HomeVideos

Hermes Agent is blowing me away...

Now Playing

Hermes Agent is blowing me away...

Transcript

542 segments

0:00

Okay, I have to be honest with you.

0:01

Hermes Agent is now better than Open

0:03

Claw. Open Claw has fumbled the bag over

0:06

the last couple months. Every update

0:09

they put out has broken the system. I

0:11

have spent more time fixing Open Claw

0:14

than using Open Claw. Now, this is fine

0:16

for me as a technical person. I don't

0:18

care. I can fix things pretty quickly.

0:19

But for most people out there, having to

0:22

fix your Open Claw every time you update

0:24

it has become a major pain in the wazoo.

0:27

If you were to learn just one AI agent

0:30

at the moment, it's got to be Hermes

0:31

Agent. In this video, I'm going to show

0:34

you how to set up your own 24/7

0:37

Hermes AI employee. An AI employee that

0:40

will literally do work for you at all

0:43

times of day and night, saving you tons

0:45

of time and creating tons of value. I'll

0:47

go over all the differences with Open

0:49

Claw, show you how to set it up, and

0:51

then show you three use cases that I

0:53

promise will change your life. By the

0:55

end of this video, you're going to walk

0:57

away with your own personal AI employee.

1:01

Now, let's lock in and get into it. All

1:03

right, so here's the story on Hermes

1:05

Agent, why it's taken the world by

1:07

storm, and why I'm now recommending it

1:09

for the time being over Open Claw.

1:11

Things change quick, but number one, and

1:13

this is the most important thing, it's

1:15

just plain reliable. Open Claw with

1:18

every single update, basically every

1:21

single update, breaks. I update Open

1:23

Claw, it breaks. I have to spend 20

1:25

minutes fixing it. It's a pain in the

1:28

butt. They release tons of updates. To

1:30

their credit, they are constantly

1:31

shipping, they're constantly updating,

1:33

but I don't know what's going on over

1:34

there. Are they not doing testing? Are

1:36

they not testing out their own tool?

1:37

Because the thing's constantly breaking.

1:39

Hermes Agent, while they do less updates

1:42

and ship a little bit less, they're more

1:45

reliable, and I would much, much, much

1:47

rather have less updates, but they don't

1:50

break and I don't spend tons of time

1:51

fixing the gosh darn agent. The updates

1:54

are better. It seems with every Open

1:56

Claw update, it's just shipping the

1:58

entire kitchen sink, right? It's a

1:59

hundred different things. They don't

2:01

really tie together. It's just a whole

2:02

bunch of things thrown together

2:04

hodgepodge style. With Hermes agent,

2:07

every update has a theme. Every update,

2:09

all the features tie together. As you

2:11

can see here, every release they do, it

2:13

has a theme, it has a name, the tenacity

2:15

release. And all the features in that

2:18

update will tie together, right? So,

2:20

this one had a Kanban board, goals, all

2:22

those things tie together. It seems like

2:24

every update is very well curated. It

2:28

doesn't seem like it's just throwing

2:29

things together. Everything kind of ties

2:31

together, which is really nice and also

2:32

just shout out the aesthetic of the

2:34

whole thing is very nice as well. So, it

2:36

it it it it feels much more well

2:38

curated, which leads to much more up

2:40

time and reliability, which I really

2:42

like. It's self-improving. This is a big

2:44

one. They've put huge emphasis on

2:46

self-improvement. How does the

2:47

self-improvement work? Check out this

2:49

chart here. Shout out Mr. Anon. I found

2:52

your newsletter. I saw this chart. It's

2:53

beautiful. You send a message to Hermes

2:56

agent. It does some sort of execution.

