HomeVideos

100 hours of Hermes Agent lessons in 19 minutes

Now Playing

100 hours of Hermes Agent lessons in 19 minutes

Transcript

534 segments

0:00

Hermes Agent is the most powerful

0:02

software ever made. When used correctly,

0:04

you literally have your own full-time AI

0:07

employee. Here's the thing though, when

0:09

used incorrectly, it's basically just a

0:12

normal chatbot. In this video, I will go

0:15

over every lesson I've learned from

0:18

using Hermes Agent for hundreds of hours

0:20

the last 4 months. If you stick with me

0:22

until the end, I promise you you will be

0:24

100 times more productive. Now, let's

0:27

lock in and get into it. I have a ton of

0:29

lessons to go through in this video, but

0:31

let's start with one of the number one

0:32

asked questions I get, which is which

0:35

model should I be using for Hermes?

0:37

Well, I have an official recommendation

0:39

for you here. If you want the absolute

0:42

best-in-class performance, there is no

0:45

model better than Opus. It is the

0:47

absolute best agentic model ever made.

0:49

And now I know what you're going to say,

0:51

"Wait, Alex, isn't that API pricing?"

0:53

Yes, yes, it is. I have spent $1,400 the

0:57

last month on Opus credits for my Hermes

1:00

Agent alone. Yes, I'm spending a

1:02

tremendous amount. Yes, on average, I am

1:04

spending about $40 a day. Yes, I was in

1:06

Cabo this one day where I didn't spend

1:08

anything at all, but I am spending a

1:11

tremendous amount on Opus for Hermes

1:14

Agent, but for good reason. It is the

1:16

only model on planet Earth where when I

1:18

plug it into Hermes Agent, I know for a

1:21

stone-cold fact when I give my prompt to

1:24

it, it's going to finish the task no

1:26

matter what. The way I like to think

1:28

about it is this, when you give Opus a

1:30

task to do in Hermes Agent, even if it

1:33

loses a leg halfway through a marathon

1:35

race, it is going to crawl its way to

1:37

the end of the finish line. With every

1:39

other model, it stubs its toe halfway

1:42

through the race and then it rolls over

1:44

and gives up. Opus, when used in Hermes,

1:47

no matter what, it finds a way to

1:49

complete your task. And for me, that is

1:52

worth a tremendous amount of money. I am

1:54

willing to spend double what I'm

1:56

spending now in order to use Opus and

1:57

Hermes cuz it runs my business so much.

2:00

But I get it, not everyone can spend

2:02

thousands dollars a month on Opus. So,

2:04

you have other options. If you're

2:06

already paying for a ChatGPT

2:08

subscription, which most people are, you

2:10

can plug that in. ChatGPT 5.5 and later

2:14

has been very good with Hermes. It's not

2:17

Opus level. It still trips up. It's

2:19

still not as warm. It's still not as fun

2:20

to talk to. It's still not as great at

2:22

tasks. But it's usable. Everything

2:25

before 5.5 was not usable. It was quite

2:28

frankly horrible. Now, if you want to

2:29

save even more money, GLM 5.2 is also a

2:33

great model. You can plug it in. It's a

2:36

fraction the price of both of these, and

2:38

it'll still get the job done. It's a

2:40

little bit more robotic than ChatGPT and

2:42

Opus, probably cuz it's distilled from

2:44

these other two, but it's still very

2:46

good, and it is very, very cheap. So, if

2:48

you're in ultra money-saving mode, I'd

2:50

recommend GLM 5.2. So, it's really up to

2:52

you. You got a menu. Do you go with the

2:54

best in class, the filet mignon Opus,

2:56

where you can be sure every task will be

2:59

done? Do you go with ChatGPT? Do you

3:01

already have a subscription ChatGPT?

