Open Claw is the most powerful AI tool

ever made. And after using it non-stop

for the last 3 months, I can confidently

say no one else on planet Earth has used

it as much as I have. I've used it to

grow my businesses, to build brand new

businesses that just raised funding.

I've exploded my revenue with it. I've

exploded all my social channels with it

and created tons and tons of content.

I've used it to improve basically every

aspect of my business and personal life.

In this video, I'll go over every single

Open Claw lesson I've learned over that

time period. Whether you are brand new

to Open Claw or a stone-cold expert, I

promise you are going to learn something

amazing in this video. So, now let's

lock in and get into it. So, before we

jump in, feel free to check out the

chapters down below. If you've used Open

Claw in the past or if you've watched

past videos I've made, I'm probably

going to cover some things that you're

already familiar with. So, feel free to

go down below, check out different

chapters, jump around. I want to make

sure I cover new things for you here.

But, let's start off with why Open Claw

is the goat and what it is. Open Claw is

a completely autonomous AI agent. It is

basically your own personal AI employee

that works for you 24/7 and does

whatever you need. I'll show you what my

Open Claw does in this video, but it is

a completely autonomous employee that's

building things for me 24/7, researching

things for me, scraping things off the

web, getting actions done, posting

content, all of that. It's

self-improving. Every time I send it a

message, it improves, builds skills,

figures out new ways to do things. It is

tenacious and incredible. It does

whatever a human can do. So, it manages

all my devices for me. It builds apps

for me. It changes settings for me. If

you notice down below all the members of

the Fin Family, I just said, "Hey, build

me new logos and put it on YouTube for

me." It designs new logos, then went on

YouTube Studio and uploaded them for me.

It literally does whatever a human can

do. And here's the best part. Here's

where it really differentiates itself

from the Claude codes, Claude co-works,

codexes of the world. It is completely

open source, meaning your data stays on

your device. It doesn't go to the

servers of AI labs or anything like

that. And you can see how the data

flows, what it does, and it is

completely customizable. If you want to

download the source code of Open Claw

and change it and fork it and do

whatever you want with the code, you can

do that. It's completely open source and

up to you. If you want to plug in

different models, you can do that.

Claude code, Claude co-work, you can

only use Claude models. Codex, you can

only use chat GPT models. With Open

Claw, since it's open source, you can

use whatever models you want. If you

want to use local models that run on

your device completely for free, you can

do that, too. And I'll cover all of that

later in the video. We're covering

everything, use cases, models, devices,

all of that, all the lessons. This is

going to be comprehensive. It's also

private and secure. So, all your data

stays on your device. And they have made

huge improvements when it comes to

security. I will also cover security on

this video, as well. Here's my favorite

part about Open Claw is it lives where

you work. I have it in Telegram. I have

a bunch of Open Claws and Hermes agents

and other agents here on the left. Every

one of these you see that has a letter

on it is a different Open Claw or Hermes

agent. I'll also cover Open Claw versus

Hermes later, as well, cuz that's a big

controversy right now on Twitter. But,

it lives where I live. I can go on

Telegram anywhere in the world on my

devices, on my phone, say, "Hey, do this

for me." and it'll go on my computer and

build a whole bunch of things out. You

don't have to have 20 different apps to

use it. Just all goes in your Telegram

and it works really, really well. Now,

let's talk about where to host your Open

Claw, what device it needs to be on.

Does it need to be on a Mac mini like

this or can it just be on anything? Let

me just start off by saying this. Host

it literally anywhere except for a VPS.

I know if you go on to literally any

other AI YouTube channel right now,

basically every video is sponsored by

Hostinger. I will tell you this, none of

the creators that are sponsored by

Hostinger are actually putting their

open claw on Hostinger. Do not put your

open claws on VPS's. They are a

considerably worse experience. They are

less secure. They are less usable. They

are less powerful. There are zero zero

zero advantages to putting your open

claw on a VPS. You need to have this on

a local device. I will never take a

sponsorship from any of these VPS

companies. I can't in my good conscience

tell you to put your open claw on a VPS

when I am not personally doing it

myself. But does that mean you need to

run out and buy a $600 Mac Mini? No,

that doesn't. You can put this on

literally anything. This is like the

most misquoted thing I say. Everyone

goes, "Oh, Alex Finn's the Mac Mini

guy." You're right, I do like it on Mac

Minis, but you do not need to buy a Mac

Mini for this. You can go take that old

Lenovo computer with the red little nub

in the middle of it in the keyboard that

you were using 15 years ago in college.

Take that out of your closet, plug it in

for the first time, and put open claw on

that. You do not need to buy a new

device for this. You will still get

incredible power out of this. But here's

the thing, if you do want to upgrade and

you do want to put it on a Mac Mini,

it's going to be an amazing experience

for many different reasons. Mac Mini is

one of the most usable, best values in

computing, I think, on planet Earth. I

love having it on my desk knowing

there's an AI agent working on it 24/7.

It's fun to play with. It's fun to

customize. You can easily remote into it

from any other computer you have. Mac

Minis are great values and great

devices. That's why a lot of people are

going on buying them for these agents

cuz there's just the easiest way to have

a little device on your desk knowing an

agent's working 24/7. You can put open

Claw on your main driver computer, the

one you use every day. That's totally

fine. I do it myself, but I also have

other devices that I'm putting on as

well. So, it has its own workspace, its

own hard drive, its own files to use,

its own tools to use. You do get

advantages from giving it its own

computer, but you don't need to do that.

