Hermes Agent 0.19 Quicksilver: més velocitat, tasques en segon pla i aprovacions intel·ligents

All right, folks. So, Hermes just released their newest release of Agent 0 V0.19, and it's come a long way. And what I want to do in this video is give you a quick recap of everything that's been updated. But instead of just reading the headlines and seeing what's good, I also want to just key in like kind of the recent updates overall and how Hermes has improved over time and what we are using to get the max out of Hermes. So, first and foremost, timing's never been better for AI. We're past that phase where like we're trying to experiment with Hermes and it kind of it fails some at some stuff. We're at a point where Hermes is becoming the de facto agent like kind of like your assistant for life. And honestly, I think I want to talk about something that's outside the release, which is the newest models uh either from GP, regardless of where it's from, either OpenAI's newest GPT 5.6 or KK3, those are my two preferred models. And because these models are a lot more capable in agentic work, agents are actually becoming super useful. So for managing your calendar, resolving conflicts, helping you book stuff, buying something, these agents are getting

really, really powerful for those use cases. So just right off the bat as well, in terms of the updates and what they've been doing recently, Hermes has been improving themselves by a lot. I mean, honestly, they've received a lot more funding, right? They completed more series of rounds, raised more than $50 million to make this assistant good. So they better deliver on some good results. the TLDD or the headline feature for Quicksilver is that it's running a lot faster. So they time the 80% depending on how you use it. Uh and the reason why that this happens is because of the way they select models and the pipeline in the background. So I feel like before what was happening with these agents where they were trying to use LOMs that were not designed for agentic workflow. They were kind of cramming in and just jerryrigging everything together and patching everything and hoping praying to God that it works. But with recent updates all the way here in version I would say I'll call this 19 cuz 0.19 uh is a mouthful. But honestly with these upgrades and these uh speed increments and with better AI models, you're going to get something that actually works really well. If you

guys want to update, I mean it's a no-brainer. Um the command is Hermes update. So just type that into Hermes. You'll update it. The way that we run it here is that we run these on a VPS. Um, we don't really run these on the desktop versions or whatnot because I do believe that a 24-hour agent is the most powerful regardless of like, you know, you don't have to use it on your computer, you can just use it on chat. That's what makes this so powerful, right? So, having the speed improvement is really good. If you guys are using password as well, one password support is a lot better. And honestly, passwords were just such a big pain in the past because of just the permissions. So, they kind of fixed two things to fix this password issue. one is um not requiring the API key, but also of course they have a smart filtering system that they released like here now you can just like uh I wanted to show you guys the video but of this but the smart approvals just mean that you can save your passwords and do what you need to do. Yet again, these agents really are supposed to help you with

those knick-knack tasks like whether you're browsing a um website that you go on to all the time, fill in something for you, they should help you with everything. And obviously, having your passwords and not asking you for approval each time is very much needed. Also, of course, there's also um speed improvement in the background as well, not just with the code boot, but also um across desktop and across multiple platforms. there's been a lot of it's essentially they made like 2,000 different commits to improve that speed um for everything that you do. So, I think that's the huge update that we see. I I do want to deliver on some concrete uses and I think the next one that's going to be really important for you guys is background work. This has always been a very big struggle and I do see that it's not going to be fully fixed here. I'm always going to hold my hand because like we've done so many upgrade videos, update videos where you know they're supposed to have fixed background work. So far what we've done and so far that's I found very useful are cron jobs that run in the background for your agent to do every morning. So

for example in the morning kermes delivers our news updates. It delivers what our competitors are doing. It delivers this basically summary of what's happening in the AI space. These are background tasks. But previously they failed a there's obviously times when they fail and they don't run. But now they've actually improved how background tasks are work um being done and how you can basically multitask. There's also transcripts now. So if you got if you got really want to go through the nitty-gritty of it, you can actually see the transcripts of what's happening in these background tasks and what these sub agents are doing. So you have a better understanding of exactly what is being done by your agent and the orchestrator that's happening uh with you. I mentioned this a little bit earlier, but obviously the smart approval is really good. That doesn't affect me too much. Um, I almost always use yolo mode. I never give my agent too much that I can't do. So, I always just say, "Hey, just yolo it." I I'm not going to spend, you know, 5 minutes, 10 minutes approving and not approving and figuring out what's happening with you. So, smart approval is much appreciated. But also, of course, you

