The Hermes Agent V0.15 Velocity update represents a major architectural refactor that transformed the agent from a monolithic script into a modular system, reducing the core code from 16,000 to 3,800 lines across 14 focused modules. This restructuring improved startup speed from 700ms to 258ms, reduced function call overhead from 399,000 to 213,000 calls, and accelerated session search by 4,500x (from 90 seconds to 20 milliseconds). The update also introduced a Kanban Swarm orchestration system for parallel task execution, new integrations with Ntfy, Krea, and Bitwarden, and enhanced security hardening against prompt injection attacks.
Inmersión profunda
Prerrequisito
- No hay datos disponibles.
Próximos pasos
- No hay datos disponibles.
Inmersión profunda
Hermes Agent Update v0.15 is POWERFUL! (Velocity Release)Añadido:
All right, folks. If you're using Hermes Agent, this is a big update. V0.15, the velocity release. Everything in this patch touches on architecture, speed, task orchestration, integrations, and security all at once. And this is one of the things I like how Hermes agent are doing things. They would be rolling out these uh highlights one by one and then they would bundle them all together into one official release so that you know you don't get bombarded with new updates every single day. In fact, we've covered most of them already in the past week, past two weeks. So, we're going to be concluding everything you need to know if you've missed out on those uh updates here in this video. Also, another interesting to note is on the same day they they also released a same day hot fix. This is the patch release. So, we're going to look into this uh and cover everything you need to know for these two updates.
So, the easiest way to describe this release is Hermes feels less like one giant script that kept growing over time and more like a real functioning system.
And that's really what this update is trying to improve. Not just the performance of your Hermes agent, but also the consistency of your Hermes agent. And uh one of the biggest changes that contribute to that is their core refactor. So the old main loop file, the run agent Python file went from 16,000 lines down to 3,800 lines. We actually briefly covered this in the speed upgrade video 8 days ago where they shrunk it down from uh 16,000 to 8 8,000. So roughly by half, but now they've shrunk it down even more. So it's a lot faster. So it's a huge reduction already in a very short amount of time. The code here was redistributed across 14 focused modules. So from a user perspective, the behavior of your Hermes agent stays compatible, but the internals become much easier to maintain and extend. So that means whenever Hermes agent makes a new release, a new update in the future, uh uh when you run the update, your agent won't break. And actually, that was one of the biggest problems of Open Claw is whenever you make an update, some things would get fixed, but some things would almost always break when you make an update. Uh that's because they their their uh main loop file was a big mess. They actually addressed this as well in the latest update, the 5.26 update. So, if you're using Open Claw, you can watch that video, but we're not going to get into that. And the next win is the startup speed. So, we also covered this in the speed upgrade video. So, it's been about 8 days, but the Hermes team managed to shave off another second off the launch time. Even something as simple as Hermes double dash version dropped from 700 milliseconds to 258 milliseconds. Uh, that is a very big drop. And I like this metric because it reflects something we all can actually feel. It's definitely feel a lot faster, smoother, not just on startups, but also mid-con conversation with your agent when they do tool calls.
They they they're less likely now to get stuck on a particular task even without having to use slashgoal. It's just feels less probabilistic now, more deterministic when they can finish their task. Of course, the really complex ones would take time, but of uh but for the simple ones, they know that these are simple, so they don't have to take that long to finish those tasks. And speaking of long task, Hermes is spending less effort on its own internal machinery and more on the actual work you asked it to do. This is because of the function call overhead, having dropped from 399,000 calls to 213,000 calls. So, that's a big reduction in per turn framework overhead. uh and that matters a lot for longer conversations which means they do less work on the hot path and focus more on the actual work you give it. Now another useful update in this release is the session search. So before the session search would take I think around like 90 seconds which is long enough that most people will simply stop relying on it unless they absolutely have to. But with this update it was rebuilt without needing an auxiliary LLM. So now it runs in about 20 milliseconds. So that's that. Oh my god.