3:03

That's fine, plug it in. You won't have

3:05

to spend more money. Or if you want to

3:07

go in ultra saving mode, GLM. Listen, if

3:09

you're running a serious online business

3:11

and you're making revenue, I say spend

3:14

the money, go with Opus. Your job will

3:16

get done. The tasks you give it will get

3:18

done, and that kind of peace of mind is

3:20

super important. So, my official

3:22

recommendation's Opus. If you already

3:23

have an account, just go with ChatGPT.

3:25

Cost savings, go GLM 5.2. Second lesson

3:28

I learned, and this is a massive one

3:30

that literally 100% of people watching

3:32

this video should do, you need to have

3:35

at least two Hermes agents running at

3:37

all times. Only having one is not

3:39

enough. Let me walk through this for

3:41

you. So, Hermes, I named my first Hermes

3:43

agent Hermes. I just like that name.

3:45

Hermes is my main agent. It's running

3:47

off Opus just as we just talked about,

3:49

but I also have a second agent,

3:51

GPT-Me's. Why do I have a second one,

3:53

GPTme's? Obviously, it's running off

3:55

ChatGPT. The reason why you need at

3:57

least a second and sometimes third and

4:00

fourth Hermes agent is they all watch

4:04

over each other. Hermes isn't perfect.

4:06

It breaks at different times. I find

4:08

Open Claw breaks a little bit more.

4:10

Hermes breaks less, but it still does

4:13

break. And when it breaks, you need

4:15

failovers. You need ways to fix it. So,

4:17

just yesterday, my account, my ChatGPT

4:20

account connected to GPTme's, went down

4:22

for some reason. The token expired or

4:24

something went wrong. If it was my only

4:26

Hermes agent, I wouldn't know what the

4:28

hell to do. I wouldn't know how the hell

4:30

to fix it. I'd probably panic. I'd

4:31

probably be like, "Oh, my best friend's

4:33

gone. I don't know where he went." But

4:35

when you have multiple Hermes agents,

4:38

they can all monitor, watch over each

4:41

other, and fix each other automatically.

4:43

So, as you can see here, when GPTme's

4:45

went down, I took a screenshot, gave it

4:48

to Hermes, my main agent, and said, "My

4:49

other agent, GPTme's, is down. Giving me

4:51

errors. Can you fix it?" And literally,

4:53

as you can see here, didn't have a

4:54

message again, came back to me, said

4:56

it's done, and it worked. GPTme's was

4:58

back up. You need to have that failover.

5:01

And so, if you have any extra accounts,

5:04

using Opus and you have a ChatGPT

5:06

account, if you're using ChatGPT and you

5:08

have a Gemini account or Google account,

5:10

you need to plug them in so that you

5:12

have those extra Hermes bots available.

5:15

If you want to take this a step further,

5:17

I'd even recommend using like an Open

5:19

Claw as your backup account, too, cuz

5:21

then you can have them not only watch

5:23

over each other to make sure they're up,

5:24

but then you can also take advantage of

5:26

any features that one or the other comes

5:29

out with that the other is not using.

5:31

Now, how do you set up a second or third

5:34

Hermes agent the best way? Well, you

5:36

basically have two options. The easiest

5:38

way is just go to one of your agents and

5:40

say, "Set me up a new Hermes profile.

5:42

Make it powered by ChatGPT." So, profile

5:45

here, Hermes profile, Profile is the

5:48

terminology in Hermes. Basically, for

5:51

other agents, they're called profiles.

5:53

So, if you set up a second Hermes

5:54

profile, you're basically setting up a

5:56

second Hermes agent. So, you just go and

5:58

say, "Hey, set me up a new Hermes

6:00

profile." Then, you let them know what

6:02

model you want to use. If you want to

6:03

get even more custom, you can say, "Name

6:05

it Harry." Or whatever you want. You put

6:07

it in there, you hit enter, it'll build

6:09

it for you. There's one other way, and

6:11

that is through the Hermes dashboard.