Now, if you want to take it up to the

next level like I do, I currently have

three Mac Studios, two Mac Minis, and a

DGX Spark. You don't need to do what I'm

doing. Don't run out and just buy all

those cuz I'm doing it. But, there are

advantages to doing that as well. You

get to run the most frontier local

models, which basically are, and I'll go

into deeper detail on this a little bit

later, models that run on your devices

that don't connect to the internet, that

are fully private and secure. It allows

you to unlock a lot more use cases. So,

it is more power. The more computers you

have, the more powerful computers have,

the more power you get. But, as long as

you just install this on I'm serious,

any device you have, any device at all,

you're going to be so far ahead of

everyone else in the world cuz you're

going to be using this powerful AI

technology locally in your office, on

your desk, and it's going to be amazing.

One last recommendation before we move

on to which models to be using, I would

scale into this slowly. A lot of people

that watch this go and spend $20,000 on

Mac Studios after watching my videos.

And listen, that's amazing. I'm proud of

you. I think you're making a good

decision. But, I think for a vast

majority of people out there, it's much

smarter to start small and then scale

up. Start small, put it on the dusty

laptop from your closet, start to get

some use out of it. And then, once

you're comfortable with the workflows,

then you can buy a Mac Mini. And then,

once you've utilized all the hardware

inside the Mac Mini and you're ready to

start hosting local models, then you can

buy a Mac Studio. And then, once you've

done that, you've maxed out the Mac

Studio, then go buy another one. Right?

Start small, scale up. I think that's a

probably a lot smarter for most people

and just going out and dropping tens of

thousands of dollars on computers. So,

let's talk about models. This is a big

one. This is where there's tons of

controversy. I'm going to give it to you

straight. I've been using Open Claw for

hundreds of hours now. I am extremely

confident about in the recommendations

I'm about to give you here.

Orchestrator, this is the brain of your

agent. This is what decides which task

to do and how to do it. You need to be

using Opus 4.7. Now, I know what you're

going to say, but didn't they ban Opus?

Didn't they ban Anthropic? Yes,

Anthropic banned the use of Open Claw

inside of its subscriptions, right? So,

you can't pay for the $20, the $200

subscription and use those subsidized

tokens with Open Claw. But, you can

still use it with the API. Now, yes,

it's going to be expensive using it with

the API. If you optimize this correctly,

it's not going to be that much more

expensive and I'll talk about

optimization a little bit later as well.

If you use this the right way, you won't

be paying much more than that $200 a

month. But, at the end of the day,

here's what it comes down to. When it

comes to orchestrating of AI agents,

when it comes to coming up with tasks

and then completing those tasks, no

model comes remotely close to Opus 4.7.

Not ChatGPT, none of them. And I'm not

getting paid by any of these companies.

Anthropic, Open AI, Google, none of them

have ever paid me in my life. This is

truly based on my experience. I've

tested them all. Claude is the only

model where if you give it a task, you

can be extremely confident it will

complete that task. I love ChatGPT. I

love using ChatGPT. I love to chat with

it. I love to code with it. But, as a

orchestrator for Open Claw or any other

AI agent out there, Hermes, it just

simply is not as reliable based on my

experiences. And I hope this changes.

Open AI has been very, very consumer

friendly when when comes to limits and

when it comes to using your subscription

for whatever you want. Anthropic has

changed limits a good amount, and they

don't want you using their subscription

with different tools. OpenAI has been

more consumer-friendly, so I am rooting

for OpenAI to improve ChatGPT with Open

Claude, but we have to face the facts,

which is Claude just beats ChatGPT for

Open Claude usage. And to give you a

quick example, a quick metaphor on why I

believe this, when giving Open Claude

powered by Claude a task, you can be

near 100% certain it will complete that

task no matter what. You can go up to

Claude with a katana sword and cut off

its leg halfway through a race, you can

be confident it will crawl its ass down

to the finish line and complete that

race. With ChatGPT, if it even stubs its

toe halfway through the race, it's just

going to roll over and give up. I just

don't have confidence when I give

ChatGPT a task in Open Claude that it's

actually going to complete that task.

This is as of the time time of filming.

That's why I do one of these videos

every single month is to update my

opinions and thoughts on which models to

use cuz it's constantly, constantly

changing. It's rumored that ChatGPT 5.5

is going to come out this week. And you

know what? It might be a thousand times

better for Open Claude than Claude when

it comes out. And if that's the case,

I'll make an updated video and I'll

switch all my Open Claude's over to

ChatGPT, but I'm being unbiased and I'm

just talking about at the moment, you

need to be using Opus. You need to be

paying for the API. And just in case 5.5

comes out and it's better for Open

Claude, make sure to subscribe and turn

on notifications down below because I

will make a video the moment that model

comes out letting you know if you need

to be switching over. So, we talked

about the orchestrator, let's talk about

coding. So, one big thing I'll talk

about later in this video is the muscles

and brains model, right? Opus 4.7 is

your brain, and then you're going to use

other models as the muscles to actually

execute the tasks. You want going

ChatGPT 5.4 as the executor for coding.

