HomeVideos

Everything You Need to Know About Grok Bot (Full Grok Build)

Now Playing

Everything You Need to Know About Grok Bot (Full Grok Build)

Transcript

574 segments

0:00

I'm paying $300 a month just so I could

0:02

use Grok bots new viral app. This is

0:04

through XAI and now I'm on the super

0:06

heavy Grok plan. I'm a big agent user.

0:08

I've been using Hermes for a long time.

0:10

Before that I was using open claw right

0:12

when it came out. My AI agent runs my

0:14

entire business. It posts to all of my

0:15

social accounts, runs all of my ads and

0:17

is my executive assistant. So I wanted

0:19

to see how Grok bot stacked up against

0:21

Hermes. So I signed up for the $300 per

0:24

month super heavy Grok so I could use

0:26

this new feature. So today I want to

0:28

bring to you exactly what my experience

0:30

was with Grok bot and whether it's worth

0:31

it for you to set this up and how to

0:33

save a bunch of money so you don't end

0:34

up wasting money like I did. Now what I

0:37

did pretty much as soon as I signed up

0:38

for Grok bot is started playing with it

0:40

but mostly I wanted to port all of my

0:42

existing bot configurations into Grok

0:45

bot. So I'll show you how that really

0:47

worked as well. And if you stick around,

0:48

there's actually a surprise lesson that

0:51

I learned through all of this and I

0:52

actually ended up changing my entire

0:54

Claude code workflow based off of what I

0:56

learned from signing up for Grok bot and

0:58

it has nothing to do with Grok bot. This

1:00

is probably the biggest finding from

1:01

this video that you're going to get. So

1:03

stick around till the end so I can share

1:04

that with you. Now some quick context

1:06

just so you know why my answer means

1:07

anything. I run several eight-figure

1:09

marketing agencies. They're Inc. 5000

1:11

companies. I've [music] got thousands of

1:12

people inside of my AI automation

1:14

insiders community. So I teach this. I

1:16

travel the country speaking about AI and

1:18

I'm actually using AI to build and run

1:21

my [music] business, not just creating

1:22

content about it. All right. Now Grok

1:23

bot came out August 11th and it's gone

1:26

ultra viral. All kinds of creators have

1:27

been doing crazy videos about it being a

1:30

complete game changer. [music]

1:32

And I found some of that to be true,

1:33

some of it to be not true. So let's

1:35

actually talk about Grok bot and what it

1:37

is and then we'll go into actual

1:38

implementation and lessons and findings.

1:40

So you can sign up for Grok bot on

1:42

x.ai/bot

1:43

and you'll be able to get some

1:44

information about Grok bot. Now the

1:46

standout information here, what makes it

1:48

really unique, is it's not free. You

1:49

need to be on one of either Grok's heavy

1:52

paid plans or Cursor's heavy paid plans.

1:55

If you'll recall, XAI actually bought

1:57

Cursor for like $60 billion pretty

1:59

recently and they acquired all of their

2:01

coding models, too. So, their coding

2:03

models have gotten a lot better. That's

2:04

very important for Grok Bot because Grok

2:07

Bot runs on these new advanced coding

2:10

models. [music] So, here's the site for

2:11

Grok Bot. As you can see, this is what

2:13

it looks like. Here's my actual Grok

2:15

Bot, so you can see it in practice. I'll

2:16

be walking through that in just a

2:18

minute. It is a app that sits on your

2:20

computer. So, I actually have a little

2:21

Grok Bot icon down here and when I open

2:23

it up, it's just a desktop app. However,

2:25

unlike most agents that you've used in

2:26

the past where you either need to spin

2:28

up a VPS, a virtual server, or run it on

2:31

a Mac Mini, with Grok Bot, each one gets

2:33

its own dedicated virtual server

2:35

immediately. That's why they have to

2:36

charge money for it cuz there's server

2:37

costs. And here's what they advertise.

2:39

It works where you work, so it actually

2:41

gets its own browser. It can sign in to

2:42

things and take action. You can show

2:44

Grok Bot how to do things just like how

2:46

you train a Claude skill. You can

2:48

actually click around and teach it what

2:49

to do. Connecting is really easy and

2:51

apparently they get smarter over time. I

2:53

have not seen that yet, but I'm sure

2:54

it's true. Now, Grok Bot's theory is

2:56

that one agent shouldn't do every job.