6:13

So, if you go into the Hermes dashboard,

6:15

and then you go over to profiles, you

6:18

can go in and click create in the top

6:21

right, and then you can go through this

6:22

nice user flow to choose a model, give

6:25

it a name, everything you want. You can

6:27

put it in there, and that'll build the

6:29

profile for you as well, and launch it.

6:31

Now, you have a second and third Hermes

6:33

agent good to go that will watch over

6:35

each other. That was my second lesson

6:37

learned. Before we get into the third

6:38

lesson, after 100 hours with Hermes

6:42

agent, one thing became clear. The more

6:44

agents you build, the better you get at

6:47

building them. You start seeing

6:48

different patterns. My friends over at

6:51

HubSpot, who have sponsored this video,

6:53

put together a free AI agent building

6:56

course just for my audience. You can

6:58

access with the link down below. It's 18

7:01

free videos on how to build your own AI

7:05

agents. They break AI agents down to

7:07

every component you need to know, from

7:10

memory to tools, to plugging in the

7:13

models. You can see memories and tools

7:14

here. How to test and publish your

7:16

different agents. This allows you to go

7:18

a lot more hands-on on building AI

7:21

agents. They also talk about a lot of

7:23

other really important, interesting

7:25

concepts when it comes to the

7:26

nitty-gritty of AI agents, like

7:29

automations versus agents, fixed step

7:31

versus dynamic reasoning, which would

7:34

have saved me a ton of headaches early

7:36

on building these AI agents out, and how

7:38

to plug your AI agents to the different

7:40

tools, like N8N, Notion, Make, Chatbase,

7:44

and a whole lot of other platforms. So,

7:46

you'll be able to plug your AI agent to

7:48

every single tool you use on your

7:50

computer. It's a free, complete course,

7:52

18 different videos. The link is down

7:55

below. Make sure to check it out, and

7:57

thank you to HubSpot for supporting the

7:58

channel. Lesson number three I've

8:00

learned over the last several months and

8:02

hundreds of hours using Hermes Agent.

8:05

The next tip I want to go over is a

8:06

massive mistake I see a lot of people

8:08

make, which is this. They buy their own

8:11

computer for their Hermes Agent. They

8:13

give it their own Gmail account. They

8:15

give it their own iMessage account. They

8:17

give it accounts for every single little

8:19

thing it does. They completely separate

8:22

it from everything they do. This is a

8:24

massive mistake. You do not need to buy

8:27

your own Mac mini for every agent you

8:29

have. You do not need to give your agent

8:31

its own Gmail account, its own Apple

8:34

account. You are adding needless

8:36

friction to your entire experience. And

8:39

I get why people do, right? They have

8:40

security concerns. Oh, Alex, if I give

8:43

my Hermes Agent access to my Gmail,

8:44

can't it leak all of my emails? Oh,

8:47

Alex, if I give it access to my iMessage

8:50

account, won't it text my ex-girlfriend?

8:52

No, that's not how any of this works.

8:54

Hermes Agent only does exactly what you

8:57

tell it to do. If you tell it to write

8:59

you a tweet, it isn't going to go into

9:01

your iPhotos and leak all your nudes. If

9:04

you tell it to get the latest AI news,

9:07

it isn't going to go then and email your

9:10

mom. That's not how any of this works.

9:13

It is an AI. It's not sentient. You give

9:15

it a prompt, it does the prompt you give

9:18

it. That's it. So, as long as you have

9:20

personal accountability and think deeply

9:23

about what you're telling your agent to

9:24

do, it's not going to do anything you

9:26

don't want it to do. My question to you

9:27

is this, how many people do you know

9:29

have had a security incident with

9:30

Hermes? Truly, tell me. How many people

9:32

do you personally know have had their

9:34

Hermes Agent go and then email everyone

9:38

on their contact list. The answer is

9:40

probably none. You probably don't know a

9:43

single person who's ever had a security

9:45

incident. That's because a lot of the

9:46

security concerns are largely overblown,

9:49

and I know what's going to happen now.

