Conectei o HERMES com o JEV: 167 Execuções por $0.006

Author: Julio | Lucrar com IA · Watch on YouTube

Video authored by the channel shown, not by IANIA; all rights belong to the creator. Transcript and summary generated automatically; the original-language content is preserved.

Original language: English

Original video: Conectei o HERMES com o JEV: 167 Execuções por $0.006 · Channel: Julio | Lucrar com IA · 15:36 · Watch on YouTube

Topics: AI agents, Task and workflow automation, AI infrastructure and hardware

Summary

The video demonstrates how to connect the Hermes agent to the JEV model by installing a TypeSafe skill and configuring API access, including an OpenRouter API key option. It then tests the setup by having JEV triage simulated emails, classifying urgency, relevance, and suspected scams at a reported low cost. The creator also shows a second simulated customer-service interface where incoming messages are routed to departments. The video concludes that this approach could help organize high-volume support workflows efficiently.

Key points

Hermes should already be installed and running before connecting it to JEV; the creator discusses using a separate computer or a VPS.
The video recommends installing the TypeSafe skill through a prompt in Hermes, with a modified setup intended to make it available across sessions.
JEV needs API access; the creator’s existing OpenRouter setup already supports it, while a separate API key can be created and configured.
The TypeSafe portal was not accepting new account sign-ups at the time of recording, so the API-key demonstration uses OpenRouter.
The creator configures a dedicated API key by saving it in an environment file, uploading the file to the VPS, and applying appropriate permissions.
In the email simulation, JEV triages 12 messages: two for an immediate reply, three for a reply that day, two for review, and five identified as likely scams.
The creator reports a triage cost of 0.004 and notes that the OpenRouter interface displays costs with limited decimal precision.
The email examples include confidence scores, urgency, scam likelihood, and a decision about whether and when to reply.
A second simulation demonstrates routing incoming WhatsApp or email messages to relevant departments in a customer-service workflow.
The creator presents low cost and fast processing as potential advantages for businesses handling high volumes of support messages.

Chapters

Start	Chapter	Summary
00:01	Introduction and Hermes setup options	The creator introduces the Hermes–JEV integration and outlines using an existing Hermes installation on a computer or VPS.
03:08	Installing the JEV skill in Hermes	The video shows how to access Hermes and install the TypeSafe skill using a prompt adapted for Hermes.
05:11	Configuring API access	The creator explains existing OpenRouter access and demonstrates creating, saving, and configuring a dedicated API key on the VPS.
08:52	Testing email triage	JEV classifies 12 simulated emails by priority and scam likelihood, and the creator checks the results and reported cost.
11:56	Reviewing classification details	The creator inspects email bodies and JEV’s confidence, urgency, and reply-decision details for specific messages.
13:32	Customer-service routing and closing	A second simulation demonstrates routing incoming support messages to departments before the creator closes with links and a call to action.

Quotes

how to configure it with the Hermes agent and how to link them. — 00:01

How will Hermes trigger the JV if it doesn’t have this API key? — 05:11

Total emails triaged: 12. — 09:52

Code decision to respond now, that is, the model’s decision to respond now. — 13:32

as soon as they arrive, the JV, in a matter of milliseconds — 14:03

Concepts and entities

Hermes, Jev, OpenRouter, Hostinger, VPS, TypeSafe Skill, API key, triagem de e-mails

Full transcript (original language)

00:01 Since JV was released, the newest type and model of AI, everyone on the internet is talking about it. In this video, I will show you something no one is talking about: how to configure it with the Hermes agent and how to link them. Beyond showing you this step-by-step, we will also explore a practical example together, using Hermes and Jeev working together to handle triage in a customer service scenario. So, get your coffee and water ready, and let’s head over to the computer screen. Let’s begin our configuration. And before that, I want

00:31 to say that the entire step-by-step process of what we do will be in a PDF for you to download for free. The PDF will be in our community “Lucrar com IA “. The link to access it is here in the description, so you don’t need to take notes, just pay attention to the video. The first step of this process is to have your Hermes agent configured, installed, and running, just like I have mine. If you don’t have the Hermes agent configured yet, here are two options for you. For those of you who have a separate machine or a spare PC, install Hermes on that other PC, separate from your daily-use one, for security reasons, right? And you will