2:59

And this is the theory that I think made

3:00

it viral. If you remember Paperclip went

3:03

viral, you know, 6 to 12 months ago, all

3:05

the agents talking to each other and

3:07

coordinating work. Kind of the same

3:08

theory. You don't want one agent doing

3:10

everything because then it has all of

3:12

these crazy prompts and it gets [music]

3:14

confused and there's context rot. So,

3:16

they advertise you can give each bot a

3:17

job and you can have one master agent

3:19

that coordinates with all of the other

3:21

agents. Now, a couple important points

3:23

about Grok Bot. This is my Hermes agent

3:25

and I'm texting with it every day. It's

3:26

sending me text messages with updates

3:29

and status reports. With Grok Bot, you

3:31

can only communicate with it through the

3:32

app, which actually is really seamless

3:34

and easy and fast. So, I actually don't

3:36

mind it at all. Whether it's on your

3:37

phone or on the computer and it's always

3:39

on because it lives in its own server.

3:41

Now, the big drawback for most people is

3:43

going to be price. So, a lot of you, if

3:45

you're using agents, you know, you can

3:46

spin up Open Claw or Hermes on a $5 a

3:50

per in server or $600 for a Mac Mini one

3:53

time and then you never have to pay a

3:55

monthly fee again. However, with Grok

3:56

Bot, there is a pretty large barrier to

3:58

entry that you need to pay in order

4:00

>> [music]

4:00

>> to use it. Now, they actually just

4:01

dropped the pricing. When I signed up,

4:03

you could not get Grok Bot for $100 per

4:05

month. You needed to be on, I think, at

4:07

least their two or $300 per month plan,

4:10

but it seems that they've brought that

4:11

down since the virality maybe has slowed

4:13

down. Now, a couple important things to

4:15

note here, you do get access with Cursor

4:17

if you want it. However, the big

4:18

difference that you're going to see is

4:19

this. Significantly higher usage across

4:21

chat, imagine, voice, and build. So, you

4:23

can either get it through paying Cursor

4:24

or paying for Grok. Either one will give

4:27

you access to Grok Bot. I recommend

4:29

paying through Grok. That way you also

4:31

get access to the next thing that I'm

4:33

going to share with you a little bit

4:34

later in this video, the big

4:36

transformational piece for my cloud code

4:37

workflow. So, I actually have their

4:39

heavy plan, which is $300 per month. I

4:41

wanted the full usage of Grok's latest

4:44

model. I wanted to fully test it. So,

4:45

this is the one that I selected.

4:46

However, it looks like now you can get

4:48

it through their $100 per month plan.

4:49

So, awesome. Once you upgrade and

4:51

install Grok Bot, you're going to see

4:53

something like this. It looks like a

4:54

really simple interface and you're only

4:56

going to have this one agent here that

4:57

you're talking to. The first thing that

4:59

you're probably going to want to do is

5:00

connect all of your things. So, down

5:02

here you'll see plugins and just like

5:04

maybe you've seen my Victor videos in

5:06

the past, you can just come through and

5:08

connect all of the things that you

5:10

normally use in your day-to-day

5:11

workflows, all of your marketing tools,

5:13

your your operational tools, your Gmail.

5:15

And if you want to see the ones that

5:16

I've connected, here they are. We've got

5:18

Appify, we've got X, we've got Super

5:20

base, you've got Higgs Field. So,

5:21

they've got a lot of these connectors

5:22

already built in. And, of course, you

5:24

can bring outside connectors in. So, if

5:26

you're like, "Hey, I want to integrate

5:27

Hyros." It'll actually collect your key

5:29

securely and it'll build an integration

5:31

for Hyros for you. So, connections,

5:33

really easy, really seamless. Of course,

5:35

that's what we expected. However, that's

5:37

not a huge advantage over Hermes because

5:40

if you are in my programs or you've

5:42

watched my videos, you'll know that we

5:43

use something called Compozio, which

5:45

makes it really easy to connect all of

5:47

your apps for free and you can just give

5:48

this to your Hermes. So, integrating

5:50

plugins and apps isn't such a

5:52

differentiator, but it was in the past.