9:50

I'm going to get a thousand comments

9:52

that say, "Alex, you jerk, you're

9:54

telling people do this dangerous" Shut

9:56

up. I don't care. If you have some

9:58

personal accountability, you act like an

9:59

adult, you're not going to need to add

10:01

20 different accounts for everything

10:03

your Hermes agent does. You're just

10:05

slowing down everything you do, adding

10:06

tremendous amounts of friction. Put it

10:08

on your main computer, put it on your

10:10

main accounts, and then be a responsible

10:12

adult. The next lesson I learned is what

10:14

platforms to be using your Hermes agent

10:16

on. I use three main platforms for

10:19

Hermes agent. When I am at my computer,

10:21

I use Hermes desktop. Hermes desktop is

10:24

excellent. It's the best user experience

10:26

for an AI agent on planet Earth. You can

10:28

quickly switch between all your

10:30

profiles. So, if you're doing

10:32

multi-agent profiles like I told you to

10:34

towards the beginning of this video, you

10:35

can switch to them very easily. You can

10:37

see all your cron jobs very easily in

10:40

here. You can pin your different

10:41

sessions instead of having to having one

10:43

massive session. You can have separate

10:45

sessions, pin them, pop them out, talk

10:47

to multiple Hermes agents at once. It's

10:49

excellent. When I'm on the go and I'm

10:52

doing deep work, I use Telegram. So, I'm

10:54

still using the Telegram app on my

10:57

phone, but only for deep work, only if

10:59

I'm doing multi-threading things on my

11:01

phone. I hope they come out with a

11:02

mobile app soon. I know OpenClaw just

11:04

came out with one. But, if I'm on the go

11:06

and I'm doing deep work, I'm using

11:07

Telegram app on my phone. But, other

11:09

than that, this is a new addition. When

11:11

I am on the go and I have quick prompts

11:13

I need to do, I'm using iMessage.

11:15

iMessage is actually fantastic with

11:17

Hermes agent. This is a brand new

11:19

addition as well to Hermes in their

11:21

latest update. They had out of the box

11:23

iMessage. I highly recommend setting

11:25

iMessage up. I added a contact called

11:27

Hermes to my phone. I pinned it to the

11:29

top of my iMessages, and now I can

11:32

message my Hermes agent through iMessage

11:35

the fastest easiest app that I'm on at

11:37

all times to get things done quickly. So

11:40

quick tasks iMessage on the go deep

11:43

tasks telegram and when I'm at my

11:45

computer Hermes desktop. That's how you

11:47

need to be using Hermes agent. The next

11:49

tip is related to what we just talked

11:51

about and that is the deep work on the

11:52

go. There is some amazing new features

11:56

inside of telegram when you're doing

11:58

deep work on mobile on the go on your

12:00

phone and that is formatting inside the

12:04

messages. You now can have so many nice

12:07

pieces of formatting in your messages.

12:09

As you can see here you have tables, you

12:11

have bold, you have paragraphs, you have

12:12

different things you can do. You have

12:14

really nice in-depth formatting now

12:16

inside of telegram. I'd highly recommend

12:18

taking advantage of this. Set up cron

12:21

jobs, ask questions where you ask for

12:24

generated tables. So I have it do stock

12:27

research on the top AI related stocks

12:29

every day. I tell her to put it into a

12:31

table. I tell her to give it a bunch of

12:33

ratings and information there and now

12:35

every morning I wake up to this really

12:37

nice table with information on AI stocks

12:40

I should be investing in. Take advantage

12:42

of this new formatting in Hermes. Set up

12:45

different cron jobs where it gives you

12:47

tables of stock information or tables

12:49

with your top performing content or

12:51

tables with your most important emails.