It's incredible at coding. You get tons

of limits. It doesn't cost a lot. It's

very, very good. So, you're going to

want Chat GPT 5.4 to do the coding that

Opus 47's going to give to it. And then

you have other tasks as well like

research and writing. This is basically

your catch-all like everything else.

Kimmy K 2.6 just came out. It's great.

But you can use pretty much any other

cheap model for research and writing. If

you already have a Google account most

people have Google accounts, you can

plug in your Google OAuth for this.

Gemini's plenty good for research and

writing. If you have the $200 a month

Chat GPT plan, the pro plan, you can

just use Chat GPT for this as well. You

have tons of usage with that plan. So,

you can just plug it in for basically

everything else. Still save tons of

money. Still get great usage. It's going

to be great. These are the models as of

the time of filming you got to be using

for everything. In a second I'm going to

go over use cases. This is the big

question. Oh, if Open Claw's so great,

what do you use it for? What have you

built? We're going to go over that in

one second. What I want to do first

though is talk about how to communicate

with Open Claw. What should you do it

in? Should you do it in Discord? You do

it in Telegram? What should you

communicate with? Well, we're going to

go over that now. Then we're going to go

over use cases. We're going to go over

advanced workflows. We're going to go

over setting up second and third agents.

We're going to go over Open Claw versus

Hermes. We're going to go over cost

savings with the brains and the muscles.

We're going to go over building your own

software, building your own mission

controls. And we're also going to go

over security. But before we do that,

let's just talk about Telegram and how

to communicate with your agent. By the

way, if you haven't installed it yet,

it's super simple. I'm not going to

spend a ton of time on installation. You

literally just go to openclaw.ai.

You take the one line from right down

here. So, you see that quick start? You

hit copy on that. This isn't

complicated. I promise you can do this

cuz we're about to do something really

scary. You go into your terminal. I

know, spooky, very spooky the terminal,

but I promise you can do this. You paste

in that line and you hit enter and

you're good to go. Open Claws installed.

That's it. That's all you need to do.

You don't need to pay for someone to

come over and install it for you. You

don't need to pay for a VPS. I promise

you, you can do this. You just copy and

paste a line into your terminal, hit

enter, and you're good to go. Once you

have that installed, you're going to be

able to choose where you communicate

with your agent from. I think it's not

even close. I think the best way to

communicate with your agent is Telegram.

Telegram has a tremendous amount of

built-in functionality that makes it

really, really nice to communicate with

your agents. Telegram's also completely

free, so make sure you get that

installed. But there's many reasons why

Telegram's the best. It's really easy to

set up new bots. You don't need advanced

API keys and all this. Like if you do

Discord, which Discord has its

advantages, it's just a lot longer of a

setup. Telegram, like you'll be set up

in 10 seconds and you'll be good to go.

But what I really like about Telegram is

you can set up chats with your agent

where you have different topics in it.

And having these different topics does

many different things. It allows you to

make sure context is nice and efficient.

So if I'm talking about content in one

channel or my community in another

channel, if I've coding academy linked

down below, or the app I'm working on,

Henry Intelligent Machines, each channel

has its own context. So I'm not flooding

the API with tons and tons of context

with every message. This saves and this

keeps my chats more organized. To set

this up, all you do is you start a new

group chat in Telegram and you add your

bot to that group chat. Once you add

your bot to the group chat, you can set

up multiple topics inside the group chat

that your bot will respond to and will

remember the context just for that

channel. Now, Telegram is changing the

rules and how group chats work with

agents all the time. Sometimes you need

to tag your agent first before it can

see the different channels, sometimes

you don't. After you've set up OpenClaw

in Telegram, do this group chat, set up

the topics. If it's not working right

off the rip, go to your main chat with

your agent and say, "Hey, the topics

aren't working. How can we get this set

up?" And it will go and set it up for

you and walk you through that really

simply. I'd walk you through it right

now, but they're literally every time I

set this up it's been it's being done

differently. But the gist of it is

you're setting up a group chat, you're

adding your agent and you're creating

different topics. If at any point during

this tutorial you get stuck or confused

on anything, just go to your agent and

say what you're stuck on and it will fix

it for you. That needs to start being

your gut reaction anytime you run into

challenges or get stuck. Okay, I'm just

going to ask my agent what to do. I'm

going to ask it how to fix this. It can

fix basically 99.9% of problems for you.

Discord is really strong as well. I have

Discord set up for myself, too. Discord

is really great for automated workflows.

So, I have my agent dropping in viral

content into Discord all the time. I

have it writing new scripts based on the

viral content. Use Discord for automated

workflows. Do not use Discord as kind of

your main chat like I do with Telegram

here. There's a lot of weaknesses with

Discord and the OpenClaw integration. In

Discord it's not sending all your

memories and all your agent rules with

every chat. It does that in Telegram.

You're not going to have quite as good

of an experience when it comes to

chatting in Discord, but it's really

good for automated workflows. I made a

video on building those in-depth

automated Discord workflows. You can

find that down below, as well. But for

main chatting, what you're going to be

spending 95% of your time on anyway,

Telegram's the way to go and having this

group chat set up will make it nice and

organized, as well as save you lots of

money on tokens. So, let's talk about

use cases. This is the big controversial

one. Whenever I talk about OpenClaw, the

first 500 replies I get every single

time is, "But what have you built with

it? But what are you doing with it? But

what is the use cases? I don't get it.

