Agente Hermes: NOVIDADES que deixaram o App 10x MELHOR!

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: Agente Hermes: NOVIDADES que deixaram o App 10x MELHOR! · Channel: Julio | Lucrar com IA · 11:24 · Watch on YouTube

Topics: AI agents, Enterprise AI applications, Task and workflow automation, Local and open-source models, Multimodal models

Summary

The video walks through recent Hermes Agent and desktop app changes, led by a live bot screen that lets users watch an agent browse and take over its controls. It also demonstrates simpler desktop layouts and a connectors menu with many services available to install. The presenter explains the app’s one-click local model setup and shows a voice feature that responds to the wake phrase “Hey Hermes.”

Key points

The new bot screen feature displays an agent’s browser and terminal activity live.
Users can take control of the agent, then hand control back so it can continue its task.
The desktop app now offers a simple mode with two cleaner layout options.
The capabilities area includes a connectors menu with services such as Airtable, Asana, Canva, ClickUp, and Notion.
Hermes can read a computer’s hardware and select, download, and configure a suitable local model.
Users can browse for local models or add a model file manually in provider settings.
The microphone feature supports the “Hey Hermes” wake phrase and can keep listening for follow-up conversation.
The presenter notes that voice transcription may not always handle Portuguese correctly.

Chapters

Start	Chapter	Summary
00:00	Live bot screen and agent control	The presenter demonstrates watching an agent browse in real time, taking over its screen, and returning control so it can finish its task.
04:39	Simple desktop layouts	The video shows how to switch between cleaner interface layouts and arrange the sidebar, chat, terminal, files, and preview.
06:42	Connectors menu	The presenter explores a menu of pre-ready integrations and lists examples of services that can be installed.
08:15	Running local models	The video explains the app’s guided local model setup and shows where to browse for models or add a downloaded file.
09:49	Voice and the Hey Hermes wake phrase	The presenter tests voice input with “Hey Hermes,” describing how the app listens, transcribes, and responds.

Quotes

the ability to watch the agents working. — 00:31

everything turns red, see? — 03:06

a simple mode for your interface now appears. — 05:10

with just one click. — 08:15

Hey Hermes, search for me when the release date for the new Avengers movie will be. — 10:21

Concepts and entities

Hermes, Nous Research, botscreen, LM Studio, Airtable, Asana, Notion, Supabase

Full transcript (original language)

00:00 Since you really liked the last video I brought about the Hermes desktop app updates, guess what we’re going to do in this video? I’m bringing the latest updates again, because they won’t stop making improvements to both the Herm agent and the app. Let’s quickly explore the main changes that have happened in these last few days, in this last week. And one of them, in fact, will completely change the way you handle and work with your subagents inside Hermes. Starting here with the

00:31 first update, which in my opinion is the biggest and best of them all, so hold onto your seat. On September 23rd, that is, two days ago, the folks at NOS Research released this new update, which is the ability to watch the agents working. I had already shown here on the channel an option for you to view it through the box near the composer, where you type messages and prompts, to see what the subagents are doing and make course corrections or even stop their work. I’ll leave that

01:02 video linked in the end card, but that function is limited compared to this one we’re about to see, which is literally watching the agent opening tabs, doing searches, and executing commands in its terminal. Let’s go to the app to access this new function called botscreen. Here in the Hermes app, obviously you need to update, right, both the application and your Herms to be able to access what I’m going to show you now. And here in the app you go to the bots tab. And here in bots you choose any one of them.

01:34 Right-click and click here, on open screen. When you click for the first time, it might ask you to install a specific package. You authorize it and go on with your life. This is the screen, right? It’s the screen of our bot here, Jarvis. We are now going to request a basic prompt for it to do a search and open our channel, just so you can see this working in practice. The prompt I will request is the following: search for Tesla’s Team G3 and then open my YouTube channel ’ Lucrar com faz tudo’ using the new bot

02:04 screen feature because I want to watch you working. Let me know when you’re finished. We’ll fire off this prompt and here on the side it already appears , look, it already opens his little screen in live mode for us to be able to follow along. Let’s look here to verify if it will actually open. And after it opens the tab here, we’ll go to the expanded view. It opened. Hey there. Let’s take a look here. It’s already searching on Google. Oh, access denied. Anyway, let it search. Here it is, look. It opened the Teslaat.com

02:35 site. So, we’re doing the search we requested of it. We can follow along on this larger screen, or if I go back, I can also view it on this smaller screen here on the side. We can even come here , look, and perform a takeover, meaning you can take control if you notice it doing something a bit strange, right? It just opened another site searching for Optimus Gen 3. Now opening another one, Humanoid Robotics. Let me see if I can perform a takeover, just to test it

03:06 . It continues doing its research. If I click takeover now, everything turns red, see? And I can click here, look, and delete things and such. Anyway, let me just hand it back to see what happens. Whoops, let me take over and close this, since I ended up leaving it on an ad. Now, let’s see if it continues after I paused and interrupted it here; look, it identified the control and handed it back to the agent. Cool. Awesome, awesome. So it executes a command to

03:38 sleep for 20 seconds, and then it keeps checking if you’ve returned control to it or not. It’s opening our channel now . Nice. And if it verifies that you’ve returned control to it, it proceeds. Top notch. So it managed to open everything. We were able to execute this successfully. I had to run this about three times, guys, before it understood what I was saying, because the desktop app already has an internal browser preview where it opens up for

04:08 you to watch and track what it’s doing. So it kept getting a bit confused; it didn’t understand right away what I meant by the new bot screen, that new feature. Until I came back here and provided the link to this pull request from the NS Research team, which explains this whole new feature and how it works. And then it understood and created a skill based on that. Every time I mention bot screen in my prompt now, it uses specifically this instead