12:54

It makes it a really nice way to use and

12:56

organize everything going on inside

12:58

Hermes. The next lesson learned is

13:00

around performance. One of the big

13:02

complaints I get is hey my agent slowing

13:04

down and it's getting stupid. It's

13:05

always busy. People have a lot of

13:07

performance issues with their Hermes

13:09

agent. What I find the number one

13:12

culprit to be is cron jobs. A lot of

13:14

your cron jobs run in the background and

13:17

much like laws and rules, it's very easy

13:19

to make new ones. It's very hard to take

13:21

them away. And so what I find is if you

13:24

need to improve the performance of your

13:26

agent, one of the best things you can do

13:29

is just eliminate and clean up old cron

13:31

jobs. If you're in the desktop app and

13:33

you click the cron button right there,

13:35

you actually see a really nice list of

13:38

all your cron jobs. And what I like to

13:39

do is every so often, I go in and I just

13:42

find which ones I don't really need

13:44

anymore and I click pause on them. So,

13:46

you can come in any that you're not

13:47

really using, pause it, and what happens

13:50

is is when these crons run, it slows

13:53

down your agent. So, if you have tons of

13:55

cron jobs going and I bet you forgot

13:57

about a lot of the ones you scheduled,

13:59

make sure to just go in every once in a

14:01

while, once a week, pause any ones

14:04

you're not really using. You'll

14:05

instantly get a performance boost on

14:07

your agent, also save usage and money. A

14:10

lot of these take up a lot more tokens

14:11

than you think. So, always clean up your

14:13

crons and the easiest way to do that is

14:15

through the desktop app experience. The

14:17

next massive tip I have for you, and

14:19

this is an important one for anyone who

14:21

has multiple devices. So, if you have a

14:23

computer and a phone, or if you have Mac

14:26

minis, or if you have DGX Sparks, any if

14:28

you have anything more than just one

14:30

device, you need to install Tailscale.

14:33

It is completely free. I am not

14:35

sponsored by Tailscale whatsoever. I

14:36

just absolutely love this application.

14:39

Basically, what it does is it creates a

14:41

private network for all your devices,

14:44

which allows your Hermes agent to then

14:47

be able to move across all your devices.

14:50

This is great if you're running local AI

14:52

models. This is great if you do vibe

14:54

coding. This is great if you have a job

14:56

and you do work and you like to move

14:58

documents around between all your

14:59

computers. You can easily go in and just

15:01

say, "Hey Hermes, go over to this other

15:03

device and do something." So, if I want

15:04

to say, "Let me know which local models

15:09

I have running on the DGX Spark." I can

15:13

hit enter on that and now my Hermes

15:15

agent will go from the Mac Studio it's

15:17

on right now, go over to my DGX Spark,

15:20

which is also on Tailscale, and be able

15:22

to tell me what's running on there. So,

15:23

as you can see, it's SSH-ing over to my

15:26

other device. It's checking on the

15:27

models running. And if I wanted to go

15:29

then go and like run a new local model,

15:32

download something new on there, run it

15:34

for me, it can go and do that for me. It

15:37

saves me so much time cuz I don't need

15:39

to plug all my devices into monitors. I

15:42

just have my Mac Studio plugged into the

15:44

monitor. My other computers, my five

15:46

other computers I have, they're not

15:47

connected to monitors. I just have my

15:49

Hermes agent go over to those computers

15:52

and run them for me. This is where your

15:54

Hermes agent really turns into like your

15:57

own AI employee cuz it's basically just

15:59

on your other computers for you doing

16:01

whatever work you need to do. If I'm

16:03

ever on the go and I have a presentation

16:05

on my other computer and it's not on my

16:06

laptop, I say, "Hey Hermes, go on my

16:07

other computer, click and drag it over

16:09

here and send it to me." It'll bring the

16:11

documents in between computers. You need

16:14

to have Tailscale. It basically allows

16:16

your Hermes to run your entire fleet of

16:17

devices. Has some other benefits, too.