5:54

So, they do make it really easy. Then

5:55

it's going to walk you through a little

5:56

bit of an interview process and Grok

5:59

bot, the initial agent that you have set

6:01

up, can actually spin up these

6:02

sub-agents based on what tasks and

6:05

scheduled tasks and tools that you're

6:07

using day-to-day. So, as you can see,

6:09

it's spun up all of these additional

6:11

little bots. I've got my cold email ops,

6:13

my long-form desk, my graphic design. It

6:15

actually built these itself, built the

6:17

descriptions for them, come up with the

6:19

the names for them. And if I click into

6:21

any one of these, I haven't actually

6:23

interacted with this one yet. It's just

6:24

a new bot that was created by my master

6:27

Grok bot. But each of these, if I click

6:28

into it, should have a description and I

6:30

should get notifications when it

6:31

finishes. So, this is like the system

6:33

prompt for what this thing actually

6:35

does. Now, in terms of integrations, you

6:37

might think that different sub-agents

6:39

have their own sets of integrations.

6:41

That's not true. They all share the

6:42

integrations. So, you don't have to

6:44

worry about connecting one integration

6:45

to multiple things. Now, what's really

6:47

cool is this master agent, if you do

6:49

have a master agent like I have here and

6:51

I've pinned it. So, as you can see, I

6:52

can unpin, I can edit the profile. And

6:54

this is really cool. You can actually

6:56

see one agent communicating with the

6:57

others and you get push notifications

6:59

when it's done after it's done talking

7:01

to it and getting information from that

7:03

other agent. And it's pretty cool to

7:04

watch them interacting, see it messaging

7:06

back, see a task getting done over here.

7:08

But the big question remains, is it

7:10

really worth it? Do these sub-agents

7:12

actually have a big change in the

7:14

results that you're getting compared to

7:16

using something like Hermes, which

7:17

manages all of my tasks, all of my

7:20

scheduled tasks, all of my tools, all

7:22

from a single agent. So, I had Grok bot

7:24

do a couple of things for me. It built a

7:25

presentation, it did some cold email

7:27

work, it did some N8N work, it did some

7:30

review work, it wrote and posted to some

7:32

Reddit threads. All of the things that

7:33

my Hermes agent was doing, I had it

7:35

temporarily paused while Grok bot took

7:38

over. Now, you might be saying, "Jay,

7:39

how did you get it trained on all of

7:41

these different sub-agents already.

7:43

Like, didn't that take a long time? Not

7:45

really. And what I recommend that you

7:46

guys do once you start using this is

7:48

actually have your Claude code. So,

7:50

let's go ahead and open up my Claude

7:51

code here. I've got my LGJ operating

7:53

system. And the first thing that I did

7:55

was say, "Hey, I just started using Grok

7:57

bot. I would like you to go over my

7:59

entire knowledge base, my scheduled

8:01

tasks, everything that my Hermes agent

8:03

does, and give me a zip folder to train

8:05

Grok bot on all of these tasks and set

8:07

up sub-agents." And it actually did some

8:09

research and it put together a zip

8:10

folder. I gave that zip file over to

8:12

this master agent and it built

8:13

everything for me. Here's kind of how

8:15

that initial conversation went. And

8:16

also, here's all of the routines that I

8:18

had running. And right now, I have all

8:20

of them paused. And I'll tell you why in

8:21

just a second. I said, "You're going to

8:22

be my master agent, coordinate with all

8:24

of the sub-agents. You are the one I

8:26

will talk to directly and delegate

8:27

tasks. I'll be sharing a zip folder that

8:29

has all the information and skills and

8:30

integrations and tooling for what we

8:32

will need. You're going to be taking

8:33

over for my Hermes bot." I just told it

8:35

I configured a bunch of plugins. How do

8:37

I sign into them? And then it prompted

8:39

me one at a time to actually initiate

8:41

the the login cuz some of them use OAuth

8:43

and some of them use API. So, we just

8:45

spent some time configuring plugins.