01:03 have the link to the video in the card here for you to see how to perform the installation directly through the app. Super easy. The second option is for those like me, who don’t have a spare machine lying around. Your best option in this case will be to hire a VPS. For those who don’t know, a VPS is a virtual computer, okay? And internally here at the agency and for our clients, we use the Hostinger VPS. They even have an option now to install Hermes with just one click. I will leave the link here in the description as well so you can go straight to that page,

01:34 because by doing it this way, you help our channel and also get the entire setup done for you. You won’t need to do any extra configuration at all. Click the link in the description; I will also leave it in the pinned comment. You land on this page, scroll down here, and choose the power of the virtual machine you want for your VPS. We usually use this second option here. Click on the one that makes the most sense for you. And here on this screen, there are a few things you need to choose. The first thing is the duration of your plan. The longer the period,

02:06 the more of a discount you will get. I suggest selecting at least 12 months here so you can get a good discount, a good cost-benefit. Scrolling down here, you don’t need to select any of these options unless you want to. This option here is important, okay? It’s the server that will host your virtual machine, your VPS. Choose the one closest to you. In my case, it already selected the United States because I live in Canada. That would be the best option for me. Maybe you want to choose Brazil here, because perhaps you are watching from Brazil. Another thing you

02:37 need to do after selecting what I showed you is to come here. Do you have a discount coupon? Yes, we do. We managed to get a discount coupon for you, which is the coupon lucrar com IA. All together just like it is here. Click on apply. You will see that in addition to the discount Hostinger gives everyone, there will be an extra 10%off the final price. Once that’s done, click continue, create your account, and proceed with the installation process, clicking next as the system guides you. Installation process finished. You will land on a

03:08 screen similar to this one. Click on manage for the VPS you just purchased. And here in this section, you click on web console. You will land on a terminal screen like this. Here you can simply type Hermes. And that’s it, you already have a Hermes agent available to use, running nicely here in the terminal. All the configuration we are going to do now with the JV, I will do directly through the desktop application. But you can do the same thing here in the terminal as well. If you are interested in seeing how to link and connect the Hermes on the VPS

03:39 to the desktop application running on your machine, I’ve already made a video here on the channel, I’ll link it. It’s a video where I show several features of the desktop app and at the end I explain how to make that connection. The next step is for you to come here to the Typesafe documentation page. Behind the Jeevar model is this prompt here. I will also leave the direct link in the PDF so you can land right on this page. It usually opens here, in the cloud code section, but you can install it using this command here by

04:09 opening the terminal and running it. However, there is a note: this command installs it in the project, installing the skill for the project you are working on. If you want to install it globally, just add dash G, as they mention. The easiest way really is to come here to this tab, copy this prompt , which is like a universal prompt, go back to your Hermes and paste it. What this prompt tells it to do is basically this: install the Type Safe Skill. If you are on Cloud Code, run this command and then that command. If you are on

04:41 another agent, run this command here and select your agent. Use only one installation method. You can read about the skill directly here, look. It already has the address for their GitHub repository there. I made some changes to the prompt to make it more optimized for installation in Hermes and also, instead of installing locally per project, it installs directly so that Hermes can use it in any session I open. To install for other agents. Then you really need to use, if it is Cloud Code in this case, you need to use the

05:11 path they provide in the documentation. Let’s send this here so that Hermes performs the installation. We will wait for it to confirm and come back with the practical demonstration. Skill installation is complete, and now there is one thing I almost forgot that is very important, right? The API key to be able to access the JV. How will Hermes trigger the JV if it doesn’t have this API key? There are several ways to configure it. In my case, I don’t need to perform any extra configuration because if you notice here, at this moment, I am using Qwen 2.5 Flash with the provider here, see,