What are you doing?" Let's put this to

rest once and for all. I'm going to show

you three use cases I do with OpenClaw.

We'll go through that in a second. I

also have like five videos that have

like 20 use cases each. I have plenty of

content out there on use cases, and

we'll cover that. But here's the

important point I want to make before we

go into this. Trying to copy other

people's use cases and asking what other

people's use cases are is kind of

pointless. OpenClaw is your own personal

AI employee. You wouldn't go to a CEO of

another company who just hired a bunch

of people and go, "Well, what are your

use cases for your employees? What are

your employees doing? How have your

employees made you money?" That's such

like a weird question to ask a business

owner because employees do different

things with every business they work

for. It comes down to like what is the

purpose of the company they're working

for, who's their boss, what are they

having them do, what are like the goals

and metrics of that company. It's

different for every single company and

every single CEO. So, going to other

people and saying, "Hey, what are you

doing with your OpenClaw?" is just kind

of weird because what they're doing

probably is not relevant for you. You

want your own personal AI employees

doing things that are personal to you.

Doing things that improve your own

personal life, your own personal

workflows. One last thing I'll go into

before I start showing you use cases is

also this. Measuring your productivity

with OpenClaw just purely on dollars

generated is also a very weird way to

think about it. A big part of how

OpenClaw works and what makes it so

amazing is it can automate basically

every manual workflow you do on your

computer. And not that outcome of every

workflow is dollars generated. A lot of

the time, automating your own workflows

just saves you time, which frees you up

to do other things. So, saving time is a

much bigger metric here than just purely

dollars generated. If you're saving time

with OpenClaw, that's a big win as well.

So, let's talk about figuring out your

own use cases, your own personal use

cases, how to get the most out of

OpenClaw for yourself, which I think

will be a lot more useful than me going

over more my use cases, which I'll do

anyway in a second. But, let's start

with you. Let's make this useful for you

first. The first thing you want to do

when going into your OpenClaw is first

brain dump the hell out of everything

about yourself. So, if I were to go

here, I'd say, "I'm Alex Finn. I have a

big AI YouTube, Twitter, own a SaaS

called Creator Buddy, making 300,000

a year, just raised pre-seed for Henry

Intelligent Machines, write a

newsletter, blah blah blah blah." On and

on and on, all about myself, my

business, what I'm working on, its

metrics. You want to do the same exact

thing. You want to brain dump everything

about yourself here. This will

automatically get stored into memory,

into your OpenClaw memory, cuz the

OpenClaw memory is constantly improving.

Then, I would do this. This is a big one

as well. I would get a little notepad.

You can get a notebook, you can get a

notepad, whatever you want to do. I

would for a day take this around with

you and write down on this piece of

paper every single manual task you do on

your computer. Whether it's writing

content, whether it's writing emails,

whether it's managing calendar, whether

it's talking to people on Slack,

whatever it is you do manually on a

computer all day, write it down on this

piece of paper like I have here. Then,

what you're going to do at the end of

the day is you're going to go to your

OpenClaw, and you're going to write down

every manual task you did on your

computer. This is key. Don't don't short

Don't shortcut this. Do what I'm telling

you to do here. Write down everything

you do manually, put it into your open

claw, and now your open claw will have

incredible context. It'll have

incredible context on who you are, as

well as what you do. Then we're going to

do this. Once your open claw's memory is

up to date on everything you are and

what you do, we're going to say this

right here. Based on everything you know

about me, my goals, my jobs and tasks,

and what I do on a daily basis, what are

some workflows and use cases we can

implement? Boom. This is amazing because

now your open claw's going to look at

everything it knows about you. It's

going to eliminate the need for you to

creatively think about all the workflows

and use cases you need to implement, and

it's just going to figure that out for

you. It's going to figure out on its own

the best way it's going to give you

value. The best way it's going to save

you money. The best way it might even

make you a few dollars. It's going to

figure that out for you. It's going to

give you a list of use cases you can

implement. Based on that list, just

choose one or two, get them implemented,

have your open claw working, and it's

going to give you a ton of value. Doing

that exercise right there that I just

showed you, I promise will be

significantly more valuable than me

sitting here and giving you all the use

cases I personally do on Although I will

do that in a second, what I just showed

you is going to be way more valuable cuz

it's going to come up with use cases

that are personal to you. The average

internet troll after I give the use

cases I do, goes, "Oh, well, that's not

relevant for me, so open claw's

useless." Obviously it's not relevant

for you because it's my own personal AI

agent. So you need to figure out what's

personal to you with your own personal

AI agent. So you do that exercise,

you'll be good to go. So let's talk

about, just for example, maybe just for

some inspiration, some of the things I

do with my agent. So this is Linear.

This is a project management tool. Right

now, I am building out my new startup

Henry Intelligent Machines. I also have

my old startup Creator Buddy I have as

well. And I have my Open Claw fully

integrated inside of Linear, right? So,

you have one of my Open Claws here,

Lola. I work with like four or five Open

Claws. I can go in and any software task

I have in Linear, for those who are not

familiar with Linear, it's a project

management tool, allows you to kind of

plan out and build your software if

you're vibe coding. So, I have all

related to Henry Intelligent Machines

inside here. I can take any task and

assign it to Lola. And automatically,

once I assign it to Lola, Lola goes and

completes that task for me. So, if I

have it writing code, it will go, take

that task, pull down the code from

GitHub, and then complete that task for

me and write the code and create a PR.

