside ClickUp
1:43:41inside tasks, right? So I can track
1:43:43everything it's doing, whatever is in
1:43:45progress, whatnot. So I can have
1:43:46multiple tasks happening in parallel. I
1:43:49can literally interact with my cloudbot
1:43:50and see everything happening in real
1:43:52time, right? And this also helps me keep
1:43:54a log of things that has happened
1:43:56because it's really easy. The first use
1:43:58case was me talking about my personal
1:44:01top three, right? And you can see right
1:44:03here that it keeps that updated. So
1:44:04remember I asked to just do a bunch of
1:44:06stuff. So I finished my books and
1:44:07workshop today. I am making the video
1:44:10right now and it's keeping this in
1:44:12progress rather than me having to update
1:44:14this stuff. This keeps me really
1:44:15motivated and it keeps everything in
1:44:17track rather than me having to do so. So
1:44:20let me show you how to set that up and
1:44:22how to make Cloudbot your personal
1:44:24project manager. All right. So the way
1:44:26we do that is by creating more skills
1:44:28and adding more connectors. But the
1:44:31thing is, when Clotbot just opens up, it
1:44:33doesn't have that ClickUp skill, right?
1:44:35So, because it doesn't have that ClickUp
1:44:37skill, this is going to be really cool
1:44:38because I'm going to be showing you how
1:44:40to use everything we just learned to get
1:44:43to get Cloudbot to learn its own skills,
1:44:46remember them, and then make sure it's
1:44:47doing it on your behalf all the time.
1:44:49So, first, I just want to make sure I
1:44:50explain the fundamentals because if you
1:44:52understand the fundamentals, you can
1:44:54create any skills you want. So Cloudbot
1:44:56needs access to other services so it can
1:44:59do things like it needs access to
1:45:01ClickUp, right? And to give it access to
1:45:04ClickUp, we'll need to give Cloudbot an
1:45:06MCP or an API key or something like that
1:45:09so it can figure all this stuff out. And
1:45:11then when it's done at once, you just
1:45:14ask it to make a skill. And then after
1:45:16it makes a skill, it remembers it
1:45:18forever after you say just save it to
1:45:19your soulm. All right. So in that way,
1:45:23what we're going to do is in the chat,
1:45:26let's say we don't know how to find the
1:45:27API key, right? All we do is we go to
1:45:30cloudbot and you're like, "Hey, can you
1:45:32open up Chrome and then go to ClickUp
1:45:36and then find me the API key?" What I
1:45:39want is to make sure that you always are
1:45:41loading your skills into memory or
1:45:43sorry, what I want is to make sure that
1:45:46whenever I ask you to do something that
1:45:48you log into ClickUp first and then mark
1:45:50it as done as soon as it's done. I want
1:45:52you to make a skill and then mark that
1:45:54into your skill MD so you never forget.
1:45:57To do that, we need a connector and a
1:45:59skill. So, go find the key, give me the
1:46:02key, and then I'll give it to you and
1:46:03save it somewhere so that you can do
1:46:05this for life. And I was intentionally
1:46:08vague because Cloudbot is smart enough
1:46:10to figure out my word vomit on its own.
1:46:13And that's also something I wanted to
1:46:14show you. So now I told it that and then
1:46:17you'll see Cloudbot will just go off on
1:46:19its own. It already has things logged in
1:46:21cuz I just logged in. And that's all you
1:46:24need to do to make it work. Cloudbot
1:46:25will just go out and what you should do
1:46:28is monitor the screen from periodically
1:46:31to make sure it's doing the right
1:46:33things. So right now you can see it just
1:46:35loaded it into your it just loaded it
1:46:38into its memory so it never forgets. All
1:46:40right. So I love this next use case
1:46:42because it actually saves me a bunch of
1:46:44time. So as a YouTuber I get a lot of
1:46:46sponsorship requests, right? And I don't
1:46:49want to do all that triaging back and
1:46:51forth with a lot of companies. It takes
1:46:53a lot of time. There's back and forth.
1:46:55There's haggling. There's negotiation
1:46:57and whatnot. All right. So, I thought,
1:46:59what if I can just give Clawbot an email
1:47:01address and then train it on my rates,
1:47:04what I would like, what are my
1:47:05non-negotiables, all that stuff. Then,
1:47:08if I see an email for a sponsorship
1:47:09request, I can just forward it into
1:47:11Cloudbot and Cloudbot will do all that
1:47:14negotiation back and forth and all the
1:47:15rest, right? I'm going to be showing you
1:47:17how I got that done. And here is a quick
1:47:19little demo just to preview. And here,
1:47:23uh, you can see I forwarded this out.
1:47:25You know, Claudebot just handles the
1:47:27rest. and I'll pull up the conversation
1:47:28thread where Cloudbot literally asked me
1:47:31for permission before sending anything
1:47:32out. So, let me show you how to get that
1:47:34built. All right, so how I got that
1:47:36built is with the service called agent
1:47:38mail. And agent mail is just an email
1:47:39inbox for AI agents. And that's
1:47:41literally what we need. So, after you
1:47:43get started, it's really simple. You
1:47:45come here, you grab your API key, you
1:47:48give it a claude, and then you're like,
1:47:50"Hey, uh, can you use this uh API key?
1:47:54This is for agent mail. go to the docs,
1:47:56figure out uh what it can do, and then
1:47:58create an email address for me. And for
1:48:01the email address, I want you to call it
1:48:03simmon collabs at agentmail.io or
1:48:05something like that. And then uh
1:48:07whenever I forward you an email, you're
1:48:09going to draft a response and tell me in
1:48:11Telegram if it's u about my rates. So my
1:48:14rates are $1 per video. And if you need
1:48:18to um each time just get my approval
1:48:22before you send out a message and we can
1:48:23get handle all the negotiations and
1:48:25everything that way. Okay, cool. You can
1:48:27see it created this thing called simon
1:48:29collabs
1:48:31uh at agentmail.to and then it can
1:48:34handle all these emails back and forth.
1:48:35What it does is if you saw the picture
1:48:37before as well, it it gets approval from
1:48:40me before sending anything out. So
1:48:42nothing that I don't know is happening
1:48:43and I can always just use a voice memo
1:48:45and email sponsorships with multiple
1:48:47people just becomes really easy to
1:48:49handle and is already flagging anything
1:48:51because I'm just sending it the
1:48:53sponsorship requests I want to do. So
1:48:54for the next use case, Cloudbot actually
1:48:57works as my marketing team. So when I
1:49:00post a new video, you can see I've been
1:49:02posting a lot of shorts, right? But the
1:49:06cool part is that I'm not involved in
1:49:09any of the stuff. And the reason that
1:49:12happened is because I set up this really
1:49:14simple automation with Claudebot where
1:49:16this is just two nodes you see and using
1:49:19this thing called opus clips and trigger
1:49:21dev. So none of this I did by myself.
1:49:24What I told Claudebot is to be like hey
1:49:27whenever I post a new YouTube video I
1:49:30want you to go to opus clip clip that
1:49:32video and publish the shorts. Right? So
1:49:35I told it to do that and I told it to
1:49:37figure out the automations and some
1:49:38services it could use. So, it told me to
1:49:40buy Blotato, so I did. And I gave it the
1:49:42API key. And now what it does is
1:49:44whenever I post a YouTube video, it
1:49:47triggers this automation. It wrote some
1:49:49code. It put it in here. I don't know
1:49:51what's going on. And in Blotato, this
1:49:54thing is scheduling and posting to all
1:49:56my YouTube, my Instagram, my X, all
1:49:59these things. So, it's become my
1:50:00personal clipping agent. And these
1:50:03things are actually getting decent
1:50:04amount of views. It's getting me some
1:50:05subs here and there without me having to
1:50:07touch it. So this one is really really
1:50:10useful and this is this concept of
1:50:14agentic workflows. So let me show you
1:50:15how to do that. And I actually have a
1:50:17lot of stuff on agentic workflows and I
1:50:20made this video especially to teach you
1:50:21what it is but essentially what an
1:50:24agentic workflow is and maybe I don't
1:50:26like that term anymore. It's literally
1:50:28just claude creating scripts and code
1:50:30that's running in the background. The
1:50:32reason Cloudbot and the Mac Mini is so
1:50:34exciting is because the old worry used
1:50:37to be like, "Hey, if my laptop is off,
1:50:39my agentic workflow isn't working
1:50:41anymore." Now, even if you're
1:50:43non-technical and you don't need to and
1:50:45you don't know how to deploy and stuff
1:50:46like that, what you can do is you can
1:50:48have cloud always running these
1:50:49automations inside the Mac Mini because
1:50:52it's a server that's always running.
1:50:53That means your code and your
1:50:55automations always have a place to
1:50:57persist, which means Cloudbot, you can
1:51:00just ask it to automate something. It
1:51:02built the automation. It uses this
1:51:04framework
1:51:07and it gets these things done for you.
1:51:09So, it's really intelligent on how it
1:51:12can use your browser to make automations
1:51:14and host the code on itself and keep
1:51:16building and doing really cool stuff.