16:20

If you're vibe coding and you have your

16:21

app running on localhost on like your

16:23

main computer, you can have it if

16:26

everyone's on the same Tailscale

16:27

network, you can access your localhost

16:30

from your phone. So, you can test your

16:31

apps out on the go if you want. So many

16:34

benefits of Tailscale. It's completely

16:36

free. There's I think there is a paid

16:38

version, but I don't even know what's on

16:40

it. I've just been using paid I've just

16:41

been using free forever. So, make sure

16:43

to check that out. My eighth lesson

16:45

learned is around reverse prompting.

16:47

Hermes agent is so powerful that we

16:51

can't even really comprehend everything

16:53

it can do. But this prompt I'm going to

16:55

give you down below that reverse prompts

16:57

Hermes agent helps you figure out how it

17:00

can help your life the most. Every

17:02

morning I do a morning interview with my

17:04

Hermes agent where I basically tell it

17:06

to ask me a bunch of questions. What my

17:08

priorities are for that day, what tasks

17:10

I'm working on, what stresses me out at

17:11

the moment, what's on my plate. It then

17:14

takes all those things and then figures

17:17

out what it can take off my plate for

17:18

me, what tasks it can do for me, what it

17:21

can automate, what it needs my

17:22

permission to do, and starts doing those

17:25

things for me. Every day I learn more

17:27

and more about what Hermes agent can do

17:30

for me, and and what it can take off my

17:31

plate. You need to be doing this morning

17:34

interview every single morning. It takes

17:36

about 5 minutes, but it saves you tons

17:39

of time after because it helps your

17:41

Hermes agent figure out what it can do

17:43

for you. I promise you're not getting

17:45

the most out of Hermes agent. I promise

17:47

you're going to be automating so much

17:48

more of your life. You just need to be

17:50

able to figure out what those things

17:52

are, and reverse prompting is the way

17:54

you figure it out. And this prompt that

17:57

has your agent interview you is the best

18:00

way to reverse prompt. So, take this

18:02

prompt, run it every single morning. I

18:03

promise you'll find two to three new

18:05

jobs for your Hermes agent to do that

18:07

will be super helpful for you. And the

18:09

ninth tip I got for you, this is an

18:11

important one. After you do your morning

18:14

brief, and it gives you a bunch of tasks

18:16

it can do, put those tasks in your

18:19

Kanban board for the agent to take over

18:21

and actually do for you. This is the

18:23

best way to stay organized on your

18:26

tasks. All you need to do to see this

18:28

Hermes Kanban board is go in, type

18:31

Hermes-dashboard

18:33

into your terminal, whether using Ghosty

18:35

or just your built-in terminal on your

18:36

computer, type Hermes-dashboard. It will

18:39

pop open this website. You click on

18:41

Kanban board on the bottom left, and you

18:43

have this nice Kanban board where you

18:44

can add in tasks, move them over, assign

18:47

them to your agent. And so, all the

18:48

tasks your agent comes up with from your

18:51

morning interview, you can put it in

18:53

here, and it'll start taking care of it

18:54

for you. This is an amazing way to stay

18:56

organized with the tasks your agent has,

18:58

with the tasks you have. You put them in

19:00

here, you're good to go. You can

19:02

organize your day really, really nicely.

19:04

Those are my nine lessons learned. If

19:06

you learned anything at all, leave a

19:08

like down below, subscribe, and turn on

19:10

notifications. All I do is make amazing

19:12

videos about AI. Let me know down below,

19:14

what do you want to see next from me? Do

19:16

you want to see more Claude code videos?

19:17

Maybe a Fable 5 video or more tips on

19:20

Hermes agent? Let me know down below.

19:22

Hope this was helpful. See you in the

19:23

next video.

Interactive Summary

This video outlines nine essential lessons learned by the creator after spending hundreds of hours utilizing the Hermes Agent. The tips cover choosing the right AI model for performance, the importance of maintaining multiple agents for failovers, debunking security myths, optimizing platforms, and using advanced features like Tailscale, cron job management, and reverse prompting to maximize productivity.

Suggested questions

4 ready-made prompts