8:47

Before it was added, it either prompted

8:49

me to sign in like this, authorize, or

8:52

asked me for an API key. So, we just

8:53

went through, configured everything. I

8:55

removed one of the plugins. As you can

8:57

see when it needs to do a sign in or I

8:59

need to do something in the computer,

9:00

this is its screen. And I can actually

9:02

click into a screen and see what it's

9:03

doing at any given time. This is its

9:05

computer that it always has access to.

9:08

I'm going to go ahead and minimize that

9:09

now so we can keep going. And here is

9:11

that zip folder that I dropped to it.

9:13

Unpacking now, it's a 12-bot provision

9:15

pack, not an import. So, it went ahead

9:17

and built all of the sub-agents based

9:20

off of this zip folder that I gave it.

9:22

From there, we configured all of those

9:23

little sub-agents and specialized bots.

9:25

And it did it. And we turned on all of

9:27

the scheduled tasks and Grok bot started

9:29

running. So, all in all, in theory,

9:31

sounds really cool. But what I noticed

9:33

was it wasn't quite doing as good of a

9:34

job as my Hermes agent was. Now, maybe

9:37

that is because I didn't spend nearly as

9:38

much time training it and refining it as

9:40

I did with my Hermes agent. But, from

9:42

what I can tell, Hermes is still the

9:44

better agent harness. Now, yes, you're

9:46

going to need to either deploy it on a

9:48

virtual server or get a Mac mini like I

9:50

have so I can iMessage it. So, there is

9:52

that kind of additional steps, kind of

9:54

additional cost whereas Grok bot you can

9:55

kind of just launch it, it already has

9:57

access to servers, you don't have to

9:58

worry about it. You don't have to worry

9:59

about integrating an LLM cuz it just

10:01

works on Grok. However, I've still found

10:04

my Hermes model to be better than Grok

10:06

bot. So, this is Bob, I text it, this is

10:08

my Hermes agent. He lives on my Mac

10:10

mini. We also can deploy these really

10:12

easily using a VPS. In fact, if you're

10:14

in my AI Automation Insiders community,

10:16

we literally deploy these for you in one

10:18

click. So, this does everything. It's

10:19

connected to all of my tools through

10:21

Compozio. It sends comments and DMs

10:23

across all of the social channels,

10:25

monitors all of my ads. If I tell it to

10:27

do anything, it's able to pretty much

10:29

power through and figure it out. It uses

10:31

my Codex subscription, which I have a

10:33

$200 per month subscription, but you can

10:35

run Hermes on a $20 per month chat GPT

10:38

subscription. And that's pretty much it.

10:40

Or you can get really frugal and get

10:41

like open router or an open source model

10:43

and run it on something cheaper. But, it

10:45

does a great job running on Codex. I

10:47

also really prefer the way that Hermes

10:49

handles context windows, auto

10:51

compaction, scheduled tasks. It just

10:53

does a really nice job and manages

10:54

everything really well. But, the

10:56

question remains, is that one agent

10:58

responsible for all of those tasks done

11:00

it going to do as good of a job at all

11:02

of those different tasks as Grok bot

11:04

will since it had it can delegate all of

11:06

those tasks to sub agents. Now, I am

11:08

somebody who does a lot of tasks.

11:10

There's a lot of different scheduled

11:11

tasks, there's a lot of tools, there's a

11:13

lot of different things that I ask it to

11:15

do. So, I feel like I would be a pretty

11:16

good use case of determining if it was

11:19

capable of it or not. And I'm here to

11:21

tell you, yes, Hermes agent is just as

11:23

capable as Grok bot at executing

11:25

multiple tasks, doing multiple things,

11:28

all without having to delegate to a

11:30

thousand different sub agents. And most

11:32

of you aren't doing things that are that

11:33

complex. You don't have a million

11:35

different tools or businesses that

11:37

you're overlooking. You're just using it

11:39

for pretty simple use cases, social

11:41

media automation, executive assistant

11:43

stuff. This is my Hermes agent running

11:45

on my Mac mini. This is the user

11:46

interface. If you ever want to log in

11:48

and see it, they do have one. And Hermes

11:50

actually does have the ability to

11:52

delegate tasks to subagents through

11:55

[music] Kanban and through agent

11:56

orchestration. They have this built in.