05:42 if I scroll, OpenRouter. This is the provider I am using. Sometimes I use it with one of these others, sometimes I use it directly from my ChatGPT account . There are also several ways to have multiple providers configured here. Inside OpenRouter, the JV is already available. Since I already have everything configured to use via OpenRouter, I don’t need to enter any API key. If this is your case, congratulations. Now, if you want an extra API key generated from within, perhaps from the Type Safe AI portal, which is Jeff’s company, or if you want to generate an extra API key inside OpenRouter to account for or perhaps

06:13 track the spending on the Jeff model, I will show you how to do that now. Here through the TypeSafe portal, at this moment that I am recording this video, it is not available. So, if you want to create your account and generate your API key directly on their platform, it won’t be possible, at least at the moment I’m recording this video; they’ve paused new account sign-ups because demand has become too high. So, we will do everything by simulating it through the opener. I have already generated a separate API key for us to perform this configuration and for me to show you. First thing, click here: get API key. If you don’t have an

06:45 account, you will need to create one. And once you create your account, you will go here to the API Keys menu and create a new API key by clicking on this menu. Fill in the information and click create. Save this API key when it is generated, because once you close this box, you won’t be able to access the API key again. See? It’s all showing here, just fragmented. After you generate the dedicated API key for JV, the next step is to save it in the environment variables file on the same machine where your Hermes is running. So, remember that in my case, the

07:16 desktop application runs here on my Mac , while Hermes itself runs on the VPS. The first thing is to create the file where I will save this dedicated API key on my Mac and then upload it to the VPS. I will save this file in the downloads folder, okay? So I’ll type here: cd, then the tilde symbol slash downloads to change to the downloads folder. Now yes, I am in the downloads folder. Now we will create the file where we will save this API key here on the Mac first, to then upload it to the VPS. In my case, I will use this

07:48 filename: typesfe_line ke ptxt. We press enter, and it opens here so that we can save both the parameter and the API key. Done, I’ve already typed it here: type save_line API_line K equals. And then paste the API key. Once that’s done, we just type control O, press enter, and then type control X. It’s been saved and the file with your API key has been created on your computer, in this case on mine, right? Now that the file with the API key has been created, we will upload it to the VPS

08:19 with this command. It will ask for the password; you provide the password and then this line will appear here with the filename and 100%indicating that it has been uploaded and saved on the VPS. After that, you will run this security command. It will ask for your password, you enter it, hit enter, and that’s it. The file has been saved. The API key will now be available with the correct permissions. The last step is to go back to Hermes and inform him so he can finish the configuration. And here it is, he has already informed me that he finished the whole setup and did what was needed. He probably had to create a

08:52 pattern for the calls every time he activates JV. And he also restarted the backend here so the configuration can be applied. Now let’s move on to our practical test. I asked Hermes himself to create an email inbox simulation here for us. The one who will triage this inbox, all these fake emails you see here, is Jev. I asked him to connect using the specific API key we created and configured in this example, okay? It was used 7 minutes ago by him

09:22 just to run a test. The cost here was 0.001 cents, that is, fractions of a cent. And now, supposedly, if everything is working, we will click here to run the triage and we will see JV classifying these 12 emails that are currently mixed up here. What he will do is reorder this list by priority and show the decision badges. Reply now, reply today, what is a scam/spam. So, Jeff himself is the one who will do this classification for us. Let’s go. I

09:52 will click here. Run triage with Jeff. We are seeing, wow, super fast. Here is the cost. Total emails. You can’t even blink. Blink and you’ll miss it. Total emails triaged: 12. Two reply now, three reply today, two review, and five that look like a scam. The total cost was 0.004. Let’s go back and see if that’s correct. I’ll refresh here so we can see it. Alright, it updated. It says here that it was used just now. It

10:23 means it was used now, but the cost the interface showed us didn’t appear. It stays here at 0.001. We had less than 0.001, right? So let’s go back and check what was classified. Ah, guys, it’s correct, I’m the one who got confused, okay? The cost was 0.004, meaning there are three zeros here before the four. And here, look, we only have two, so it makes sense. That’s exactly it. It’s because OpenRouter here doesn’t categorize, it doesn’t go beyond three decimal places. After the period, we have the first