2:59

So, whatever you tell it to do, it

3:01

triggers a skill. So, everything you do

3:04

comes from your skills. And then, if

3:07

that skill doesn't exist or you ask it

3:09

to do something new that it's never done

3:11

before, it improves or creates that

3:14

skill. So, everything you're doing makes

3:17

the Hermes agent better, improves the

3:19

Hermes agent. So, the more you use it,

3:22

the better it gets. They put tons of

3:24

emphasis on this, which is really

3:26

amazing and you see it in real time. You

3:28

can see it here. I had to take a

3:29

transcript of one of my YouTube videos,

3:32

and you can see it goes, gets the skill,

3:34

then it takes the skill, improves it,

3:36

writes at top of it, adds whatever it

3:38

did, and it's just always improving.

3:40

Everything it does is a little bit

3:41

faster and a little bit better. It's

3:43

really, really cool. And I also just

3:44

like that they show you a lot of

3:46

transparency behind all the tool calls

3:48

it does and everything it's doing. So,

3:49

you can really track your AI agent

3:51

employee. And lastly, there's an

3:53

emphasis on experimentation. I really

3:55

like this. These people at new research,

3:58

it seems like they truly are like

4:00

hackers. I really love it. They give you

4:03

tons of tools around using local models

4:06

and things like that. It's training

4:07

models, Laura's, tons of tools around

4:11

machine learning and experimentation.

4:13

So, if you're a little bit of a tinkerer

4:14

like I am, it's really, really cool. If

4:16

you're buying tons of hardware and GPUs

4:18

like I've been telling you to do for the

4:19

last couple months, there's going to be

4:21

a lot of cool things for you to play

4:22

around with here. They are very pro

4:24

local model and tinkering. And one last

4:26

thing I'll add before we get into

4:28

installation and all the really cool use

4:29

cases I promise will blow your mind.

4:31

Swarms, they really built this to be out

4:34

of the box made from using multiple

4:36

agents, spinning up new agents quickly,

4:39

and like sending swarms of agents to do

4:41

things. All you have to say is build a

4:43

new Hermes profile and boom, it spins up

4:45

a new agent that you can now talk to and

4:47

use however you want. I do tons of

4:49

multi-agent stuff, and I'll show you all

4:51

that in a second as well. So, it's

4:53

really built for swarms, which I really

4:55

love as well. All this being said, this

4:58

is why Hermes has kind of taken the lead

5:00

is just this focus on reliability and

5:02

just working has really been amazing.

5:05

Now, let's talk about installation. I

5:06

won't spend too much time here. If

5:08

you're already set up and installed,

5:09

feel free to skip between the chapters

5:11

down below, really easy to do. But, much

5:14

like every other AI agent out there like

5:16

OpenClaw, it's just copying a command

5:19

and then putting it into your terminal.

5:21

So, I do a new terminal here. I press

5:23

paste, you hit enter, and it installs.

5:25

It gives you some options on how to

5:27

install. So, it's going to give you

5:30

options around what messaging service to

5:31

use. Just as I recommend with every

5:34

other AI agent including OpenClaw,

5:36

Telegram is the absolute best. Telegram

5:38

is king for AI agents. They're

5:40

constantly adding new AI functionality

5:42

to Telegram. The formatting is really

5:44

nice. You can see every tool call they

5:47

give you here. They have built-in

5:48

approval, so they ask your approval and

5:50

you can just hit accept in the buttons.

5:51

Everything is built around Telegram

5:54

really, really nicely. As for models,

5:57

this is the big question I get all the

5:58

time. This is where the big debate is.

6:00

After many, many, many tests, I am still

6:04

saying Opus is the best model for AI

6:07

agents. It is still the best model for

6:09

open claw and Hermes agent. It just does

6:12

things best. It has the best taste. It

6:13

builds the best UI. It gets the work

6:15

done the most reliably and efficiently.