11:58

So, it's essentially the same thing that

12:00

Grokbot is doing just behind the scenes.

12:02

All right. Now, before I go into what

12:04

actually really works and what I

12:05

actually implemented, and it's not

12:07

Grokbox, I'm probably not going to be

12:09

using that long term. I'm going to stick

12:10

to Hermes. It's a great solution. And I

12:12

still think it's the best possible

12:13

solution for most of you that want to

12:15

train a bot over time. Now, if you don't

12:16

want to worry about setting up a VPS or

12:19

getting a Mac mini, we can actually

12:20

deploy them for you and just fill out

12:22

this form and within a minute you'll

12:24

have a Hermes agent that you can message

12:26

back and forth that just works for you.

12:28

It also comes preloaded with all agent

12:30

skills. Or if you don't want us to host

12:32

your Hermes agent for you, that's fine,

12:34

too. You can go to legionai.com/bob.

12:36

Bob is the name of my Hermes agent, it's

12:38

what I call it, as you can see from my

12:39

text messages. I'm messaging back and

12:41

forth with Bob. This way you can just

12:42

copy our configuration into your own

12:45

Hermes agent so that you can host it

12:46

yourself on whatever platform that you

12:48

want. And for those complete

12:50

do-it-yourselfers, just know Hermes is

12:52

free. It's at this GitHub repo, Hermes

12:54

agent. It's updated all of the time.

12:56

It's got 235,000 stars. It is probably

12:59

the most popular and most powerful AI

13:01

agent harness at this time. And in my

13:03

humble opinion, after trying multiple

13:05

different agent harnesses, open claw,

13:07

paper claw, Hermes by far my favorite,

13:10

by far the best. However, [music] if you

13:12

do plan on subscribing to one of Grok's

13:14

heavier plans, which you might, then you

13:16

might not need Hermes agent. And now I

13:18

might just talk you into subscribing to

13:20

one of Grok's paid [music] plans,

13:21

anyways. Now, here's why I recommend

13:23

subscribing to Grok, whether or not you

13:25

decide to use Grokbot. If you already

13:28

have a Hermes agent and it's working

13:29

great, awesome. If not, you might want

13:31

to sign up for Grokbot anyways because

13:33

of this. Grok's latest model 4.6 just

13:36

came out and it's significantly better

13:38

than GPT Soul in most use cases. If you

13:40

just click through most of the

13:41

benchmarks, you'll see that it's pretty

13:43

much just crushing them and I think

13:44

they're going to continue to crush them.

13:46

They're also extremely generous with

13:48

credits just like ChatGPT is. Why is

13:50

this important? Since you're probably

13:52

using Claude code, I'm using Claude code

13:54

as my primary agent driver. So, let's go

13:56

ahead and open up my Claude code. So,

13:57

now that we're in my terminal, this is

14:00

running on tmux, but what's important is

14:01

this is all running Claude code. All of

14:03

these sessions are Claude code. These

14:04

are actual real sessions that I'm

14:06

running right now. Now, before signing

14:08

up for Grok, I had these two skills that

14:11

I would use pretty frequently. They were

14:12

{slash} codex review where I would

14:15

actually have GPT 5.6 Soul where it

14:18

would actually go back and forth [music]

14:19

with Claude to do an adversarial review

14:21

and make sure that the plan that we're

14:23

coming up with and reviewing is actually

14:25

good. And it was a game-changer. It

14:27

absolutely caught way more bugs than

14:29

Claude did on its own and really

14:31

improved my developing process. However,

14:33

GPT 5.6 took a long time to go back and

14:36

forth and review with Claude. It would

14:38

miss things all the time and then it

14:39

would catch things and it just wasn't

14:41

the best possible co-pilot. I thought it

14:43

was at the time and then I would

14:44

actually use another command called

14:45

codex execute to actually write the

14:47

code. Since it's much cheaper than using

14:49

Fable to write code, it would use GPT

14:52

5.6 Soul to actually write the code and

14:54

make the fewest mistakes. And this ran

14:56

on my $200 per month ChatGPT

14:59

subscription. So, it saved me a ton of

15:01

money in Claude code and stopped me from

15:02

hitting my limits. And by the way, guys,

15:04

if you want access to any of my Claude

15:06

code skills, you can go to

15:07

leadgenjay.com/skills

15:09

and I put all of my most important ones

15:11

here so you can come in here at any time

15:13

and kind of see what I'm actually using

15:14

in my day-to-day and install them for

15:16

yourself. So, if any of these commands

15:17

look interesting, they're all in here.