It'll put it in its own branch, so it's

nice and secure. And then I can go in

and I can see here, there's the uh new

branch it created. I can click on that

and see the work it did. This helps me

automate my vibe coding so, so much.

I've gotten so much more done building

out Henry Intelligent Machines because

of Lola and integrating my Open Claw

into Linear. You can do this same thing

as well. If you're vibe coding or

anything like that, I'd highly recommend

downloading Linear. It's completely

free. Set up a bunch of tasks and

projects inside Linear to track your

vibe coding. If you want to do it

automatically, you can even ask your

Open Claw, "Hey, just download Linear,

go in and set up projects and tasks for

the project we're working on." It will

do it for you. And then you'll say,

"Hey, I want you inside Linear. Let's

build out this integration so I can

assign tasks to you and you can vibe

code for me in there." And it will set

that all up for you. This has been one

of the best use cases I ever implemented

with Open Claw. If you want a full deep

dive on this, by the way, Open Claw and

Linear and vibe coding and all that, let

me know down below. I'll do like a

20-minute deep dive video on this. This

is probably the most powerful use case

I've done so far. So, let me know down

below or any other use cases you'd want

me to dive into. Here's another amazing

use case I use all the time, and that is

for content scouting, right? Maybe

you're not a content creator like me,

and that's totally fine. But, maybe you

still want to stay on top of trends.

This is in a fantastic way to do that.

So, first thing I do is I have Henry

going and actually going out and reading

all the posts on Twitter and YouTube

about Claude code, about Open Claw,

every couple hours or so. Then, when

content starts trending, it alerts me

here in my alerts channel in Discord.

So, just drops it in. And now I have a

really nice record of all the content

that's trending about Open Claw, about

Claude code, and all that. But, here's

where it goes to a new level. It will go

and it will actually draft scripts and

draft content based on what's trending.

So, based on what's going on out there,

what people are talking about, it will

create new scripts for me that I can use

and then create content on those

trending topics. If you're content

creator, you know if you want to go

viral, you need to be creating content

on trending topics. This is how it's

done. It's creating scripts for me all

day autonomously. This is one of the

really good use cases for Discord. These

automated workflows where I'm not really

chatting with my Open Claw in here. I'm

just getting content dropped into

different channels and having it

organized for me so I can stay on top of

information. Another huge reason it is

for stock research. So, I'm a big

investor. I like to invest in different

AI stocks. I'm trying to invest in the

AI build-out as much as I possibly can.

I have my Open Claw going and looking

for different investing opportunities

when it comes to AI all the time. It's

looking for different bottlenecks and

scarcities in the AI build-out supply

chain and surfacing those bottlenecks to

me every single day. So, every morning I

wake up to a brand new stock research

report with where all the scarcities are

at the moment. You can see here, right

now, the power grid and electrical

infrastructure is the big bottleneck.

And so, I'm on top of all the investing

opportunities out there as well. I'd

highly, highly, highly recommend, in a

world where the dollar is being printed

endlessly, to get your money into

assets, not financial advice, but at the

same time kind of financial advice, and

using Open Law as a researcher for

companies and investing opportunities

has been massive for me. I'd set that up

as well. If you want to set up any of

these workflows, feel free to do one or

two things. One, just tell your Open Law

what you want to do, and it'll set it up

for you. But, two, and maybe even

better, is click share down below, grab

the link to this video, and hand it to

your Open Law, and say, "Hey, find the

use cases in here and implement them for

me." It will go find the transcript of

the video, and actually build out these

use cases for you. Those are three of

the most powerful use cases for me, but

again, you can copy those. You're more

than welcome to copy those use cases.

I'm telling you though, the most

powerful thing you can do is figure out

use cases that are relevant for

yourself, that are personal for

yourself. It'll help you get way more

out of Open Law. This is your own

personal AI employee, so it needs to be

doing personal things for you.

Implementing my use cases may not do

much for you if none of this is relevant

for you. So, find your own use cases,

reverse prompt, reverse prompt, reverse

prompt. One other thing I love to do,

I'll just leave this with you here in

the use case section. Next, we're going

to talk about a multi-agent approach.

We'll talk about Hermes in a second as

well. But, one last thing I'll leave you

off with is my favorite One of my

favorite things to do with Open Law is

have it build out prototypes on the go.

So, if I'm on the go, I'm out getting a

Big Mac at McDonald's, I'm out at the

gym getting a huge pump on, right? I

will have my phone on me, and whenever I

get new ideas, and most ideas I come up

with either at the gym or when I'm

eating a Big Mac, I'll go on Telegram

and I'll say, "Hey, can you build out a

prototype for this idea for me?" And

I'll just spam it with ideas all day

long. And then when I get back to my

computer, Open Claw will have built out

all those prototypes for me and I can

then just go and test them very easily.