6:18

If you have money, if you have a little

6:20

bit money to dedicate to this, I say use

6:22

Opus. And listen, you might say, "Oh

6:24

man, I'm going to spend like 3 to $400 a

6:27

month on Opus API. That's a lot of

6:28

money." Do some reflection here. Using

6:31

Hermes agent is going to improve your

6:34

life. It will add more efficiently, it

6:36

will add more efficiency, it'll help you

6:38

get more done, it'll make you better at

6:39

your job, your career, everything you're

6:41

doing. This is an investment in

6:44

yourself. If you're spending money right

6:46

now on Netflix, if you're spending money

6:48

right now on Xbox Game Pass, that's not

6:50

an investment in yourself. That's an

6:52

investment in you getting distracted,

6:54

right? So, before you go, "Oh, $300 a

6:57

month, that's a lot to spend on Opus

6:58

API."

7:00

This isn't the same thing as buying

7:01

Netflix. You can't compare those two

7:03

services, right? This is an investment

7:05

in yourself, which I think has a decent

7:08

chance of having a return on investment.

7:11

So, just keep that in mind. ChatGPT,

7:13

I'll be honest, ChatGPT 5.4 was

7:16

completely useless for AI agents. It got

7:19

nothing done. It was just a pain in the

7:21

wazoo. But, the good news here is

7:24

ChatGPT 5.5 is actually quite good. Is

7:27

it Opus level? No, it's not. It's not

7:29

Opus level. The user interface it

7:31

creates stink. A lot of decisions it

7:33

makes is kind of questionable. But, as

7:35

opposed to ChatGPT 5.4, it's actually

7:38

usable. It gets the job done. So, if

7:40

you're tight on budget, you can use your

7:42

chat GPT subscription. You could be on

7:44

the $20 chat GPT plan and still get

7:46

pretty good usage. So, I'd say uh you

7:50

know, if you're really tight on budget,

7:51

chat GPT gets the work done and you'll

7:53

still be able to get really good results

7:55

out of your Hermes agent. I'm not taking

7:57

you step-by-step in the terminal setting

7:59

up. It's really easy. Just copy and

8:01

paste the line in, hit enter, and it'll

8:03

take you through step-by-step. Really,

8:05

the only two decisions you have to make

8:07

are models and how to talk to it. But,

8:09

Telegram, Opus, those are ways to go,

8:11

but you can get away with chat GPT here.

8:13

You also have the option for models uh

8:15

to do local models. It's built for local

8:18

models, too. I have a video coming out

8:20

in the next few days that shows you how

8:23

to set up local models and power your

8:25

Hermes agent with it. They get the job

8:28

done. They're really good and they're

8:29

completely free. This is my Hermes agent

8:32

powered by Qwen 3.6, which is running on

8:34

my DGX Spark, and it is great. It does

8:37

really good coding. It can answer my

8:40

questions quickly and I'm not spending a

8:42

single dollar on it uh other than the

8:44

power going into the computer. So, this

8:47

is another option as well if you have

8:48

really good hardware. My recommendation

8:51

is if you're going to go local, set up

8:53

two agents. So, have two agents, one

8:55

local, one cloud, so you can give the

8:57

right task to each agent. If it's a

8:58

super complex task you need done

9:00

quickly, probably want to go cloud. If

9:02

it is a simpler task, then you can go

9:04

local and save tons of money. So, let's

9:06

get into the first things you should be

9:08

doing after you set up your Hermes

9:11

agent. I'm going to show you how to make

9:13

it personalized and make it your own.

9:15

Once we do that, we'll get into the

9:17

actual execution and the use case.

9:19

Again, feel free to jump around down

9:21

below if you need to. Also, if you

9:22

learned anything at all so far, make

9:24

sure to leave a like down below. Tell me

9:26

in the comments what you want me to make

9:28

a video on next. Hermes use cases, vibe

9:31

coding stuff with Claude code. Super

9:33

curious, let me know down below. Also,

9:35

subscribe and turn on notifications cuz

9:37

all I do is make amazing videos about

9:40

AI. I have a second channel I just

9:42

started, Alex Finn Lab. Subscribe to

9:44

that down below. I'm going to do more

9:45

off-the-cuff, just talking about what's

9:47

on my minds when it comes to AI. And

9:50

then the last plug, I swear to God, Vibe

9:52

Coding Academy. If you want to join my

9:53

community, I do live bootcamps every

9:55

Friday. I do Hermes bootcamps in there.

