Hermes Desktop + Ollama: Run a Self-Improving AI Agent on Your Own Server

Añadido: 2026-06-04

[00:00:01]Hermes agent desktop is here. Let's get it installed in this video and we are also going to integrate it with local Olama based model. This is my Ubuntu system. I already have Ola installed with this quen 3.6 model and I also have Hermes installed with this older version. If you don't have it installed, all you need to do is to run this single oneliner command. If you are having issues, just search my channel with Hermes agent and there are heaps of videos which you can use in order to get it installed. Once it is installed, all you need to do is to run this Hermes update. If this is the second time or third time you're installing it and it is going to bring it up to date which is required and you can just simply press Y here. If you want to retain your changes if not that is fine. I will also show you how you can integrate it with local based models. This is Fad Pza and I welcome you to the channel. Please also follow me on X if you're looking for AI updates.

[00:01:08]And you can also ignore these warnings and then it is going to take bit of a time 1 to 2 minute in order to update everything all the NodeJS packages and stuff.

[00:01:19]Remegent is updated. Let me quickly show you the latest version. So you can see that everything is up to date. And now you can simply launch that desktop by running this per desktop command. You can also use the same steps on Windows or Mac OS. Let me launch the desktop.

[00:01:40]First time it just installs all these dependencies.

[00:01:51]And now it is installing all the packages. And now it is launching our Hermes agent desktop first time. The back end is being configured and installed. And there you go. Everything is ready. I will just increase the screen size. And from here you can also set your appearances and lot of other stuff. You can see that on the top right there's a settings button. I have clicked there. Let's go to my appearance. I would really go with a dark theme here. So this is bit darker.

[00:02:22]And then you can also select a theme.

[00:02:25]For instance, you can go with midnight or you can go with amber which is quite interesting. Cyberpunk is also quite interesting. You know, normally this is the default one uh which I have seen with Hermes. And then there are lot of other things which you can configure.

[00:02:44]For example, if you want to go with API based models, you can go with lot of API um based LLM providers like Gemini, GLM, Kim and the list goes on and on and we have covered all of these. You can also go with your own um custombased one and you can see that it already has detected that my minimax API key is already there. Another cool thing is that you can also connect it to other open-source tools like open code and few others uh hosted ones and of course the local providers are always there and you see that the list goes on and on. You can also connect lot of MCP servers. You can also configure your own the model context protocol. I mean where you can use lot of tools which are exposed from various external providers or you can configure your own.

[00:03:38]You can also configure chat and other things. But if you go to the chat here you can see that you can uh set the personality here. Let me see if I can just make this bit bigger.

[00:03:51]I couldn't find any way of increasing the font. Maybe this is for them in the to-do list. Anyway, you can have multiple models here. For example, you can also select uh any other model. You can see that our Olama based model is there. So, and it's also selected and this also brings up your own local model or you can just simply go with API based or whatever model is configured. I was also running a local model earlier with VLLM and then you can simply apply there. I'm just going to cancel it from here. And this is what the interface at the main page looks like. And I'm just going to send some follow-up. I'll just say, "Hey, Smile, you're on camera. I'm recording this for a YouTube video." And let's wait for it. This is a new session which is untitled at the moment. I will let it run so that you can also see what this experience looks like in terms of the response. First time it takes time of course it loads the model uh in the GPU and it has come back with the response which looks pretty good by the way and this is um very well formatted. I would say all the emojis are there. Now if you can just go up you can see that this is a title. You can of course rename it.

[00:05:08]You can just call it anything you like like post chat or whatever. This is my session. You can pin this session here.

[00:05:17]Now you can see that it is pinned. You can remove it of course and then you can even export it in your own format. You can save it locally and then I'm just going to also show you how you can do the copy ID. You can simply archive it and then you can unarchive it. You can also delete it. This is fairly straightforward. The great thing about which I feel is that you can very easily configure the skills and tools and you can see that lot of skills are already there for most of the agentic uh coders there including clot code open code. I'm very curious to see if open claw is there or not because they really don't get along that well as far as I can tell from the social media wars anyway. But you can see that also there are some MCP clients are available which we have been covering a lot on the channel. You can simply uh for example this MCP native one you can toggle on or toggle off. You can also use it. Sorry. There are also various um skills some you know which you can readily use or you can create your own. You can also um filter them out at the top with you know for example if you click on general these are the general ones creative data science and I'm more than sure this list is going to uh grow a lot. These are the messaging channels like you know you can configure the telegram uh discord. We have done lot of videos where we have integrated these but this of course makes it simple instead of remembering and using those CLI commands which I prefer personally but not everyone. So you can just simply put your bot token or your user ID here.

[00:06:59]same experience and then don't forget to save it and you can use the same and whatever artifact it creates like you know um you know that I normally create those HTML file and stuff self-contained all of these get stored in one place I believe uh it also works on projects so for example if you create a new session you can also just I believe you can create a new persona and new project and new files so for example at the top right you These are all the profiles or personas which you can define. And if you just click on settings here, you can define either here or even during the chat. While we are here, you can also not only go with the text models but also these are lot of uh voice models and other um stuff. It also comes with its own memory or you can configure some external memories. We already have covered this agent memory in other video if you're interested. And you can specify lot of other parameters like compression but most of these work good out of the box. Advanced parameters allow you to set lot of hyperparameters or some of the inference parameters. You can also select refresh rates and retries and all that exponential backoff if you like. a very very new tool where you can not only use your local gateway which I'm using or they also offer I think some sort of hosted cloud-based gateway just for scalability and all that stuff but I won't be uh using that cloud-based for us local is more than enough so this is pretty much it where you can um have your desktop you don't have to have put a new GUI on top of it because there have been some open source tools which you can put on top of it but I think that this experience is quite better. The only thing which I can uh suggest to them is to just you know allow us to just make a font bit better.

[00:08:55]But it's not bad either. That's it. Let me know what do you think about this new agent. Please follow me on X and if you want to help out the channel, please become a member. Thank you for all the support.