So, using Open Claw as a prototype

builder when you're out on the go,

fantastic way to use it as well. And

real quick, before we talk about the

multi-agent approach, so talking about

having agents, multiple Open Claws,

having Hermes set up as well. This is a

question I get all the time, when to use

Claude Code, when to use Open Claw and

Hermes. Claude Code is great for

building out consumer applications, so

big consumer-facing applications, very

complex, a lot to it, you want to be

using Claude Code because Claude Code is

built specifically for vibe coding. I

also like to use Claude Code just for

quick tweaks, changing button colors,

things like that, just cuz it's easier

to go into and get that changed. Open

Claw Hermes is great for being an

advisor to building those apps out,

right? You want to have an agent for

building and an agent for advising. Open

Claw, great for advising, mostly because

it knows the most about you. It has huge

memory about you and everything you do.

So, it's going to be really good at

advising on those coding projects. And

it's also great at managing files on

your computer. So, you're changing

different things on your computer, can

manage those files as well. So, from a

coding perspective, I use Claude Code as

the main driver for building consumer

apps and then I use Open Claw as an

advisor or as an integration in Linear

as I showed you before where I'll be

working with Claude Code to build things

and then I'll assign tasks to my Open

Claw in Linear to build out other

features as well. This is the split and

there's way more to it as well. I'm just

going over this quick. I mean, Claude

Code, you're only using Anthropic

models. At the time of this filming,

yes, Anthropic models are the best when

it comes to coding, but if you want to

use other models, if you want to use

local models, which we'll talk about

soon, the only way to do that will be

Open Claw because it's open source and

customizable. So, that's the split

between these two. So, let's talk about

the multi-agent approach real quick.

Everyone and their mothers is talking

about Hermes right now on Twitter.

Hermes is another agent just like Open

Claw. This is Hermes agent website right

here. It's fantastic. It's the same

exact setup with just the one line that

you paste into your terminal. It's

great. I find it to be actually slightly

more performant and less bloated than

Open Claw. It moves a bit faster. It

uses less tokens. That's its advantage.

As a disadvantage, I found that the

memory isn't quite as good in Open Claw.

And this is based on my personal

experience. Whenever I say this, I get a

tremendous amount of people angry at me

because for some reason there's this

weird tribalism in AI. But, based on my

own personal experiences, Hermes agent

still great. It's just the memory

doesn't quite match Open Claw. It seems

to just forget things a lot for me. But,

it is more performant. So, I will say

this, I don't think you can go wrong

with either. I think they're a lot

closer in performance than most people

on the internet making out to be. For

some reason on the internet, most people

are like, "Oh, no, Hermes is the

absolute best. Never use Open Claw." Or,

"Open Claw is the best. Never use

Hermes." I think that's mostly just

tribalism. I I feel like people use each

of these agents as like a a basketball

team for some reason where it's like,

"No, my basketball team's the best. Your

basketball team stinks." I don't know

why. You'll be a lot more productive if

you don't have this kind of tribalism

mentality where one thing has to be the

best and one thing has to be the worst.

I think they're both great. I think they

both have their own advantages and

disadvantages. I think using them

together is the best possible way to be

using them cuz then you get the best of

both worlds and you see which one fits

into your workflows. One might be better

for your own personal workflows than the

other. And that's totally fine. I'm not

paid by Open Claw. I'm not paid by

Hermes. I'm not paid by any of these

companies cuz I don't really take

sponsorships from anyone cuz I want to

give you unbiased opinions. So, here

would be my recommendation. Use them

both. Use them side by side. How do you

use them side by side is the question.

Well, let me show you. Inside of

Telegram, I have two things going on

here. I have my Henry group chat, which

you see as this blue square right there,

and then I have my Hermes regular chat,

and I can speak to Hermes right through

here. I have them working side by side

on this computer. You put them on

separate computers if you want. The only

real advantage to that is they'll have

separate workspaces, but if you have a

good enough computer, it doesn't really

matter. But, I use them side by side. If

OpenClaw's working on one thing, I'll

have Hermes go and work on something

else completely different. You don't

need to give them kind of separate

roles. Some people say, "Oh, Hermes is

their coder, OpenClaw's their

researcher, or whatever." You don't need

to do that. They're both more than

capable of doing whatever you want. I

kind of use OpenClaw's more of an

orchestrator, where if I'm doing

something really complex, I'll say,

"Hey, talk to Hermes and have them do

this thing as well." That's totally

fine, but the biggest advantage to

having a multi-agent approach and why I

think you should do this and why I think

you should set up one OpenClaw and one

Hermes is just for insurance. These apps

break quite a bit, and most of all, they

break when you update them quite a bit

as well. The best way to fix them is by

having a second AI agent there as

insurance that can go and then fix the

other agent for you. They can go and see

what you did wrong, see what the update

broke, and fix it for you, and update

the config, and all that. This has kept

my uptime up a lot higher than it was

before. Right before when things would

break, it'd take me a couple hours to

fix it cuz I didn't have an AI there to

help me out. But, when you have two AI

agents going, a Hermes and an OpenClaw,

when one breaks, you can have the other

go and fix it. And it's not like it's

more expensive. These are completely

free tools. You're just paying for the

tokens, right? And so, if you just plug

this in, even if you just plug Hermes

into like your ChatGPT subscription, or

even a local model, which I'll talk to

you about in the next chapter of this

video, you're not spending much more in

order to have a lot more insurance, in

order to get a lot more productivity.

So, the number one reason to set up two

agents is so that they're insurance for

each other to make sure they don't go

down. The number two reason to see which

agent fits better in your workflow, Open

Claw or Hermes. I truly don't have a

horse in the race. I use them both. I

think they're both great. I don't think

one's like way worse or better than the

other. So, I just use them side by side.