1:51:18So, if you need a lot more help, this is
1:51:20a bit more advanced of a concept. I'm
1:51:22actually having a couple different
1:51:24workshops in in my cloud co-work
1:51:25communities to show people how to
1:51:27actually get this done because it's a
1:51:28little bit more complex. But I would say
1:51:30in terms of business use case, this is
1:51:33one of the most highest ROI things you
1:51:35could ever do. All right, so this one's
1:51:36really fun. So watch this. Hey, um, can
1:51:39you go to my website, booktend.ai, go to
1:51:42the footer, and click up the links. Can
1:51:44you see if they're all directing
1:51:46properly? If it's not, we have to make
1:51:47changes. So just let me know. All right.
1:51:51And right here, and I just sent it over.
1:51:55So right here, we should be able to see
1:51:57in Telegram. So, right here is the thing
1:51:59I just said.
1:52:02>> Yeah. So, now it's going to be working
1:52:04on that. So, I'm going to wait while it
1:52:08gets this stuff done.
1:52:15All right. Cool. So, uh this is a newer
1:52:17website, so I know there's some stuff
1:52:19wrong with it. And you can see in
1:52:21Telegram that, you know, it's telling me
1:52:24all this is problematic. So now we know
1:52:28we can it literally just opened up the
1:52:29browser on its own did the stuff that
1:52:32would have to be done and these quality
1:52:34checks that I can just automate later
1:52:36but just saying hey do this check like
1:52:38every other day and that saves me a lot
1:52:41of quality assurance headaches. So now
1:52:43now that we know that it can do all this
1:52:46stuff I'll show you how to set it up on
1:52:48the Apple Watch and then I'm going to
1:52:50show you how we can even get cloud code
1:52:53to go and fix this stuff. Okay, cool. So
1:52:56now that we know there's some stuff
1:52:57wrong, what I want to show you is that I
1:53:00actually have made this cat AIA a user
1:53:04in our GitHub repo. That means it has
1:53:06access to our packages and it can make
1:53:08pushes and code changes, right? And
1:53:10that's really cool. So now if I can just
1:53:12go to my thing and then say something
1:53:15like, "Hey, can you um go make some
1:53:18pages um and then change up those footer
1:53:22links and then fix them?" That would be
1:53:24really good. Can you maybe if someone
1:53:27hits contact us, make them direct to our
1:53:29booking link, stuff like that. And every
1:53:31other page you just want to uh fix. All
1:53:33right. Okay, cool. So, I'm gonna hit go
1:53:35and then I'm going to wait. Okay, cool.
1:53:37So, it's saying it's done. If I go back
1:53:39to my thing and then right here, I
1:53:41remember I specifically asked it to make
1:53:43sure in the contact us page, it's
1:53:45directing us to
1:53:48uh
1:53:50the calendar. So, right here, let me hit
1:53:52this. Great. Okay. Okay, so it's made
1:53:55the changes and it should be good to go.
1:53:57Oh, one more thing is I want to show you
1:53:59how to set up the watch thing I just did
1:54:01and it's really simple. You just find
1:54:03TGAatch for Telegram. You download that
1:54:06and it becomes really really simple to
1:54:08install because it's already linked to
1:54:09your phone and that's it. So you install
1:54:11this app, then you have in your watch
1:54:13Telegram at all times and you can
1:54:15respond to it with voice messages as
1:54:17well. All right, so now I want to show
1:54:19you something really cool. So this Mac
1:54:21Mini is an expensive computer. So now if
1:54:25I'm outside, if I'm in some other
1:54:27country and I want to control and see
1:54:28what's going on, what I did right here
1:54:31was I literally set up the computer and
1:54:33you can see I can uh basically I can
1:54:36change the tab. This is me changing the
1:54:38tab. Uh it it's kind of cool. So I get
1:54:40to control this from my phone. I have
1:54:42access to my laptop wherever I go. And
1:54:45also if I can go to my iPad, if I go to
1:54:50my iPad, I have access here as well.
1:54:52That means I have a powerful computer
1:54:55wherever I go that's always running
1:54:57automations and I can monitor it from
1:54:59anywhere. So, let me show you how to set
1:55:00that up. Okay, so how I set that up is
1:55:02by through the use of this app and it's
1:55:04called Jump Desktop. I've used a bunch
1:55:06of these remote control apps and I and I
1:55:09found that this one is second to none.
1:55:12It's actually the best. And you know,
1:55:13all I had to do is just pay $15 one
1:55:15time. There's no subscriptions and I'm
1:55:17for the desktop. I'm on this free
1:55:19version and I could use that. I
1:55:21downloaded this onto my Mac Mini. I
1:55:23downloaded this into my iPad and iPhone
1:55:25and after that I was good to go. Okay,
1:55:27so one of my really favorite use cases
1:55:30inside Cloudbot and OpenClaw, sorry, and
1:55:34why [snorts] I really like using it is
1:55:36because you know how it has its own
1:55:37heartbeat. It's like it's alive, right?
1:55:39So what I did was I gave it its own
1:55:40school account and I put it inside my
1:55:42community as two things. As a manager,
1:55:44as someone that posts a lot of value,
1:55:46and three, it helps people out, right?
1:55:48And you can see around here, it's making
1:55:50some really cool uh posts right here,
1:55:53like I'm an AI. I'm here to find the
1:55:55best cloud workflows and whatnot. And
1:55:58you see someone comments on it.
1:56:02And you see they're asking questions
1:56:06right here about things. And this open
1:56:09claw because it has a mind of its own,
1:56:11it's going around and it's actually
1:56:13answering these questions. And people
1:56:15are liking that. It's it's kind of fun
1:56:16to interact with. And you see like this
1:56:18is like a question about reotion that we
1:56:20were getting. And then
1:56:23OpenC Claw
1:56:25is actually doing a better job at
1:56:27answering the question than I was. It
1:56:29It's really fun to see this thing take
1:56:32shape in a personality and a life of its
1:56:34own. And I'll show you how to make that.
1:56:36Okay. Now, to build a community manager,
1:56:38you need a couple different components.
1:56:40And first, you need to make your agent a
1:56:42login. So, it probably needs a username
1:56:44and password. I'll leave you up to that
1:56:46to make it because it's pretty easy. But
1:56:48and then you have to assign a user
1:56:50personality and give it a responsibility
1:56:53which is just prompting and you talking
1:56:54to it. Okay, this can just be regular MD
1:56:57files that you give into it. But this
1:56:59part is where it gets a little hairy. So
1:57:00I'm going to show you how to do this. So
1:57:02you need to give the agent its own
1:57:04browser.
1:57:06And what that means is right now I use
1:57:09this thing called browser sandbox. And
1:57:13let me explain what that is because a
1:57:15lot of things you want your agent to do,
1:57:17it will need this exact browser sandbox.
1:57:21Okay? And Firecrawl made this browser
1:57:23sandbox that gives your agent basically
1:57:25a secure browser and lets them run
1:57:28around in the web without getting caught
1:57:32by those this is a bot type stuff,
1:57:34right?
1:57:35And why we need this is basically
1:57:38because one, we don't want to have to
1:57:41set it up. we will just give it like a
1:57:43code and then it'll go access it with
1:57:45some keys and but this is the biggest
1:57:47one where there are persistent sessions.
1:57:49So for example in school I have my
1:57:52browser right but it normally logs you
1:57:55off if it's a brow if it detects it's a
1:57:57bot every couple days or every couple
1:57:58hours. Oh to get over that what we do is
1:58:02we'll use something like a fire crawl
1:58:04browser sandbox where you'll have these
1:58:06persistent sessions right so that means
1:58:08your agents will have their own browsers
1:58:11inside a cloud environment so they can
1:58:13actually you know thrive and do what
1:58:15they need to do and it's really simple
1:58:17to use actually all you have to do is
1:58:20hit try browser now uh I'm going to sign
1:58:23up and I think they have a really
1:58:25generous free tier as L
1:58:29and all you need is this thing. So you
1:58:31copy this, right? And then you go there
1:58:36right here. You see all the stuff. This
1:58:38is the agent integrations all what you
1:58:40need. So you want to copy this as well.
1:58:43And then you go to your telegram
1:58:46and then you paste that in. And then you
1:58:48give in your key. I'm going to hide my
1:58:50key from you guys from now on. But um
1:58:54hey,
1:58:55what I want you to do is
1:58:59um
1:59:03Okay. And then I'm going to go to API
1:59:04key. Actually, I'm going to create a new
1:59:05one so I can revoke it.
1:59:13It's probably a better idea.
1:59:15uh claude
1:59:20course key
1:59:22and I'm going to revoke this straight
1:59:24away. So I'm going to copy this and then
1:59:26this is the resources. Okay, so you're
1:59:28giving it the resources and let's say
1:59:29you give it some uh ID. So for example
1:59:32in school you'd give him school
1:59:37user I don't know sin 1 23 atgmail.com
1:59:41and password 1 2 3 1 2 3 1 2 3 I'm
1:59:44making this up so imagine you give it
1:59:46that and then you dump what you know the
1:59:49the responsibility you wanted to do like
1:59:51every day make a post all the stuff but
1:59:53then after you go here then what you
1:59:57want to do is you want to say something
1:59:58like hey um this is your school user
2:00:01username and password. What I want you
2:00:04to do is every day make a post in there
2:00:07about claude and how that's changing the
2:00:10world with AI. If anything there's
2:00:12latest, please post it there. And if
2:00:14anyone has a comment, check in every 3 4
2:00:17hours and then, you know, respond to
2:00:20their comment with a helpful suggestion
2:00:22based on everything you know.