9:58

Link for that is down below. Join, join,

10:00

join. It'll be the best decision you

10:01

ever made in your life. So, here's the

10:03

first things you need to do when you set

10:05

up your Hermes agent. These are the

10:06

first things you need to do, no matter

10:07

what AI agent you're setting up. Super

10:10

critical, sets the stage for what you do

10:13

next, makes every task your agent does

10:16

way more impactful. And that is two

10:18

things. Two things we're about to do

10:19

here. Brain dump and reverse prompt.

10:22

What are those two things? Brain dumping

10:24

is first telling your agent everything

10:26

about yourself. Your career, your

10:28

ambitions, your goals, your interests,

10:30

your hobbies, everything about yourself,

10:32

you brain dump it in here. Now, let me

10:34

show you. So, here's a quick one about

10:36

me. I am Alex Finn, an entrepreneur and

10:38

content creator. I have 200,000 subs on

10:40

YouTube, 450 on X. I'm working on a new

10:42

startup, Henry Intelligent Machines,

10:44

which will help people build their own

10:45

businesses. I am super into AI. All my

10:47

content's about AI updates. I love AI

10:50

agents and like Openclaw and Hermes. I

10:52

also enjoy Boston sports like the

10:53

Celtics and Patriots. I live stream

10:55

Monday, Wednesday, Friday, 11:00 a.m.

10:57

Pacific. You brain dump everything about

10:59

yourself. This is the first message

11:01

you're sending, right? It allow the

11:03

agent to create a whole bunch of

11:05

important memories about you. Oh, Hermes

11:08

has an incredible memory system. So, you

11:11

brain dump it in, Hermes learns

11:13

everything about you, and you're good to

11:15

go. You should be more detailed than

11:17

this. I just did it kind of light for

11:19

demo purposes, but you should go in and

11:21

put in as much as humanly possible about

11:24

yourself, cuz it will create all of

11:26

those memories. Next, we are going to

11:28

reverse prompt. If you've ever watched

11:30

one of my videos before, you know what

11:32

reverse prompting is. I'm the biggest

11:34

reverse prompting proponent on planet

11:36

Earth. What does that mean? That means

11:38

instead of you prompting the agent, you

11:40

have the agent prompt you. You go, "What

11:42

are things we can do? Based on what you

11:45

know about me, what should we be doing

11:47

here?" Right? It's asking the agent

11:49

questions and seeing what they think.

11:52

Cuz remember, these agents are smarter

11:53

than us. This This is super

11:54

intelligence, right? So, here's what

11:56

we're going to do. Based on what you

11:57

know about me, what are the best

11:59

workflows we can implement? How could

12:01

you improve my life? Ask me any

12:03

questions you have for me. This is a

12:06

reverse prompt. The issue people have is

12:08

they go into these agents and they go,

12:10

"Uh I don't have any use cases. I quit."

12:12

Right? You should never be like that. If

12:15

you don't know what to do, if you don't

12:17

have any use cases, you go to your agent

12:19

and ask, "Hey, what can you do for me?"