15:19

Anyways, now that I had this $300 per

15:21

month super generous Grok subscription

15:24

with all of these new Grok credits. I

15:25

needed to see how well this could

15:26

perform, how good it was at coding. So,

15:28

what I did is I ran a head-to-head and I

15:30

just posted about this in my AI

15:32

automation insiders community. It is

15:34

free if anyone wants to join. This is

15:35

where I give the latest updates, all of

15:37

my tactics, all of my skills, you can

15:38

download them all within here. But, I

15:40

posted this a while ago and actually

15:42

showed the benchmarks. Here we go. So,

15:43

let's actually review these now. Grok

15:45

did the same task in 12 minutes 53

15:47

seconds compared to about 30 minutes

15:49

with Codex. And that's with 45 test

15:51

passes compared to Codex's 33 test

15:54

passes. So, it was more than two times

15:55

faster and more accurate and ran more

15:58

tests. Now, in order to use this inside

16:00

of your Claude code, you need to install

16:01

the bridge. And this bridge is actually

16:03

a free GitHub repo that you can just

16:05

install. It's from Grok. So, it's this

16:07

Grok build plugin CC. And this is the

16:09

Grok build Claude code bridge. So,

16:11

you'll give Claude this GitHub repo and

16:14

it'll install all of those skills and

16:16

commands that I use inside of my actual

16:19

Claude code terminal. So, now when I'm

16:20

doing anything new, I'm actually going

16:22

to type {slash} Grok and it will call

16:25

Grok on my paid subscription to review

16:27

and write the code. So, now instead of

16:29

just using Fable for everything, cuz

16:31

yeah, if you look at the benchmarks,

16:33

Fable still crushes everything, but it's

16:35

a token consumer. So, now I can use Opus

16:37

5 and have it review and write code with

16:40

Grok's latest model, which is

16:41

significantly better than Opus. And I'll

16:43

get a much better and more accurate

16:45

result. So, that was my big takeaway

16:47

from this whole Grok experiment. Now,

16:48

I'm probably not going to continue to

16:50

use Grok bot. However, I probably am

16:52

going to keep my Grok subscription. I

16:54

don't know whether I keep the $300 or

16:56

downgrade to the $200, but it is proving

16:59

so far to be worth it. Mostly as a

17:01

Claude code copilot. So, I hope this

17:03

video was helpful. Hopefully, it saves

17:05

some of you some time, some money, some

17:07

experimentation. And if you want us to

17:09

deploy a Hermes agent for you,

17:11

pre-trained, ready to go, you can join

17:13

our AI automation insiders community.

17:15

The link is going to be down below. It's

17:16

also got live coaching with me, tons of

17:18

lessons, tons of modules. We go really

17:20

into the weeds about specific topics.

17:22

Hope to see some of you in there. If

17:23

not, I've got a free community that you

17:25

can join as well. I'll also link all of

17:27

the resources that I talked about in

17:28

this video down below in the

17:29

description. And finally, I'd appreciate

17:31

if you hit the like button. Show your

17:33

support. Thank me for making this free.

17:35

And in the comments, I'd love to hear

17:36

any of your feedback. Have you

17:37

experimented with Grok bot? What do you

17:39

think about it? Make sure to subscribe

17:40

to the channel if you love AI for more

17:43

like this. I'll see you soon.

Interactive Summary

The video details the creator's experience with the newly released Grok bot, comparing its performance as an AI agent against his existing 'Hermes' setup. While Grok bot offers a unique, dedicated server architecture and the ability to spin up specialized sub-agents, the creator concludes that Hermes remains more capable and flexible for his professional workflow. The most significant finding, however, is not the bot itself but the integration of Grok's latest coding model into his existing 'Claude code' workflow, which resulted in faster and more accurate coding tasks.

Suggested questions

3 ready-made prompts