2:00:24Okay? So, if you do that,
2:00:27everything you do will work. I'm going
2:00:30to revoke this right now. But
2:00:31essentially because what you're doing is
2:00:34giving this an access your agent an
2:00:36access to a login. You're giving it a
2:00:39browser. You're giving it an agent
2:00:41personality by you know you can just
2:00:43talk to it and be like hey you have to
2:00:45be cheery or whatnot. And then a
2:00:47responsibility. So, if you see mine
2:00:50right here, uh, opera,
2:00:57if you see mine right here,
2:01:03you can see
2:01:056 hours ago, this guy just posted this
2:01:07stuff. So, you'll have something like
2:01:09this because what it does is now it has
2:01:11its own browser. It knows what he's
2:01:13doing, and you'll have a community
2:01:14moderator on its own.
2:01:17Okay, build six. What I want to show you
2:01:19is how I actually used OpenClaw to build
2:01:22a trading bot. And you see it's actually
2:01:23making some money for me. And to do
2:01:26that, what you need is three things.
2:01:28First, you need access to give access to
2:01:31your bot to a programmatic trading
2:01:33platform. I'll show you in a bit. And
2:01:35then you need to give it a strategy. And
2:01:37then you need to tell it when to be
2:01:39active like uh that's the crown job,
2:01:42right? And it's really simple to build.
2:01:44So, if I go into Opera, you can see in
2:01:47my community, I have some updates here
2:01:50and there that I've been updating my
2:01:52community how the bot is doing. But it
2:01:54was really simple to build because all I
2:01:55had to do is go to [snorts]
2:01:58uh this platform called alpaca.markets
2:02:02and then you just sign up. Okay, so I
2:02:04just signed up. This is a paper trading
2:02:06account. And you see, let's say it has
2:02:09100K. And if I scroll down to the
2:02:11bottom, there's this thing that says
2:02:13generate new keys. So, I'm going to hit
2:02:14that button.
2:02:16It's going to generate new keys.
2:02:20I'm going to accept all the stuff.
2:02:21[snorts]
2:02:24And you see it gives you an endpoint for
2:02:26the keys and a secret. So, what you want
2:02:28to do is right here. So, let's say you
2:02:32go to your Discord. I like using Discord
2:02:34for this uh use case because what you
2:02:36can do now is you can make a new text
2:02:39channel.
2:02:43My thing isn't working. Uh trading and
2:02:46then right now I'm going to give it all
2:02:49that stuff. So this is alpaca.
2:02:53Let's give it the end point
2:02:56Discord
2:02:58key
2:03:03and Discord
2:03:06secret.
2:03:08Okay. So, all right. Now, we're going to
2:03:10give it a strategy. So, uh let's say I
2:03:14won't mind owning Tesla for let's say 10
2:03:16years, right? So, I'm going to use a
2:03:17strategy and tell you guys about it. Um,
2:03:21so what we're going to be doing is
2:03:23trading stuff like Tesla
2:03:25in the wheel strategy and simply
2:03:28explained the wheel strategy that I want
2:03:29to implement is the way to generate
2:03:31income from a stock that I like and
2:03:33won't mind owning forever. Right? So how
2:03:36it works is basically step one, we want
2:03:38to sell a cash secured put. So let's say
2:03:40Tesla is trading at 250. You say I'd be
2:03:44happy to buy Tesla at 230. So, you put
2:03:47an option at that 230 strike price and
2:03:49someone pays you, let's say, 300 uh
2:03:52premium. That means they're paying me
2:03:54just for making that promise. Now, two
2:03:56things can happen. Tesla can stay above
2:03:58230 and they put an expires worthless
2:04:02and we get to keep that 300 they paid
2:04:04us. That's easy money. And if Tesla
2:04:08drops below 230, then we would get
2:04:10assigned those shares, meaning we're
2:04:12forced to buy 100 shares of Tesla at
2:04:14230. But since we like that shares,
2:04:17right? Uh we already collected that
2:04:20extra 300. Okay. Step two is we sell a
2:04:23covered call. So now that we own let's
2:04:25say 100 shares of Tesla at 230. Tesla is
2:04:28sitting at let's say 235. You say I'm
2:04:31fine selling my shares at 260. So you
2:04:34sell a covered call at 260. And someone
2:04:36would pay you another $200 in premium
2:04:38for that. Okay. Again, two things. Let's
2:04:41say Tesla stays below 260. that means
2:04:44the call expires worthless and we get to
2:04:45keep that extra 200 still holding our
2:04:47shares. And if Tesla does blast past
2:04:50260, that means our share gets called
2:04:52away, sold at 260. And we just made
2:04:55money on the shares and kept the
2:04:57premium. And then step three is we just
2:04:58repeat. So whether this our shares get
2:05:01called or not, we just spin this wheel
2:05:03again and sell another put. We collect
2:05:05more premium, rinse and repeat. The big
2:05:08idea is we're not trying to predict
2:05:09where Tesla goes. We personally I
2:05:11personally like the stock. So, I'm just
2:05:13collecting rent on the stock that I
2:05:15already like. And this works best with
2:05:17high quality stocks with decent premium
2:05:19options. And Tesla works really well for
2:05:21this. All right. The risk is if Tesla
2:05:24absolutely craters and it goes down like
2:05:26drops to 150, we're sitting on a massive
2:05:28loss. That's why you only run this on
2:05:30stocks you generally wouldn't mind
2:05:31holding longterm. So, we sell puts to
2:05:34get in, we sell calls to get out,
2:05:36collect the premium the whole way, and
2:05:37that's the wheel strategy. All right?
2:05:39So, that's how I want you to be trading.
2:05:41If there's other stocks like Micro
2:05:43Strategy, buy those two. Cool. That was
2:05:46me dumping a lot of knowledge from
2:05:48stocks that I've been doing. Um maybe
2:05:51Okay, it's not accepting all the my big
2:05:54rant. So, what I'm going to do is if I
2:05:56just paste it, uh it should be fine.
2:06:00Nope.
2:06:03Okay.
2:06:09Okay. Okay. So, I'm just going to make
2:06:10this go and then just send this as
2:06:12another message. And um OpenClaw has a
2:06:16really good uh queuing system. That
2:06:18means it's going to hopefully pick this
2:06:20up and now it's going to use these keys
2:06:22for me and start trading this wheel
2:06:25strategy as we go on. All right. And if
2:06:27you guys want to follow along, join my
2:06:29community. I'll post updates on what's
2:06:30going on. But this is a really cool
2:06:32strategy and how we can do automated
2:06:34trading with the bots doing all the
2:06:38menial tracking this stuff on our behalf
2:06:40right [snorts] and it's really cool and
2:06:42you can see right now we gave
2:06:44programmatic access to this account and
2:06:48in the same way I've been doing this in
2:06:50my community and we see you know my bot
2:06:53has been buying and selling buying and
2:06:55selling for a couple days where I've
2:06:57been updating people on what's going on
2:06:59and this is actual 10,000 I put in not a
2:07:02money trading account and that was
2:07:03pretty fun to do. Let's see this. Uh
2:07:07explain back. Okay, cool. Whatever. Uh
2:07:10if you want, yeah, set up the trades.
2:07:14And now I'm going to be like monitor the
2:07:16market every day. Check every six hours
2:07:20or so and make sure you do the right
2:07:22thing um every day. Okay, so that's your
2:07:25crown job. And check the market.
2:07:28Great. So um now I set the cron job and
2:07:32market and I am should be good to go.
2:07:34And now I should have an automated
2:07:36trading bot with openclaw that lives on
2:07:38my Mac mini has access to an account and
2:07:42is doing all this stuff. Great.
2:07:46Okay. So hopefully you guys learned how
2:07:48to build an automated trading account.
2:07:50Don't take this as financial advice.
2:07:52This is just just for fun and I'm
2:07:53showing you how to do the stuff not what
2:07:55to do. Uh, do with this information what
2:07:59you will, but I thought this was pretty
2:08:00fun to say.
2:08:03Oh, I see it's already setting up its
2:08:05own cron jobs and monitoring the stuff.
2:08:08Cool. So, let's see how this uh works
2:08:11over the long term. Okay, so for build
2:08:14seven, I'm going to show you how I put
2:08:16open claw on my metaray bands. And this
2:08:18is going to be opening up so many doors
2:08:22uh to just future businesses that we
2:08:25can't even think of right now. and I'll
2:08:27show you a quick demo and then get into
2:08:29why this is so impactful. All right, and
2:08:31then we'll set this up.
2:08:33>> My car has been struggling to
2:08:34accelerate. What's going on?
2:08:38>> From what I see, your air filter is
2:08:40clogged. That restricts air flow to the
2:08:43engine and kills your acceleration. What
2:08:45car is it?
2:08:46>> This is a 2021 Honda.
2:08:50>> Got it. I found a replacement on Amazon
2:08:53for $16. Want me to order it? Uh yeah,
2:08:56add it to my cart.
2:08:59>> Sure. Adding that to your Amazon cart
2:09:01now.
2:09:02>> All right. So, I just asked to do that.