can also have yolo mode that can just go and straightforward and do your tasks honestly. Um, long term. Okay. Okay. So, they they're really pushing their subscription management service. I I understand why because they need to make money to kind of pay back the investors that uh funded them. But that being said, uh Newsportal is not too bad. I do say that overall I don't use News Portal that much. Um it's one of the first things I do turn off when I use Hermes. I'll just be honest with you guys. The reason why is because I don't want to use the models through them. I still want to use Kimmy independently of uh Hermes like I use the Kimmy model uh Kimmy K3 and I use codeex model. So if you actually look at uh let me just finish the update there. Uh replace let me just go there. But I tend to want to use the models um Hermes update and just pull that there. But I tend to want to use those models outside of Hermes. I don't only uniquely use Hermes. Which is why if I'm thinking of where to spend my money, well, I buy two. Well, not I buy pretty much buy all

of them, but my primary focus has always been uh for now on OpenAI um and on Kimmy. Kimmy for the cheap inference. And I know it's kind of interesting because um Hermes uh we just updated. I'll just show you guys here. But um Hermes can use multi model and you can choose multiple agents. Um, right now it's on Terra, but 100% you can use Kimmy on here. But what I kind of find is like I kind of stick with a certain model. I don't really need to use multimodels anymore. Um, the reason is because is it's so much cheaper to get AI now. Like every it just like we're in the golden age where every um AI provider is trying to compete for our attention and a $20 plan gets you really far. So I actually don't find myself burning through that much usage. So uh in terms of personal usage um I have two options. I have GPT 5.6 Terra and Kimmy K3. Kimmy K3 new new entry on the block but it's been performing really well on enginic flows and the rates of hallucinations have dropped dramatically which is why yet again Harmes is becoming a really really good model. Not even just not even because

of them but because of the whole AI landscape because everything's improving. Well, it improves with you. If you want to switch models, of course, you can do Hermes model and you can go through the picker. Uh, either open AI or you can do Kimmy Moonshot. Uh, let me just double check here. Here, yeah, you can get Kimmy K3 on here. It's already in built in into the plan. So, you can choose Kimmy and then run your Hermes on there. This will make your this well, you can probably fit everything within a $20 plan and you're going to get an assistant for $20. I I've never really thought of seeing this day because yet again uh we run a company and we do have uh assistants to help us with our work but more and more of that work is like even the scheduling is becoming uh a Kimmy task rather than an assistant task and it's saving a lot of our assistance work for them to work more on company filing company registration or whatever whatever like workload they have to do. So this actually took a lot of um uh stuff off the plates of our assistants so they can do something else. Um, and for

what? For $20. Come on. That's [sighs] We're in a crazy crazy uh landscape. And uh very drastically, we have actually improved our company uh scheduling handling handling over to the Kimmy side or not the Kimmy side to Hermes side. Um and that's helped us a lot in terms of both savings and uh work efficiency because yet again, these assistants work 24 hours a day. If you want to hire a 24-hour assistant, first of all, it's not possible. Second of all, if you want to call your assistant on a Sunday, it's not fun. Okay? You have to be please, please, please get this done because it's not their work hours, right? So, yet again, I do feel like this has been just a very drastic change into how uh companies are operated. So, we put all the updates as well onto our new so-learn website. It's in a weird place. It's in news and it's in the Hermes updates. If you want to read the full log on what we think about that, that's in the website. I'll put a link down below as well. But upgrading is just super easy. Hermes update and you update it and you can get um 0 Hermes you can get 0.19

or version 19 I'm going to call it here. So that's around it. Uh we're going to do more videos as well on how to maximize the use of your Hermes. So stay tuned for that. And I actually have a separate video that's coming out which is talking about when you should use Hermes and when you should use clot code. Uh this is it's not as clear-cut as you think because uh some agent work like say for example I just give you let's look at a good example here. uh we used to do uh this type of work where we analyze videos, update comments, read comments, update our links. This used to be all handled by a Hermes agent, but over time I actually built my own dashboard for this because I want deterministic results where I know exactly what's happening and I have a very good visual on what's happening. So I actually have a video coming up on when you should use Hermes versus when you should use CL code to build a custom solution for you. But with that guys, that's uh that's pretty much the wrap on the updates. Thank you guys so much for watching this video. See you guys in the next one.