04:39 of opening the screen here, the browser preview, or using another skill or navigating behind the scenes without showing anything here. The second update is this one right here. It’s the simple mode for the desktop app, where you get a cleaner interface, without so many tabs for those who are more technical or work with development. Let’s go back to the app and see how it works. Back in the app, to enable this

05:10 simpler and cleaner option, just click this button here, layout editor. And when you click, a simple mode for your interface now appears. I was using advanced mode, but actually the basic version of advanced. Basically, that’s what I was using. And with this setup, I have the sidebar here, the chat in the middle, and the preview. When it opens the browser, it will open it right here. And I have the chat in this area and the terminal tucked away down

05:40 here. In the default option, it removes the terminal from here and moves it over there, which I think is much better, actually. I was using the basic one, but that’s fine. Terminal focus comes down here, and we have files here and the preview here. Nice. Terminal deck. The terminal stays down here. The files are here, sidebar and the chat. Lastly, quad, which is also part of advanced, we have the preview down here , terminal, chat, and the sidebar. In simple mode, there are only these two

06:11 options, sidebar left, meaning the sidebar here with our session, and the chat. The other option is with the sidebar on this side. This is the cleanest option if you want to use it. I’ll go back to advanced and maybe keep the default option here, which wasn’t what I was using. Let me hit done. And when we confirm with done, it looks like this, with the terminal appearing down here. If I have any files or am working on a project, it will appear here. And the preview will also be in

06:42 this corner. It obviously expands or retracts according to use. For the next update, we’ll check it out together directly in the app, because to be honest, I haven’t seen them say anything about what I’m going to show now. Maybe because it went unnoticed, maybe because it’s not that relevant. It could also be that I didn’t pay enough attention to this tab I’m going to show you now. But anyway, let’s check it out together. Coming over here to capabilities, we didn’t have this connectors menu. What was here was mp, which, well, is practically the same

07:13 thing, right? Only the name changed, but I don’t remember having this here that you’re seeing now. Several connectors are already kind of pre-ready there, just for you to install. We have Airtable, we have Asana here, a bunch of options. Canva, ClickUp, oh, Confluence Cloud here from Confluence, I think that’s it. Cloudinary, Figma, Dropbox, Grafana, GitLab, Hugging Face, Indeed. Monday.com, the tool as well, right?

07:43 Like ClickUp for team management there, n8n, Notion, Unreal Engine, Vercel. Man , there’s a bunch of things here, WordPress, which I don’t know if that’s Supabase, look here, Stripe, Supabase, Square. Man, I don’t know if I hadn’t paid attention to this before, but I don’t remember having all these options kind of pre-ready there for you to connect and install. Comment below if you had already noticed this tab, if it was actually like this, or if I’m imagining things, but I honestly don’t remember. This is new to me. The fourth

08:15 update, which isn’t that recent, but I didn’t cover it here on the channel for you, so I’ll talk about it anyway because I don’t run local models at the moment; that might change, but right now I’m not running any local models. We now have the function inside the Hermes desktop application for it to do the whole local model setup for you with just one click. How does this work ? It reads all your hardware, understands your setup, chooses the best model for your machine, then downloads and configures everything for

08:46 you. Similar to what you would have in LM Studio and other management, installation, and usage applications for local models. Here in the Hermes application, you go to settings, click right here on providers. And inside providers, you will have the option run models locally, meaning, run the model locally here on your machine. It’s already giving me the available options here, see, of what I would have and all of them are too large for my machine.

09:18 We would have the option to browse here , meaning, search for models, and we could also, by scrolling down, add a file, right? You download the model and add it here for you to use. When we click, see, on browse models or navigate, right, to find models, we have it here scrolling down with this option for us to search for a model. I searched here for Llama, which is a new model that came out, it’s not an LLM, it’s a rival to Jeff, and Llama theoretically would work here for my computer. The next one I want to share

09:49 with you isn’t exactly an update, but a feature that’s kind of hidden here inside the Hermes app, which you might know, or you might have never heard of. You know when you say that little word to call your Siri and she starts listening to you and does it? Isn’t it what you asked her for? Or at least tries to. You can do the same thing here with the Hermes app. The function is hidden here, look, at this little microphone, where we have the speaker icon, which is for Hermes to respond out loud, that is, read the answers to you. And here we have an ear icon,

10:21 which is precisely for it to be able to hear, to pick up when you say the wake word, right, which would be Hey Hermes. Let’s do a test here to see if it works . Hey Hermes, search for me when the release date for the new Avengers movie will be. Avengers. Cool. So it stays there running. You saw that I pressed

10:53 end, but it stays kind of with the microphone open waiting for you to speak, because you can have this conversation here with it. It managed to understand. Sometimes when I speak Portuguese, it doesn’t transcribe correctly. I don’t know which model is being used to do this, if it’s DeepSeek itself or if there’s some other model behind the scenes. But this time and a few others it actually managed to understand what I wanted in Portuguese. It transcribes to English, performs the search, and returns the result to you. That’s it for this video, everyone. Tell me in the comments which of these

11:24 new updates or features I showed will make the most difference for you in your daily routine using the Hermes agent, right? Leave a like as well if you enjoyed the content. Subscribe to our channel to keep receiving notifications when we upload new videos and also to help us out, giving us a boost to keep producing content for you . And I’ll leave linked here in the end card another updates video that I mentioned at the very beginning of this video for you to watch in case you missed it. So click here and I’ll see you there. See you later.