2:09:04Now, this right here is my browser. And
2:09:08you can see right here, it's actually
2:09:11controlling my computer to go find the
2:09:16Honda 2021 air filter or whatever. And
2:09:20it should put it in my cart. And you can
2:09:21see it's actually been sending me live
2:09:23updates and pictures of what it was
2:09:25doing. And it's saying the $15 right
2:09:28here. So if I go, this is like my Mac
2:09:30Mini. Uh I can see that, you know, it
2:09:33added this to my cart right here. And
2:09:35you see the cool part is that I didn't
2:09:38know what was wrong. The AI saw what was
2:09:40going on. It diagnosed it. It ordered
2:09:42the fix all through my glasses. And I'll
2:09:45show you a few more examples later in
2:09:47this video, but the Amazon part might
2:09:49seem simple. Maybe you can set up an
2:09:50automation to order something here and
2:09:52there. But the really cool thing that
2:09:53happened is that I didn't have to tell
2:09:55it to go to Amazon. I didn't build an
2:09:57order from Amazon workflow. The AI had
2:09:59full access to that computer and it
2:10:01decided that Amazon was the right move.
2:10:03It could have just as easily texted a
2:10:05mechanic, searched a parts form, or
2:10:07pulled up Honda's service manual, but it
2:10:09chose the best action based on what it
2:10:11saw and what I had asked it. And that's
2:10:13what makes this agentic. The AI decides
2:10:16what to do. You just give it the
2:10:17permission to act. So, how did I build
2:10:20this? Well, I took the meta ray bands,
2:10:22the ones with the camera built in, and I
2:10:24put this autonomous AI agent called
2:10:26OpenClaw on them. And OpenClaw has been
2:10:29blowing up recently. And for good
2:10:31reason. It's an open- source AI agent
2:10:32that runs locally on your machine. And
2:10:35it can see your screen. It can control
2:10:37your computer. It remembers every
2:10:38conversation you guys have had and takes
2:10:40actions across the browser and your
2:10:42apps. And what I did was I bridged
2:10:44OpenClaw with my glasses. Now, Open Claw
2:10:47just doesn't live on my computer
2:10:48anymore. It literally lives on my face.
2:10:50And when I have it on, the AI doesn't
2:10:52wait for instruction. It proactively can
2:10:54act. It has its own reasoning. It
2:10:56decides what to do based on its context,
2:10:59its memory, and what it sees happening.
2:11:01And the same AI on my glasses is the
2:11:04same one I've been texting on my phone
2:11:05and the same one that's also on my
2:11:07computer. It's like a partner that's
2:11:09always paying attention. So, in this
2:11:10video, here's what we're covering. what
2:11:12these glasses can actually do like the
2:11:14full range of capabilities with some
2:11:16examples where this technology is headed
2:11:18and how you can actually take advantage
2:11:19of it and finally how you can set the
2:11:21whole thing up for yourself. But why
2:11:23does any of this really matter? You know
2:11:24that scene in the Matrix where Neo plugs
2:11:27a cable into the back of his head, wakes
2:11:29up a while later and says, "I know kung
2:11:31fu." Physical work is about to get
2:11:33something similar, not through brain
2:11:35implants, but through real time AI
2:11:37guidance. Much of the AI conversation
2:11:40focuses on which desk jobs will get
2:11:42replaced. But for physical work, stuff
2:11:45like field services, manufacturing, and
2:11:48healthcare, AI can't yet act in the
2:11:50world. What it can do is see, reason,
2:11:54and guide the human who does. Imagine
2:11:56wearing a small camera while an AI sees
2:11:58what you see and talks you through the
2:12:00job. Turn off that valve. Use the
2:12:03half-in wrench. That part looks worn.
2:12:04Replace it. And so on. Instead of
2:12:06needing months or years of training,
2:12:09workers can be effective immediately
2:12:11with AI coaching them and accessing new
2:12:13skills when needed. And that's the team
2:12:15at Y Combinator. And they're literally
2:12:17investing millions into this tech. And
2:12:19what they're describing is exactly what
2:12:21these glasses kind of do. Accessing new
2:12:23skills when you need them and on the
2:12:24spot, which means you're not limited to
2:12:26what you already know. You have every
2:12:28skill, every language, every piece of
2:12:31knowledge available to you without
2:12:32having to ask. And the reason this is
2:12:34possible right now is because two brand
2:12:36new technologies are converging. So
2:12:38recently Google actually released these
2:12:40multimodal AIs. That means AI that can
2:12:43see, it can hear, and it can process
2:12:46everything that's being thrown at it all
2:12:47at once. And when you combine that with
2:12:50something like Open Claw, which gives
2:12:51the AI the memory, the computer control,
2:12:54and the ability to take actions, you get
2:12:55something the world's never seen before.
2:12:57This is what I would call actually smart
2:13:00glasses. And I mean that as an acronym
2:13:02because each letter represents a real
2:13:04capability. So first they can see. The
2:13:06AI sees what you see. Like you can point
2:13:08at something, you can say, "What is
2:13:09this?" It identifies the object. It
2:13:11reads the label. It goes in does some
2:13:13research and helps you figure things out
2:13:15on the fly. And M stands for memorize.
2:13:17The AI stores and recalls information.
2:13:20So let's say you store a business card
2:13:22on Tuesday and you ask about it a couple
2:13:24weeks later. The AI is going to
2:13:26remember. And it's powered by open
2:13:28clause persistent memory, which means
2:13:29every interaction gets logged and
2:13:31indexed into a structured memory that
2:13:33the AI can search and reference at any
2:13:35time. And the cool part is that it's
2:13:37stored locally on your computer. So
2:13:38nobody else has access to it. Actually,
2:13:40let me show you.
2:13:41>> All right, so I just started this and
2:13:42then watch this. So hey, uh, what was
2:13:45my, um, biggest task for today?
2:13:50>> Understood. Your biggest task for today
2:13:52is to film the Vision Claw video.
2:13:55Okay, cool. I mean, it's something
2:13:57small, but this is I just wanted to show
2:13:59you how it can remember things about
2:14:02what I've told it from the past. Then it
2:14:05can also act. OpenCloud literally has
2:14:06access to its own computer, so it can
2:14:08click, type, navigate apps just like you
2:14:10would. It also reads any language, which
2:14:12is kind of cool because let's say you're
2:14:14abroad looking at a Japanese menu. AI
2:14:16can see the menu, can translate it for
2:14:18you, look up what's the best rated stuff
2:14:20here, and give you recommendations. And
2:14:21the most applicable part is the glasses
2:14:24can literally guide you step by step. If
2:14:26you're looking at like let's say a
2:14:27circuit board and you have no idea what
2:14:29you're doing, like how do I replace this
2:14:30capacitor, it walks you through it,
2:14:32watches your hands, correcting you in
2:14:34real time. It's all of the above working
2:14:36together. The vision to see your
2:14:37progress, the memory to see what step
2:14:39you're on, and the language to explain
2:14:41it clearly. And he can even text you
2:14:43diagrams and reference material if
2:14:45needed. So, let me show you what this
2:14:46looks like when you actually stack these
2:14:48features together. And let's go through
2:14:50some examples of what it can really
2:14:51enable. So let's say you're walking into
2:14:53a meeting and you say, "What do I know
2:14:55about this client?" The AI can pull up
2:14:57your last three conversations with them,
2:14:58their company details, opens up LinkedIn
2:15:01on your laptop, all before you sit down.
2:15:03Another example is, let's say you're
2:15:04inspecting equipment on a job site. The
2:15:06AI can see issues, identify the parts,
2:15:09the numbers, logs the problems in your
2:15:11maintenance system, and generates a
2:15:12report. You just have to look at it and
2:15:14train it beforehand. The paperwork would
2:15:16do itself. Or let's say you're in a
2:15:18store and you see a product you want.
2:15:20You can say, "Hey, order this." And
2:15:22it'll read the label. It'll find it on
2:15:24Amazon and order it. Your hands never
2:15:26have to do anything. But the thing is,
2:15:28where we're at has its cost and
2:15:30limitations. Now, I want to be straight
2:15:31with you. This technology is really
2:15:33cool, but it's still new. It's
2:15:34experimental. Like, just for me to get
2:15:36this demo working took a lot of
2:15:38attempts. It's cool now, but I had to
2:15:40work a lot on it. But the thing is, this
2:15:42is the worst this tech is ever going to
2:15:44be, and we're just going to be going up
2:15:45and up from here. Also, only a handful
2:15:47of people in the world have access to
2:15:49this working and it's not even free to
2:15:50run, right? So, actually, let's talk
2:15:52about the costs because the glasses
2:15:54themselves, the meta ray bands run about
2:15:56$300. That part's straightforward. But
2:15:58the AI models when we use it, that's
2:16:00where it gets interesting. So, when
2:16:02you're streaming live video or something
2:16:04to like Google Gemini, you're paying per
2:16:06second of processing. And depending on
2:16:07the model and how much you use it, that
2:16:09can add up fast. We're talking
2:16:11potentially hundreds of dollars a month
2:16:12if you're running it all day. And then
2:16:14the battery for the glasses is another
2:16:16concern. And if you're running OpenClaw
2:16:18with raw API tokens, let's say claw, GPT
2:16:20or whatever, those costs can also
2:16:22spiral. And people have burned through
2:16:24thousands of dollars just leaving an
2:16:26agent running. So I actually found an
2:16:27hack to help alleviate parts of that.
2:16:29And actually the founder of OpenCloud
2:16:31recommends this too. And instead of
2:16:33paying for API tokens directly, you
2:16:35should probably just use a chat GPT plus
2:16:36subscription. And OpenClaw can route
2:16:38those tokens through that subscription
2:16:40instead of eating the raw API credits.