Get them both set up though, and you'll

make sure that your agents are always

up. You can say, "Hey, Open Claw just

went down. Please fix it, Hermes." or

vice versa, and they'll keep them up and

going. So, let's talk about cost

savings. The best way for cost savings,

this is not only about cost savings,

this will just get you better

performance overall, is the brain and

muscle approach. What's the brain and

muscle approach? The brain is when you

have an orchestrator, right? That you're

That's the model you're interfacing

with. That's the model that's

determining the task to work on. And

then that brain gives tasks it

determines to do to each of the muscles

it uses, right? So, this could be, you

know, a coding muscle, a research

muscle, a scraping muscle, things like

that. Choosing the appropriate model for

each one of the brains and muscles is

how you A get the most performance, and

B do the most cost savings. Right? Cuz

if you just use Opus for everything,

you're going to spend a tremendous

amount of money. So, you want to have

the right model doing each muscle. So,

let's talk about what's the best for

each. We talked about it earlier. When

it comes to the brain, the orchestrator,

nothing comes close to Opus at the

moment, at the recording of this video.

Hopefully in the next week, ChatGPT 5.5

comes out, and I would be very happy if

it's the best model out there for Open

Claw. I'd be very happy. I'm rooting for

it. But at the moment, we have to face

facts, we have to face reality. Opus is

the best. But, then you have different

muscles, and we talked about this

earlier. I'm not going to reiterate

every single thing here. But, ChatGPT

for the coding muscle is the best.

Writing and research, any sort of cheap

model like Gemini will be good here.

Gemini Flash 3 Flash is really good.

But, let's talk about local models here

cuz local models make fantastic muscles

as well. What is a local AI model? A

local AI model is an AI model that runs

purely on your device. At the moment,

when you use AI, when you talk to Opus

or ChatGPT, it's sending your prompts to

the cloud. It's sending your prompts to

servers in the middle of a big farm or a

soon in space that processes your

prompt, figures out the answer, and

sends it back to you. This means you

require an internet connection. This

means all your chats go to servers in AI

labs that they can read it and train

their models on. That means you're

paying for every single one of these

prompts. When you do local models that

run on your computers, you don't need

the internet. It's completely free, just

the cost of your power, whatever it is

plugged into the wall, and it's

completely secure. Nothing goes to the

cloud, nothing goes to servers where AI

employees can go and read it. This

brings a lot of advantages too, as in

you're not Again, you're not paying for

these calls. So, if you use a local

model as like your research muscle,

which is what I do right now, I'm using

Qwen 3.6 as well as GLM 5.1

as my research model, where it's going

on the internet and researching content

and things for me all day. Literally

every 20 minutes one of my local models

goes on the internet and scrapes a

website looking for content and

opportunities. This, again, is

completely free. It runs completely

locally, and it unlocks new use cases

cuz I can run it 24/7 365. If you ran

Opus 24/7 365, you'd be paying a million

dollars a month. But, because I use

local models, I can have it running

every 20 minutes, going online,

scraping, scraping, scraping. So, tying

in local models here as your muscles

gives you really, really big advantages.

Now, what hardware do you need for local

models? There's good news and bad news.

Good news is you can use basically any

device on planet Earth right now and run

a local model on it. Now, if you have a

base model Mac Mini, it's not that

powerful, you're not going to run the

most powerful local models on planet

Earth. You could run like a Gemma 4, a

small Gemma 4 that can handle your

embedding, which is basically your

memory management, and there's

advantages to that. If you have a Mac

Studio, a super powerful one, you can

run bigger models like GLM 5.1, which is

probably the best local model on planet

Earth right now, that can do big things

for you. It can code, it can research,

it can do a lot of different things. So,

depending on the hardware you have, you

can run a local model, but you should

still be paying attention to it. what

device you have, you should figure out

which local models run on it, and then

find the best use cases for those local

models. Best way to figure out which

local model you should be using, go to

your OpenClaw, say "Hey, here's the

device I have, here's the memory, which

local models can I run? And of those

local models, what are the best

workflows and use cases for them based

on what you know about me?" You do that,

you'll get a personalized custom

recommendation for which local models

you can use for which use cases and

which muscles. You'll save a ton of

money, and you'll be learning about

cutting-edge AI stuff like local models,

which is amazing. If you're running a

local model, you're ahead of 99% of

people, which is incredible. By the way,

if you learned anything so far, I run

live OpenClaw bootcamps every single

Friday in the Vibe Coding Academy. It's

the number one AI community on planet

Earth, over 1,500 people in there

learning about OpenClaw all day, asking

me questions, talking to me live all

day. Link down below. I promise it'll be

the best decision you ever make. Check

that out down below. Also, subscribe and

turn on notifications if you haven't

already. All I do is make amazing videos

about AI. You learned anything so far,

leave a like. It's the most free way to

support the channel. I appreciate you.