12:21

Right? Based on what you know about me,

12:22

what can you do? The issue is a lot of

12:25

people, they just copy other people's

12:26

use cases, which are fine. I mean, I'm

12:28

going to give you a few of my own right

12:30

after this, and you should definitely

12:31

copy any you find interesting. But, the

12:33

most important thing here is you should

12:35

be using use cases that work for you,

12:37

that are custom for you, that your agent

12:39

does based on what knows about you. So,

12:41

you want to do this reverse prompt, and

12:43

it gets you custom use cases built

12:46

specifically for you that will improve

12:48

your life. I promise, if you do those

12:49

two things, brain dump and reverse

12:51

prompt, you're going to walk away with a

12:53

whole set of automations that will make

12:56

your life so much better. So, do that

12:59

exercise. Next, I want to go over three

13:01

use cases: a beginner use case, an

13:03

intermediate use case, and an advanced

13:05

use case. Let's go through those three

13:07

right now. Use case number one, and I

13:09

will put all these prompts down below so

13:11

you can copy and paste them and get set

13:12

up right away. This first use case is

13:15

for discovering and using and staying on

13:17

top of new AI tools. There are many,

13:20

many AI tools out there, most of which

13:23

you're not going to use, but you should

13:24

be experimenting and using new ones

13:26

every day. So, here's how this is going

13:28

to work. Every morning, tell me the

13:30

latest AI news with an emphasis on new

13:33

AI tools I've never used. Recommend a

13:35

new AI tool I need to use based on my

13:38

workflows. You're going to hit enter on

13:39

that. What's going to happen when you

13:41

hit enter on this, it's going to set up

13:43

a cron job for every single morning.

13:46

Then what's going to happen is when that

13:47

cron job triggers, your agent's going to

13:50

go, it's going to look at your

13:51

workflows, it's going to look at AI

13:53

tools you've used with it before, it's

13:55

going to look at different things you've

13:56

done with AI, and it's going to go and

13:59

find new and popular AI tools that it

14:02

will recommend based on your workflows.

14:04

Now, every morning you will discover new

14:07

AI tools to use that will improve your

14:09

life because they're custom based on

14:12

what you do. This is one of the

14:14

advantages of Hermes Agent, and this is

14:15

one of the things you should be doing is

14:17

taking advantage of its incredible

14:19

memory. It knows everything about you.

14:21

It knows everything you do, and now it

14:24

can make you custom recommendations

14:26

based on what it knows about you, and it

14:28

can help you stay on top of those latest

14:30

AI tools. So, this is huge. You want to

14:33

use this. I'm going to send this to you

14:34

down below in the description. Copy,

14:36

paste that, hit enter, you're good to

14:38

go. That's the beginner one. Here's an

14:40

intermediate one. Here's one I think

14:41

will be helpful for absolutely everyone

14:43

watching this video, and that's the

14:45

proactivity check-in. I think one of the

14:47

number one questions I get is, "How do I

14:49

make my agent more proactive, Alex? How

14:51

do I make it do things without me having

14:54

to ask?" Well, this is one of the ways

14:56

to do it, and this is a super powerful

14:58

method. Again, prompt down below, the

14:59

proactive check-in. Here's the prompt

15:02

down below. I want a proactive check-in.

15:04

Every morning, ask what my one big

15:06

priority is that day. Based on that,

15:08

update your memories about me and make a

15:10

recommendation to me on how you can help

15:12

my life or insights on things I can do

15:15

better. Again, we're taking advantage of

15:17

the memories. We're taking advantage of

15:19

the self-improvement and the skills.

15:21

Every morning, it's going to check in on

15:23

me. I'm going to give it my one big

15:25

priority that day and then it's going to

15:28

do recommendations for me. It's going to

15:29

recommend workflows so it can automate

15:31

those tasks for me or it's just going to

15:33

give me insights on my life that will

15:35

improve what I'm doing. This is huge and

15:37

this will make your agent way more

15:39

proactive. Again, this will set a cron

15:41

job for every morning. For those who

15:43

don't know what cron jobs are, they're

15:44

just scheduled triggers for your agent

15:46

that wakes them up. It'll schedule it

15:48

for every morning and help your agent be

15:50

way more proactive. So let's do a more

15:52

advanced one. This is really cool. So

15:54

this is using the new hype frames skill.

15:57

For those who don't know, there's tons

15:59

of new AI video creation skills coming

16:02

out for these agents that make it really

16:04

powerful for your agent to make videos.

16:07

There are many different reasons you'd

16:08

want to make AI generated motion videos.

16:12

For me, I'm doing it so I can make more

16:14

content so I can put out more helpful

16:17

educational graphics on my social media.