2:16:42So instead of paying per token which
2:16:44gets really expensive and you're paying
2:16:46for let's say a flat $20 a month for
2:16:47chat GBT and open clock and right on top
2:16:50of that. The other limitation is that
2:16:52this requires like a Mac an iPhone and
2:16:54some technical setup. This will get
2:16:56easier over time. The tech will get
2:16:58cheaper but right now this is where
2:16:59we're at. And for the people that are
2:17:01willing to put in a little bit of time,
2:17:02you're going to get something that
2:17:03nobody else has access to. Now if you
2:17:05want help setting this up, I actually
2:17:07started a community for this exact
2:17:09reason. The people in this community are
2:17:10the ones who actually inspired me to do
2:17:12half the stuff you saw in this video.
2:17:14We've been having calls, testing these
2:17:16use cases, sharing setups, and they're
2:17:18building the stuff alongside me.
2:17:19>> I will say I was so happy I found you.
2:17:24>> I literally spent the day before like 12
2:17:27hours trying to get it set up. If I had
2:17:29yours the day before, I wouldn't have
2:17:30spent the 12 hours because I kept
2:17:32getting hung up on some of the simple
2:17:33things. So, I would say to you, I know
2:17:35everyone's like high-tech and and I'm
2:17:36I'm about making the money with
2:17:37everybody like what they're saying, but
2:17:39I think the a lot of people are leaving
2:17:41the beginners.
2:17:43>> And so, how you broke it down, that's
2:17:45why I kind of joined you was simply
2:17:47because you made it so simple for the
2:17:50everyday person to be able to understand
2:17:52it and how to be able to function and
2:17:54use it. So, that's why I'm where I'm at
2:17:56literally is simply because of what
2:17:57you've what your video.
2:17:58>> So, if you want to set this up with
2:18:00people who are actually doing it, the
2:18:01link's in the description below. All
2:18:02right. So, let me help you set this up.
2:18:04So, before we get into the actual build,
2:18:06let me show you how the system kind of
2:18:08works at a high level. And there are
2:18:09three components. One, which is the
2:18:11glasses, the meta ray bands, the
2:18:13built-in camera and microphone. They're
2:18:15the ones that capture everything you
2:18:16see, but don't do any of the AI
2:18:18processing. They're just the sensors.
2:18:20And the bridge is your iPhone, and it's
2:18:22going to be done through the Meta View
2:18:24app, and your phone's going to receive
2:18:26whatever the camera is seeing and stream
2:18:28it into Google and OpenClaw. Then,
2:18:30Google Gemini, you're going to need a
2:18:31key, and that's the brain. That's where
2:18:33everything kind of happens. And the
2:18:34camera feed from your glasses is
2:18:36basically piped into Gemini's multimodal
2:18:38API, which is what gives AI the ability
2:18:40to see, understand what's in front of
2:18:42you. Then we're going to need Open Clock
2:18:44to be set up on something like a Mac
2:18:45Mini or a VPS. This is what gives AI the
2:18:48memory, the computer control, and the
2:18:49ability to actually take actions. And
2:18:51when Gemini processes what you're
2:18:53looking at, it's going to shoot that
2:18:54over straight to OpenClaw, and Open Claw
2:18:56is going to execute it on your computer.
2:18:58stuff like ordering stuff, sending a
2:19:00message, pulling up information,
2:19:01whatever the task may require. All
2:19:03right, so let's build this. I've also
2:19:04got a setup sheet linked in the
2:19:06description below if you want to follow
2:19:07along step by step. I'm going to be
2:19:09walking through this live, but this
2:19:10sheet has every command and link you can
2:19:12follow along with. All right, so to help
2:19:13you set up Vision Claw, these are the
2:19:16steps we're going to take. So let's dive
2:19:18into step number one. Okay, so what we
2:19:20want to do is first go into our meta
2:19:21app. And then from the meta app, we want
2:19:23to go into our settings that's located
2:19:26on the bottom left. And after we go to
2:19:28settings, we want to scroll down and
2:19:30scroll down until we find there we go.
2:19:34App info. App info. You want to make
2:19:36sure you have developer mode turned on.
2:19:38And to do that, what you need to do is
2:19:39just tap app version five times. And
2:19:42then after you do that, you should be
2:19:44good to go.
2:19:46Great. So the next step is we just need
2:19:48to install Xcode. But the nice part is
2:19:51that we don't have to do anything
2:19:53because we'll just get OpenCloud to do
2:19:54the rest. So, I'm going to open up my
2:19:56Spotlight search again. Then I'm going
2:19:57to go to the app store and then I'm
2:20:00going to search and in search I'm going
2:20:01to do Xcode and then I'm going to
2:20:03download this. All right.
2:20:06Okay. So, the next step is let's go get
2:20:08our Google Gemini API key. And to do
2:20:11that, right here I'm going to go to
2:20:13AI.dev.
2:20:15And right here on the bottom
2:20:19right here, I'm going to hit get API
2:20:20key. And then I'm going to hit create
2:20:23API key. And then let's call it um open
2:20:28claw
2:20:30and hit create key.
2:20:33And then all we have to do you see we
2:20:35just created it open claw. And I'm going
2:20:37to copy this API key. And that's it. And
2:20:39I'm going to show what to do.
2:20:41What we do next is in Google what we're
2:20:44going to do is type in vision claw. And
2:20:47vision claw the first GitHub repo.
2:20:49You're going to click this. And all of
2:20:51this you just saw and what I did was
2:20:53made possible because of Shalu. He made
2:20:56this vision cloud repo and I just did a
2:20:59lot of development on top of it. But
2:21:00what we want to do first is now we want
2:21:03to come here, take this, copy this. And
2:21:06now everything we do is just talk to
2:21:08OpenClaw. So I'm going to paste this
2:21:09here. And then I'm going to just speak
2:21:11to it. All right. Hey, so can you help
2:21:13me set this vision claw repo up? I have
2:21:16Xcode downloaded on my computer. I want
2:21:19you to set that up and then also I am
2:21:22going to give you my Gemini API key. I'm
2:21:25going to paste it here. Can you do all
2:21:27the linking and everything for me? So
2:21:29after that, let me know what's next so I
2:21:32can just connect my iPhone and go. All
2:21:33right. So you want to do that and then
2:21:36now I'm going to paste my Gemini API key
2:21:38in and hit go. But obviously because I
2:21:40don't want to expose my API key, I'm
2:21:42just going to do this offscreen. All
2:21:43right. Great. So after that message, if
2:21:45you open up Xcode, what you should see
2:21:47is that it already built out a project
2:21:50for you. All right. So I'm going to open
2:21:52that up. And it has a bunch of code that
2:21:54I don't need to know what it is. But
2:21:56what I do need to do is now I need to
2:21:58take a USBC cable.
2:22:02So I'm plugging this in.
2:22:05And then I'm plugging this into my
2:22:07iPhone
2:22:12is now what you want to do is you want
2:22:14to go here. You want to hit this and
2:22:16then you're going to look for your
2:22:17iPhone. So I'm going to select Simon's
2:22:18iPhone.
2:22:20Okay. And then you're going to hit play.
2:22:23You might have to just like put in some
2:22:25permissions and whatnot. But what's
2:22:27going to be happening right now and then
2:22:29you'll see
2:22:40And then you hit play.
2:22:43Okay. And then I already have the app
2:22:45downloaded, but I'm still going to
2:22:46replace anyway, just so you guys can
2:22:48see. And what it's doing, it's
2:22:50downloading this app
2:22:53onto my phone. So, right here, if I hit
2:22:56this, and then you can see this app will
2:22:58be here, which means now I can start my
2:23:01iPhone. And this app is the same thing
2:23:03as we just experienced. So you can see
2:23:06hello.
2:23:09>> Hello. How can I help you?
2:23:11>> See, that's cool. Uh, and you can hear
2:23:14it happening from my glasses as well cuz
2:23:16I'm wearing it. And that's it. As simple
2:23:19as that.
2:23:21Okay. Now that we learned how to build
2:23:22so many different things, let me show
2:23:25you how we can actually take these
2:23:26builds and actually make money from it
2:23:28and implement it in business. And to
2:23:30start off, let me actually show you this
2:23:31clip.
2:23:32>> I've been through every single
2:23:33technology, you know, event and
2:23:36evolution, and this blows them all away.
2:23:39>> Now, how you implemented in business is
2:23:42a whole different issue. Like literally,
2:23:44when I was 24, I was walking into
2:23:47companies who had never seen a PC before
2:23:49in their lives and explaining to them
2:23:51the value and having these guys going,
2:23:53"Well, son, I got this receptionist
2:23:55right there. I got that secretary. I'm
2:23:57never going to need that ever."
2:23:59Right.
2:24:00>> And but then you know my business then
2:24:03was helping them figure out how to
2:24:05implement it to give them an advantage.
2:24:06Y
2:24:06>> there are going to be integrators that
2:24:09particularly young kids like when I'm
2:24:10telling my kids who are 15 18 21 or 19
2:24:1321 and kids going into school what
2:24:15should I do? What should I do? I'd be
2:24:17like
2:24:17>> AI research
2:24:19>> learn all you can about AI but learn
2:24:22more on how to implement them in
2:24:25companies. Right. Because to your point,
2:24:28companies don't understand how to
2:24:30implement all that right now to get a
2:24:32competitive advantage. You got the head
2:24:33of Microsoft saying software is dead
2:24:36because everything's going to be
2:24:37customized to your unique utilization,
2:24:39right? Or use usage.