10:55 emails from Stripe and Carlos Mendes. The lead flow stalled again. Ouch, what a pain for the GR bot. Grow bots here. Ah, YouTube subscribers. Alright. Spam. Wonderful. It classified it correctly. These two are important, they really need an immediate response. These three here, spam. Bank security alert. Notice , your account is locked. This is a scam. Fernanda Souza, a unique opportunity to invest. Also a scam. So come here to reply today. Dr. Marcos

11:26 Vieira, YouTube referral. Automation for clinic scheduling, module 3 of Ana Beatriz’s community here. Rodrigo Tanaca, Thursday Coffee, partnership for small and medium businesses. Cool. And these last ones here, read when you can. Empreenda Podcast, invite to be on a podcast, barbecue on Saturday. This one is a priority, right? Just kidding. And we have these others here saying they can be archived. Cool, cool, it worked well. We actually had Jeff classify all of this for us in a real example from the agent’s own email

11:56 inbox, okay? Or yours too, I don’t know , in this case I think we can put it in a real inbox, okay? Because there’s nothing being removed, created, received, or sent. It’s just classification. So I think we can apply it to a real Gmail account. By clicking here on the fake email, we can expand what the email body is, right? In this case, the message here says we couldn’t collect the payment for your monthly subscription, please update your payment method within 48 hours to avoid service interruption. These were the

12:29 options Jeev had for financial classification during triage: financial , scam, lead, client, community, personal, interpersonal; it gave 99% confidence that it is financial and 0.1 %that it could be a scam. In this case, it’s correct because the domain came from@stripe.com. In our simulation scenario. Here is a more detailed breakdown of its judgment. Urgency 0.7, meaning it is urgent. We also have here “worth replying” 0.77, “suspected scam”

13:00 0.29. Decision: reply now, top of the queue. In this other email example from Carlos Mendes: Júlio, the lead qualification system you implemented stopped responding since last night. I have 200 new leads in the funnel. My sales team is at a standstill. This is a headache. This is costing me real sales. I need an update today. Ah, in the morning, please. I think there’s a positioning, right? Typing error. Here in the test. We have the customer classification with 100%certainty. Here we also have urgency at maximum,

13:32 meaning super urgent. Worth responding, it’s at the maximum. Suspected scam 0.05. Worth responding almost at maximum, right? 93%here. Code decision to respond now, that is, the model’s decision to respond now. Top of the queue, urgency 2.0, ongoing financial loss. Cool. Let’s go to another example . In this second example, I asked Hermes to create a simulation of a service system interface for triage. So imagine you are a large company dealing with many messages arriving via WhatsApp and email. You want to centralize and organize all of this in one place. Here is a perfect example

14:03 where JV would apply with great efficiency and at a very low cost. We are seeing simulated messages arriving in this column, and as soon as they arrive, the JV, in a matter of milliseconds and spending, look, a tiny fraction of a dollar, it can manage cents of a dollar, right? It can perform the triage, sending the message according to its content to each of the departments. To build this type of solution using conventional LLM models, the cost to evaluate these messages would be very high, and the evaluation

14:35 speed would also be much slower, which for a company dealing with a high volume of support, might not be the best option or even feasible. So, in these scenarios and others that you can even mention here, in case you are interested in me bringing a demonstration to the channel, we can apply JV to have much more efficiency and build better systems for clients. Also, if you are interested in learning how to do this here, creating automation systems for companies, the link for you to join our private area for profiting with—actually, not to

15:06 join, but to sign up for the interest list, because we haven’t launched this private area yet. The link will be here in the description if you are interested in learning to profit with AI. In the first link of the description and also in the pinned comment, I will leave the link for you to access the Hostinger page, contract your VPS, and start using the Hermes agent to create things that make your life and business easier. Did you like the content? Leave your like, subscribe to our channel so you don’t miss the next videos we are bringing and will bring here. And watch this other video

15:36 that I’ll leave in the end card. In case you want to understand more, right , about Jeff, what this model is, why it’s different, why it’s so efficient, click here, watch it, and I’ll see you there. See you later. Oh.