16:20

But for you, there could be many

16:21

different reasons you do this whether

16:22

you're getting into content, whether you

16:24

just want to help people out. You want

16:26

to use hype frames. The other big one is

16:28

Remotion. I think hype frames is pretty

16:29

cool. And what this will do is it will

16:32

have your agent build you out motion

16:34

graphic videos that you can use for any

16:36

reason. So use the hype frames skill to

16:38

build me a 60-second short on why people

16:41

should be using the Hermes agent. Make

16:43

the screen changes quick, visually

16:44

beautiful and eye-catching. Actually,

16:46

I'm going to make this one 30 seconds.

16:47

Make it a little bit quicker. I'm going

16:49

to hit enter. We're going to see what

16:50

this video looks like. I promise you

16:51

it's going to look cool. But there's so

16:53

many things you can be doing here with

16:55

these new skills. You want to stay on

16:56

top of the new skills. That could be

16:58

another kind of proactive thing you have

17:00

it do every morning is have it search

17:02

for other skills other Hermes agents

17:04

have and recommend it to you for how to

17:07

learn. See, this is amazing. It didn't

17:09

have the hype frames skill installed so

17:11

it went search for the hype frames

17:12

skill, found it. Now it's installing it.

17:14

It's hyperframes. Apologies,

17:16

hyperframes. I had a typo in there. Now,

17:18

it's going to install it and get it set

17:20

up and working and now we're going to

17:21

see a really cool video it's going to

17:23

generate for us. All right, boom, here

17:25

we go. That was actually pretty quick.

17:27

That took like 5 minutes.

17:29

Uh let's see what we got here. So, this

17:30

is a fully generated short. I can put

17:33

this on YouTube if I want. Here we go.

17:35

This is the future. This is your AI. It

17:36

runs itself. Look at this. So, it gives

17:38

you nice motion graphics, really nice

17:40

colors. I like the way things go in and

17:42

out. Now, would I just post this raw?

17:44

No, but maybe now I put a voiceover on

17:47

it. Maybe I put my face in the bottom

17:48

corner. And now, instead of having to

17:50

spend thousands of dollars on an editor

17:52

or hire a team, my Hermes agent just did

17:54

it for me and saved me thousands of

17:55

dollars. You now have unlimited control

17:58

to build any sort of videos or motion

18:00

graphics you want. You don't need to be

18:01

a professional anymore. The gatekeeping

18:03

has been taken away, which is really

18:05

amazing. That was the more advanced use

18:08

case. What I'm thinking about doing in

18:09

the next few days is doing like a video

18:11

where it's like 10 use cases on Hermes.

18:14

Just give you a whole gamut use case

18:17

focused video. I think that would go

18:19

hard. Let me know if you want that down

18:21

below as well. Again, if you learn

18:23

anything, leave a like, subscribe, turn

18:24

on notifications, super super critical.

18:27

Vibe Coding Academy doing another live

18:29

boot camp on Friday on Hermes agent.

18:31

Feel free to join, ask me questions,

18:33

chat with me, whatever you want. Link

18:35

for that is down below as well. Hope

18:37

this was helpful. I absolutely love

18:39

making these videos for you guys. It is

18:42

an honor you would even spend 30 seconds

18:44

watching my videos of me. I do this out

18:46

of love, so thank you so much for

18:48

watching these. Hope it was helpful and

18:49

I will see you in the next video.

Interactive Summary

This video presents the 'Hermes Agent' as a more reliable and well-curated alternative to the Open Claw AI agent. The host explains the benefits of switching to Hermes, including its self-improving capabilities, support for multi-agent 'swarms,' and its focus on stability. The video covers the setup process using Telegram, discusses model recommendations (such as using Claude Opus or GPT for varying budgets), and provides practical, life-improving use cases like brain dumping, proactive check-ins, and using automated motion graphic tools.

Suggested questions

3 ready-made prompts