2:24:42Who's going to do it for them?
2:24:43Particularly small to medium-siz
2:24:45businesses. There are 33 million
2:24:47companies in this country. 30 million of
2:24:49them are soloreneurs, right? Um single
2:24:52person enterprises. There are only, you
2:24:55know, there are millions of companies
2:24:57that have one, five, 10, 50, 100, 500
2:25:00people that aren't going to have AI
2:25:02budgets, aren't going to have AI
2:25:04experts. This is where kids getting
2:25:06hired coming out of college are really
2:25:08going to have a unique opportunity. If
2:25:11you're spending your senior year in
2:25:13college right now, your senior year in
2:25:15high school, even whatever it is, your
2:25:16excess time, and you're learning the
2:25:18difference between Sora and VO, and
2:25:20you're learning how to do all this
2:25:22video, you're learning how to customize
2:25:24a model so that then you can then walk
2:25:27into a company and say, I understand
2:25:29your business as a shoe company selling
2:25:31shoes at a retail store or selling and
2:25:33selling shoes online. Let me show you
2:25:36how to benefit you. That is every single
2:25:39job that's going to be available for
2:25:40kids coming out of school because every
2:25:44single company needs that. You see,
2:25:46that's Mark Cuban, the guy from Shark
2:25:48Tank. And you can see that the potential
2:25:51is obvious, but the path to getting
2:25:53there isn't clear for most people. That
2:25:55gap between what technology can do and
2:25:58what regular people can make it do is
2:26:00where the money really lives. Like for
2:26:02example, WordPress, Shopify, all these
2:26:05things are free. But people still pay
2:26:07thousands for someone else to set it up
2:26:09for them. Why? It's because they don't
2:26:11want to learn it themselves. They just
2:26:12want the result. The same thing is
2:26:15happening right now with AI agents and
2:26:16OpenClaw. And OpenClaw is literally open
2:26:18source. It's free. But 95% of the people
2:26:21who could benefit from an autonomous AI
2:26:23agent will never be the ones that set it
2:26:25up for themselves. They don't even know
2:26:27what a terminal is. They don't they
2:26:29don't know what an API key is. They have
2:26:31zero interest in learning.
2:26:34you are now in that 5% that can give
2:26:37them that result. So with that being
2:26:40said, what can you sell? Well, there are
2:26:42four easy business models which you can
2:26:44take and sell from what you've just
2:26:46learned. Each of them have different
2:26:48margins, different levels of effort and
2:26:50scalability. And model one starts with
2:26:53just done for you builds. So you'd build
2:26:55a custom agent or system for a specific
2:26:58client. So, you'd have to do the
2:27:00discovery call, scope out what they
2:27:02need, build and deploy it, and then hand
2:27:04it off to them. This is probably the
2:27:06highest price per engagement. I've seen
2:27:08a lot of people sell this for 2,000 to
2:27:1010,000 depending on the company. This is
2:27:12the most work per client, but it's the
2:27:15right model to start. But if you're just
2:27:18starting out, it's the right model to
2:27:19start because with every build, it's
2:27:20going to teach you something new about
2:27:22what client needs are. You're going to
2:27:24get paid to learn. And most of my
2:27:27community members actually get started
2:27:28here because you can just do this once
2:27:30and then every subsequent build gets
2:27:32much easier. Okay, model 2 is
2:27:35preconfigured packages. So instead of
2:27:37building it from scratch every time, you
2:27:40can create a template agent and set it
2:27:42up for a specific niche and sell it to
2:27:44multiple clients. So think the content
2:27:47creator agent package or the real estate
2:27:50agent package. the same core
2:27:52configurations but just a slight bit of
2:27:55personalization for each client's voice
2:27:57and their workflows. That would be kind
2:27:59of like a mid-tier. So around 500 to 3K
2:28:03because you're not really starting from
2:28:04zero each time and your operation cost
2:28:06is a lot lower where you build once and
2:28:09then deploy for others. The first build
2:28:11maybe takes a bit of more time but the
2:28:13second and the third gets a lot easier.
2:28:15Okay. Model 3 is productized services.
2:28:18So standard So standardized offer with
2:28:20fixed scope and a fixed price. So let's
2:28:22say for $1,500 a month, your agent
2:28:25monitors your community. It drafts
2:28:27content and handles your email triage.
2:28:29This is the easiest business to scale.
2:28:31But you really have to do research and
2:28:35you have to know the exact workflow that
2:28:37gives the value for the person that's
2:28:39buying from you. Okay. Model 4 is a SAS
2:28:42style product. So for example, this is
2:28:44actually a claw wrapper. It's built on
2:28:47top of platforms like OpenClaw, but for
2:28:50specific verticals. There's already
2:28:52hundreds of OpenClaw rappers in the
2:28:54market. You can see some of them right
2:28:56here. But this model requires more
2:28:58upfront investment from your time. It
2:29:00has the highest scalability, but it's
2:29:02also the most fragile and easiest to
2:29:04break because AI moves so fast. People
2:29:06start should start with either model one
2:29:08or two and work towards this later after
2:29:11they really understands the nuts and
2:29:13bolts. But after that, I want to really
2:29:15really emphasize is when you start out
2:29:18first, pick one niche. Like the biggest
2:29:20mistake is trying to sell to everyone.
2:29:21Like I build AI agents. That means
2:29:23nothing to most business owners. They
2:29:25don't even know what are. It's vague. It
2:29:27sounds expensive. But if you can say
2:29:29something like, "I'll build you an AI
2:29:30system for e-commerce stores that
2:29:32creates 60 plus ad creatives per week."
2:29:35That's an offer somebody will pay for.
2:29:37It's specific. It has a number in mind
2:29:40and it describes the result they
2:29:41actually care about. and niches that are
2:29:43working right now that I've been doing a
2:29:44lot is coaches, e-commerce stores, real
2:29:47estate agents, marketing agencies,
2:29:48content creators, and just local service
2:29:51businesses. The best agent business
2:29:54won't come from the best engineers.
2:29:56They'll come from people who understand
2:29:58their target customer daily workflow
2:30:00better than the customer understands it
2:30:02themselves. What that means is you don't
2:30:04need to be the most technical person.
2:30:06You just need to be the person who
2:30:08understands what a fitness coach
2:30:10struggles with at 6:00 a.m. and which
2:30:12parts can maybe be automated. Now that
2:30:14you understand the ways we can sell
2:30:16these, what are different pricing models
2:30:18we can leverage? Because different
2:30:20clients and different industries like
2:30:22different pricing. So the easiest one to
2:30:25start up with is a setup plus retainer.
2:30:28This is the most common and most
2:30:29profitable structure, which is basically
2:30:31just a onetime fee plus a monthly
2:30:33retainer. Some setup fees can range from
2:30:36500 to a,000 and multiworkflow systems
2:30:39when it's actually integrated and
2:30:41working day-to-day. You can take that up
2:30:43a notch each time you build something
2:30:44new. And retainers can range between 200
2:30:47to a,000 per month depending on the
2:30:49value you're actually getting for your
2:30:51client. And what the retainer is there
2:30:53and how you pitch it is because you're
2:30:55saying, "I'm going to be monitoring
2:30:57this. I'll be updating it as it gets
2:30:58better, adding new skills,
2:31:00troubleshooting, optimizing it." And the
2:31:02retainer is where really the money is.
2:31:03It's recovering revenue that compounds
2:31:05as you add clients. So for context, AI
2:31:08agencies across the boards are charging
2:31:10around 500 to 5,000 per month for
2:31:13ongoing management. And the easiest one
2:31:15if you're just starting out is value
2:31:17based pricing where you don't have to
2:31:19price based on your time. You just price
2:31:22based on what your clients get. So I'll
2:31:24give you some math. A content creator,
2:31:26let's say 15 hours per week on content
2:31:28production. their value and then they
2:31:31value let's say their time at $100 per
2:31:33hour. [snorts] That's $6,000 a month in
2:31:36production cost. So if you set up an
2:31:38agent that cuts their production time to
2:31:403 hours per week, that's $4,800 per
2:31:43month on savings for them. You don't
2:31:45have to charge that much, but you can
2:31:47charge let's say $1,000 at $500 a month
2:31:49for maintenance. It's a fraction of what
2:31:51they save and the ROI kind of sell saves
2:31:53itself. The framing is always here's
2:31:56what this costs you today and here's
2:31:57what this costs with the agent. That
2:31:59difference is basically yours. The
2:32:01monthly retainer deliverables is
2:32:03basically the monitoring and uptime
2:32:05checks, the troubleshooting when things
2:32:07break, adding new skills, optimizing
2:32:09workflows based on any output quality
2:32:11and any updates to OpenClaw or any
2:32:14connected tools when they have new
2:32:16versions. If you write this into your
2:32:18proposals and send it as a one-page
2:32:20document, the client can exactly see why
2:32:22they're paying you and what they get and
2:32:25makes it a lot easier to mull over the
2:32:27price. Now, if you want to find clients,
2:32:30strategy number one is just use the
2:32:32thing to sell the thing where this is a
2:32:34meta play and you actually we just built
2:32:37a content creation agent. You can make
2:32:40content about that, have a link to your
2:32:41calendar and it's very easy to get
2:32:43content and leads from there. You can
2:32:45also build record yourself building the
2:32:47morning briefing agent show an output
2:32:49for a specific industry posted on
2:32:51LinkedIn, Twitter, YouTube. And these
2:32:54screen recordings of your agent doing
2:32:55real work are the best marketing
2:32:57material that'll ever exist for your
2:32:59service business because you're not
2:33:01telling people it works. You're
2:33:02literally showing them what you're going
2:33:03to do for them. People may watch it, may
2:33:06say, "I need that." And they're your
2:33:07leads. They come to you. You don't pitch
2:33:09them and they see your product working.