So, let's talk next about mission

control. Mission control is probably one

of the most important concepts when it

comes to open claw. This is a concept I

invented a few months ago that now

everyone's talking about. And this is

basically your own custom software

building platform. Any tools you need

for your open claw, you can build out in

your mission control. I have a lot going

on here. I have a Kanban board that has

all our different tasks in it between me

and my open claw. I have an agent

overview section where I can see all my

agents working and seeing what they're

doing. I have an office view so I can

literally watch my agents work in real

time as these cool 2D pixelated

characters. I even have a factory where

I can watch them vibe code for me in

real time, which is amazing. Here's the

point though. Mission control is your

place to build out any custom software

you need to improve your open claw

experience. These should all be custom

tools built for your own workflows,

whether it's organization and project

management, whether it's just fun things

like this where you get to watch your

open claws work, whether it's a custom

memory solution. These are all custom to

you. The number one question I get when

I talk about mission controls oh, open

source your mission control. Open source

it. Give it to us. That's really I'm

going to be honest and this is no

offense to anyone here. I think that's

really stupid question because this is

all custom to me. This is all custom to

my workflows. This is like me showing

you like a really cool pair of Jordan 1s

I just bought in my size and you go, oh,

give me those. Those are mine. Give them

to me. Well, they're in my size, they

have my feet stank in them, they're

built just for me. I don't know why you

want them. You can go out and get your

own in your own size, do whatever you

want in the own colors you like. I don't

know why you'd want mine. You should be

building your own mission control for

your own use cases that are custom to

you. And you need to be doing this. Even

if you've never coded before, you need

to be building a mission control. Cuz

here's how I think about it. Whenever

I'm doing something with Henry, my

OpenClaw, and it doesn't have the tools

it needs to do that task, I just go,

"Hey, build that out in our mission

control. Add that as a new tool." And

it'll build it out, and it'll be able to

use it moving forward. Like when it

wasn't remembering things for me, I had

to build out its own memory system and

put it in the mission control. When it

was creating docs for me, and it was

just being placed in random places on my

computer, and I couldn't find them, I

said, "Hey, make a doc section so I can

read all those docs." And anytime you

need a tool, you just go to your

OpenClaw and you say, "Hey, build the

tool out for me, put it in our mission

control, and start using it moving

forward." I'm thinking about

open-sourcing mine. If I do it, I'm

going to put it in the Vibe Coding

Academy. I think it's better for people

that I don't do it, to be quite honest,

only because you really should be making

this custom for your own use cases.

Copying other people, whether it's use

cases or mission controls, it doesn't do

you any good. First of all, half the fun

is building these things out by

yourself. The other half is these need

to be custom for what you're doing. So,

mission control, build it out. Great

reverse prompting opportunity. Hey,

OpenClaw, based on what you know about

me, what tools can we add to our mission

control? It'll give you a list of them

to say, "Hey, build them out." And

you'll be good to go. So, let's talk

about security. This is a big one. This

is one I get most questions on is

security. And for good reasons, right?

You have an AI agent living on your

computer that can do anything, that has

access to all your nudes in your

iPhotos, and everything else you do. You

need to have good security. But here's

the thing, though. There is no good

blanket security recommendation I can

give to you, right? It's an AI employee

that can do anything on your computer.

If I sat here and gave you every

security recommendation, we'd be here

for 15 hours plus. So, here's what you

need to do. If you can master this,

you'll be good on security. You need to

use good judgment. What do I mean by

that? Your OpenClaw isn't going to do

things that you don't tell it to do.

What I mean by that is you're not going

to say, "Hey, write me a PowerPoint

presentation on local models." And then

it goes and leaks all your iMessages to

Twitter. It's not going to do that.

That's not how it works. It only does

exactly what you tell it to do. So, that

means you need to have some personal

accountability, and you need to use good

judgment. You can't say, "Go and check

out all my emails, and then delete any

emails you think are spam." Cuz if you

do that, you're opening yourself up for

OpenClaw to do things that you don't

want it to do. You need to use good

judgment. You need to use precise

languaging on what you ask it to do, and

you need to think about the

repercussions of it doing what you ask

it to do. My question is this, because

security makes a lot of people really

nervous when it comes to OpenClaw. How

many people do you know has had

catastrophic security events because of

OpenClaw? My guess would be for 99.9999%

of you, you don't know a single person

who's had a catastrophic security event.

That doesn't mean there aren't

vulnerabilities. That just means there's

a lot of panic over nothing here. If you

just use good judgment, think about

what's going to happen with every prompt

you give it, I'm very confident bad

things won't happen to you. So, just

have some personal accountability, use

good judgment with your prompts, think

about what could happen, and you'll be

good to go. Again, you don't say, "Hey,

write me a tweet." And then it goes and

it downloads 20 viruses onto your

computer. That's not how this works.

OpenClaw does what you say. So, if you

want to have good security, think about

what you're asking it to do, and make

sure you're only asking it to do pretty

safe things. Those are all my lessons

learned after 300 plus hours of using

OpenClaw. The game is constantly

changing. AI models are constantly

changing. OpenClaw is constantly

changing. These will update as we go all

the time. I mean, tomorrow a new model

can drop that will change everything I

just said. So, make sure you're

subscribed and notifications on. So, as

things change, as things go, you get the

update straight from me. Leave a like if

you learned anything at all. Make sure

to check out the Vibe Coding Academy,

number one community in AI. Link for

that down below. Hope you learned

something. I'm so, so grateful if you

made it past this entire video that you

would sit here listening to me all day.

I still can't believe how much this

channel's blown up over the last few

months. So, I am so, so, so, so grateful

for you watching and listening and

enjoying the content. I will see you in

the next video.