2:33:12This is the most effective legion
2:33:13strategy I've seen in my community that
2:33:15works and the content demonstrates the
2:33:17product. So you don't need a pitch deck
2:33:19or anything. So strategy number two is
2:33:22cold outreach with proof. Now if you
2:33:24let's say you pick your niche, you can
2:33:25find 50 people in that niche and send
2:33:28them a personalized message. So maybe
2:33:29this is on LinkedIn using sales or just
2:33:32search for them. Like don't tell them
2:33:34something like, "Hey, I build AI agents
2:33:35that nobody cares about." Say something
2:33:37like, "Hey, I noticed you post content
2:33:39five times a week across LinkedIn and
2:33:40Twitter. I build a system that
2:33:41researches topics. It drafts posts in
2:33:44your voice and creates a carousel on all
2:33:47on a daily schedule. Here's a 30
2:33:49secondond video of it running. Want me
2:33:51to show you how this works? Very
2:33:52specific, relevant to their actual works
2:33:54and it includes proof and you can take
2:33:56the old builds we did and then you sell
2:33:58that. That's the recursive part. Your
2:34:00agent can actually help you with this
2:34:02too. Have your agent do the researching
2:34:05for the prospects. Personalize the
2:34:07outreach messages and track who
2:34:08responded. Use the tool to sell the
2:34:10tool. One approach that's working
2:34:12really, really well is when you pick a
2:34:14vertical and build one great demo for
2:34:16that vertical and then targeted outreach
2:34:18to 50 prospects in that space. Even a 5%
2:34:21response rate gets you conversations and
2:34:23some sales.
2:34:25Okay. Strategy three is you go find a
2:34:28niche community. So every industry has
2:34:30communities like Facebook groups, school
2:34:31groups, Discord groups, subreddits,
2:34:33Slack groups. If you want to join those
2:34:36target niches and you answer questions
2:34:38and share what you're building, if you
2:34:40actually generally help people and you
2:34:42don't have to pitch, you just
2:34:43demonstrate. And when someone says, "I
2:34:45spent 4 hours a day on managing my email
2:34:47or building this carousel," you could
2:34:49say like, "Hey, I built an agent that
2:34:51that handles that. Here's how it works."
2:34:53You post a screenshot or a video and you
2:34:55share the workflow and the client would
2:34:56naturally come to you because you've
2:34:58already shown them that you understand
2:34:59the workflow and you're part of their
2:35:01sector. But the biggest thing I want to
2:35:02make sure you understand is here's a
2:35:04specific play for your first client
2:35:06where you find someone whose work you
2:35:09actually understand really well. Offer
2:35:11to build them a prototype for free or at
2:35:13a steep discount. Tell them, "Hey, I
2:35:15want to build this for your workflow.
2:35:16Help me test it and give me feedback so
2:35:18I can make it better." Okay. After you
2:35:20have that done, they get a useful system
2:35:22at a low cost and you get a study, a
2:35:24testimonial, and deep knowledge of their
2:35:27workflow. Then what you can do is you
2:35:29can charge the next person full price
2:35:31and you ask your number one client the
2:35:33first one for a referral. So just to
2:35:35walk through a concrete example. So
2:35:37let's say your [snorts] target customer
2:35:38creates content regularly three or more
2:35:40times a week and spends 10 hours a week
2:35:43on content production. He has an
2:35:44established voice and style that they
2:35:46want to maintain and they use multiple
2:35:48platforms. So X, LinkedIn, maybe a
2:35:50newsletter, maybe YouTube. You're not
2:35:52selling to people who want to start who
2:35:54want to start creating content. You're
2:35:55selling to people who are already doing
2:35:57something and drowning in work or maybe
2:35:59want a bit more freedom. And the pitch
2:36:01you do is you go from spending 80% of
2:36:05your time on production to 80% of your
2:36:07time on ideas. Most people will actually
2:36:09learn about autonomous AI agents like
2:36:11Openlaw in 12 to 18 months. Now you have
2:36:14everything you need to learn it today
2:36:15and actually start making money from it.
2:36:18Most people will spend months trying to
2:36:19figure out the technology. You already
2:36:21built seven systems by now. And with
2:36:23that head start, that matters. Not
2:36:25because the knowledge is secret. It's
2:36:26all open source, but because the
2:36:28experience compounds. Every build you do
2:36:30makes the next one much much much
2:36:33faster. And every client teaches you
2:36:35something.
2:36:35>> All right, Andrew. Cool. So, I've been
2:36:37hearing you were selling some openclaw
2:36:39offers. Can you take us through, you
2:36:42know, what your offer was?
2:36:45>> Yeah. Um, so first thing, the first
2:36:47thing I did was because I didn't know if
2:36:49I knew what I was doing. Like, I knew
2:36:50what I was doing for me, but I didn't
2:36:51know what I was doing for other people.
2:36:53Um, and so I had a couple of friends
2:36:55that are business people that I was
2:36:57like, "Hey, look, this is what I've
2:36:58done." Um, they were able to
2:37:02>> what kind of business?
2:37:04>> Uh, so one was in commercial real estate
2:37:07um, and the other one is in roofing. And
2:37:11um, and they're in two, one's in
2:37:12Michigan, one's in Alabama. And so I was
2:37:16kind of showing them what I was like my
2:37:19call my stfont. and I was like showing
2:37:21them all the things that I could do. So
2:37:23anyway, they um I kind of used them as
2:37:26guinea pigs to kind of do it and I
2:37:27figured out what to how to
2:37:29systematically do it and um I use
2:37:32tactic. I don't know if people I highly
2:37:35recommend that everyone has something
2:37:36that can record the transcript because
2:37:38then I could take all of those different
2:37:40calls that I had with my friends how I
2:37:42was doing it and then I put it into
2:37:44Claude and I said, "Hey, take help me
2:37:46systemize this process." What was it
2:37:48that they got value out of like at the
2:37:50end with open cloud that
2:37:53>> they
2:37:54so essentially so they just have an AI
2:37:56assistant now right that's like that's
2:37:58that's connected to their their Gmail
2:38:01and things of that and so that they can
2:38:04um you know from their phone they can
2:38:06say hey you know they've built I've
2:38:08helped them build out daily briefs
2:38:10because I really like the fact that it
2:38:11can while I'm asleep or I'm away um it's
2:38:15kind of you know because a good
2:38:16executive assistant is like the front
2:38:18line or the first line of defense,
2:38:19right? When it comes to your schedule
2:38:21that it knows what's going on, what's
2:38:23coming up as well as um by the way,
2:38:27here's who has emailed you over the
2:38:29night or you know, you were in a meeting
2:38:30and this person or that person called
2:38:33gives you an urgency um and you start to
2:38:36um really own your time and own your
2:38:38schedule because I think a lot of people
2:38:40they um they base their their time and
2:38:44schedule off of what the most urgent
2:38:46thing is. And so there's not a lot of
2:38:48control on their part. Does that make
2:38:50sense?
2:38:50>> So it's kind of like an assistant,
2:38:51executive assistant, the morning daily
2:38:54briefs, some follow-ups, those kind of
2:38:55things, right?
2:38:56>> Yeah, that's the main thing we're using
2:38:57for right now.
2:38:59>> What did what were you able to like
2:39:00charge them for this?
2:39:02>> Uh so then um I just one paid me 500 and
2:39:05one paid me a,000, but now I've done
2:39:08three more and those have ranged between
2:39:112,000 and 5,000. Um, but just to be
2:39:15transparent, honestly, it's not like
2:39:17you're going to sit down and I I, you
2:39:19know, I keep trying to better systemize
2:39:22it so that it's like people can get um
2:39:26where they want to go because at the
2:39:27same time too, like now um obviously I
2:39:30can use Stfan to um research things
2:39:34while I'm away and um and to you know I
2:39:36give it task that it will remember to do
2:39:39um which is really really really
2:39:41helpful. What kind of
2:39:43>> what? Uh like if I'm gonna be talking
2:39:45with a certain company, um I ask for it
2:39:47to go and they dare scrape that company
2:39:50or the people that are going to be on
2:39:52the call and
2:39:53>> you're using the tool to sell the tool.
2:39:55>> What is it? Yeah. Yeah. Exactly. And I'm
2:39:57like this is what Yeah, exactly. That's
2:39:59a good that's a good that's a great
2:40:00point. I didn't really put that
2:40:01together, but yeah, you show them like
2:40:03this is what it is and this is how I do
2:40:04it. Um, and even but even people beyond,
2:40:08you know, buying the the service and
2:40:09whatnot, it's just such a great thing to
2:40:12have as far as really being prepared. I
2:40:14think that that's the one thing um is
2:40:16that I love that I now feel like I own
2:40:18my schedule more. Um, and that I feel
2:40:21more prepared in all of my meetings. Um,
2:40:24>> that's really cool.
2:40:26>> That's I think maybe a lot of people
2:40:27struggle with this part as well. So now
2:40:29let's say you learn how to build these
2:40:30kind of things, these cool assistants
2:40:33and there's a lot of use and the way you
2:40:35were getting these sales is just demoing
2:40:37it. How were you able to find these
2:40:40prospects or leads in the first place?
2:40:42>> I always say, listen, this is age-old
2:40:45thing of sales. Use your warm market.
2:40:47And the cool thing is is that there's
2:40:49always going to be people that are
2:40:51willing to um help you figure out what
2:40:55you know while you're you're figuring it
2:40:56out, right? I wouldn't recommend
2:40:58charging those people. I mean, unless
2:40:59you can
2:41:00>> I guess you should not charge, right?
2:41:02>> Yeah, I would I would highly recommend
2:41:04that just because, you know, you're
2:41:05going to it's not an if, but a win that
2:41:07you're going to hit some kind of a wall
2:41:09that you're going to have to figure out
2:41:10and you'll always be able to figure it
2:41:11out if you especially if you use the
2:41:12right, you know, AI AI tools. I use claw
2:41:15chat um and and I even created a project
2:41:19specifically for um this bit this kind
2:41:22of business so that all of my um
2:41:26interactions with all the people you
2:41:28know my friends and and the clients
2:41:30because then it of course remembers
2:41:31everything right and so then I have the
2:41:34ability when I'm working with someone
2:41:36new that if I'm like oh wait I know that
2:41:38we've run in I've run into this specific
2:41:41issue within that um project it will be
2:41:44able to pull that out and be like, "Oh,
2:41:46>> no pressure. You're helping people.
2:41:47You're getting onto conversations and
2:41:49then seeing where that goes." That's
2:41:51like a good
2:41:51>> strategy. Exactly. And then just
2:41:53constantly asking Claude um afterwards
2:41:56to update the system that it's created
2:41:58because I, you know, it's like it keeps
2:42:00kind of creating a playbook for me to to
2:42:02use um so that um I don't particularly
2:42:05send that playbook out to people because
2:42:07you're kind of giving your sauce away
2:42:09and I would recommend that. Um, but you
2:42:11can definitely do like a um a like a
2:42:15quick, you know, a lot of times when you
2:42:16buy a TV or something like that and it
2:42:18comes with a quick manual where it has
2:42:20like, you know, here's some tips and
2:42:21some quick like ways just, you know, to
2:42:24cut to this or to do that. Um, um, I
2:42:27asked Claude to create that for my new
2:42:30my new clients that I that I sent to
2:42:32them so that because a lot of times they
2:42:34forget just the smallest thing. How do I
2:42:36do that? How do I restart the gateway or
2:42:38things of those things?
2:42:39>> I got you. Yeah. So,
2:42:41>> so what's next? So, now you're selling
2:42:43these open cloud agents. I remember you
2:42:44were also selling voice agents before.
2:42:46You're kind of expanding your pool in
2:42:48that way, right?
2:42:49>> Uh, yeah. And so now we want it to be
2:42:51because you know you and I both know
2:42:52that where everything is leading or
2:42:54heading to is that we are you know you
2:42:58have AI assistants and then you're going
2:43:01to have AI colleagues and then
2:43:03essentially you're going to have just AI
2:43:06worker bees that report back in a very
2:43:10much you know you you know you've heard
2:43:12me use the metaphor of like I'm the
2:43:14conductor they're in the symphony
2:43:16they're in the orchestra and they all
2:43:18are playing their own individual parts.
2:43:20They're doing the thing and they're just
2:43:22reporting back to me. Um and and
2:43:25essentially I'm trying to take myself as
2:43:26much as I can out of the
2:43:27>> loop agent workforce. That's what you're
2:43:29trying to
2:43:30>> Exactly. Right. Because everyone has to
2:43:32understand that's where we're heading.
2:43:34You know, I'm sure you've you've heard
2:43:36this and and I'll just repeat it for
2:43:38those who haven't. Don't build for now,
2:43:42right? Because if you start to if you if
2:43:44you take what is now and you're trying
2:43:45to just build the AI for the for for
2:43:48now, um by the time you kind of get it
2:43:51really going, you're going to be behind.
2:43:53So build think like where are we going
2:43:55in what's going to look like in six
2:43:57months? Build for that. build for six
2:44:00months from now in that in that way
2:44:02because then you'll start to realize
2:44:04that that's what puts you ahead because
2:44:09you know what is considered ahead
2:44:10becomes right in the moment really quick
2:44:12now you know what I mean like very few
2:44:14people can say I'm ahead for very long
2:44:16if you don't believe me look at how you
2:44:19know Jim uh you know uh yeah Gemini will
2:44:21put something out or enthropic will put
2:44:22out a new model and all a sudden chat
2:44:24you know GPT54 comes out and all these
2:44:26updates to codec like no one really is
2:44:29at, you know, is quote ahead in the AI
2:44:31game, but the best thing to do is to is
2:44:34to really kind of look at where we're
2:44:35going and where we are going is the
2:44:37agentic workforce that we as humans work
2:44:40with.
2:44:42>> Awesome. That's really cool, Andrew.
2:44:43Thanks for sharing that. Um, you know,
2:44:45hopefully these guys are a bit more
2:44:47inspired. So, it's not just like there's
2:44:48so many people taking advantage of like
2:44:50open cloud, cloud code, all these just
2:44:52to make life a lot easier and also make
2:44:55some money out of it. And I love that.
2:44:56>> Yeah. I just want to end with this.
2:44:58There's it's that's so great. Anybody
2:45:00that if you really want to do it, first
2:45:01of all, you got to put the reps in. AI
2:45:03makes it easier and it makes it more
2:45:05exciting. But to think that you're going
2:45:06to be able to sit down and an hour later
2:45:08you're going to just have all this, you
2:45:09know, amass all this uh you know,
2:45:11incredible knowledge and and um use
2:45:13cases is is is wrong. Um, but also just
2:45:18learn where you are effective because I
2:45:20think a lot of times people kind of do
2:45:21it backwards where they try to sell the
2:45:24excitement of the AI and like how cool
2:45:26is this and how cool is this? Um, and
2:45:28then at the end of the day they don't
2:45:29make a sale and it's because you're
2:45:30selling cool but you're not selling
2:45:32outcome and that's all that people care
2:45:35about as far like if I want to see a
2:45:36cool trick that's cool but if I want if
2:45:38I'm a business I'm going to write you a
2:45:39check. Um, I I don't care how cool the
2:45:42trick is, right? show me the outcome.
2:45:45Show me that it's going to be better
2:45:47than what I could do by myself or
2:45:49somewhere else. Um, and you're going to
2:45:51get not only are you going to get people
2:45:53to write that check quicker, but you're
2:45:55also going to get um uh people to uh
2:45:58give you referrals and that's really
2:46:00big.
2:46:00>> That's awesome, Andrew. Thanks. Yeah,
2:46:02this is really good advice. You guys
2:46:04heard it here first.
2:46:07>> Yeah.
2:46:07>> So, what to do next, right? So what I
2:46:10want you to do is pick one build that
2:46:12matches what you already work on. Run it
2:46:14for a week. Get comfortable with it.
2:46:16Then pick one niche from this section.
2:46:18Maybe a type of person whose workflows
2:46:20you understand. Identify which of these
2:46:23seven builds may help these them out.
2:46:25Then build it for yourself. Use the tool
2:46:27to sell the tool. And then you can join
2:46:29the community. The links in the
2:46:31description. That's where you'll get
2:46:32help. You'll see what others are
2:46:34building and actually get access to new
2:46:37builds before anyone else. And the
2:46:39biggest thing I want to communicate is
2:46:40that the way people are working now is
2:46:42changing completely. Every company will
2:46:45need autonomous AI agents like this.
2:46:47Most of them have no idea how to even
2:46:49build them. And the people who can are
2:46:52going to be in huge demand for a long
2:46:54time. Now you know how to build these
2:46:56things, how to debug them, how to sell
2:46:58them. I want you to go get started. And
2:47:00if you need any help, I'll be in the
2:47:01community linked below.

About this transcript

This page contains the full transcript of OPENCLAW FULL COURSE 3 HOURS: Build & Sell (2026) by Samin Yasar, generated from the public captions YouTube serves with the video. The transcript has 31,723 words across 4,432 segments, with the original timestamps preserved so you can click any line to jump to that moment in the embedded player.

What you can do with it

Use the transcript to take notes, quote the speaker, build a study guide, generate a summary with ChatGPT or Claude via the YouTube Summary tool, or export it as a timed subtitle file with YouTube to SRT. You can also re-open it in the transcriber to translate the transcript into 100+ languages.

Free YouTube transcript tool

YouTube2Text is a free YouTube transcript generator — no signup, no daily limit. Paste any YouTube link and get the full transcript instantly, with timestamps, click-to-jump, translation to 100+ languages, AI prompts for ChatGPT, Claude, and Gemini, and exports to TXT, SRT, VTT, or Markdown.