90 seconds to 20 milliseconds. That's actually 4,500 times faster. Uh which also removes the extra model cost. So it uh for for you it means uh if you're deep into a long Hermes session and you want to find something from earlier maybe a config choice a repo path uh a deployment note whatever those sort of searches they're less clunky anymore but you don't even have to go through uh that same method anymore. What you can do now is you can just uh instead of going to a log, what you can do is go to Hermes sessions list, take any ID of your previous conversation, resume it, and what you see now is you can see the the previous conversation uh in verbatim with your agent. So before if you resume an older session they would not show you the previous conversations but now you have the full uh conversation with your Hermes agent.
So, this is super super useful if you ever want to go back to a previous session that uh and you don't have to go to the logs anymore cuz the logs was a big pain cuz it doesn't just show your conversation, but it also shows their uh whatever they invoke, any tool call, any reasoning, bunch of uh numbers, hexadesimals, whatever. But if you just want the actual verbatim conversation, yeah, this is one of that. So, those two updates are pretty much like side by side. one for you, one for your agent.
So your agent can use session search to to check its previous verbatim conversation with you. But you know if you really want to be a bit anal about it, then you can check it yourself just by going back to a previous session and then read. Okay, moving on. Now uh if you're running Hermes on Android or mobile style setups, you also get a nice speed upgrade. So it's very nice to see they included that community there. Uh now this one, the cananban, this is one of my favorite features that have rolled out. Now they're making a big upgrade for it too. What they've done is they've added a Hermes canban swarm command. And what it does is it can build a graph with a root task and then you can have parallel workers, a verifier, and a synthesizer all coordinating around that one root task. And this feature alone, I'll do a a separate video on it cuz this is very very helpful because uh before this update when the canban first rolled out, there's different types of work that can be quite messy especially if it's multifaceted. So if it's not just research, but there's also additional elements you've added there.
So let's say maybe you you want research and then done that into a presentation and then do that into a website. So you see how it's like a lot of work, right?
So if you ask your Hermes to build that, the job itself rarely stays as one clean prompt. What often happens is it it branches. So you would end up with setup work, verification work, uh I think some edge cases, cleanup, and maybe a few parallel subtasks that could be done independently. So that's very messy because you have a lot of stuff lying around that needs cleaning up. So what this canban swarm is really doing is it's giving Hermes a more natural structure for that kind of work. So the big idea is Hermes is now getting better at breaking the work apart instead of forcing everything uh through one conversation stream. And this actually makes it useful for coding projects. I think this feature alone can actually make Hermes more capable of being a coding harness rather than just a general agent harness because of the operational detail that's built around it and the integrations that's layered on top. So they have added uh work tree per task. It means coding jobs can run with their own branch and working path instead of colliding in one directory.
It's very useful if you want your Hermes exploring multiple fixes or multiple approaches in parallel. They also have claimed TTL. This means a task does not stay stuck forever if one worker grabs it and then hangs. They also have retry fingerprinting which helps avoid uh those pointless loops where the same failure just repeats and repeats again.
And then they also have uh stale task detection with respawn guards that can help the system recover in a more controlled way when something goes wrong. So we'll do a dedicated video covering the Hermes cambban swarm. This is a pretty long topic, but it's certainly very very helpful uh for making your Hermes agent a lot more capable for coding. Uh, in other news, we also have Notify being the 23rd messaging platform, which is great. But honestly, just don't use any messaging platform. Just stick to TUI and CLI.
That's I think what Hermes agent is designed for. I mean, if you want to use messaging platforms, then then use OpenClaw for that. And they also have Korea for image gen, but this one you need a Korea API key. I actually thought the subscription plan of the news portal covers this, but it does not. Uh now bit warden this one it sounds boring at first but if you are managing a lot of provider keys services or environments uh you need to handle these a lot better. Now I know you can just put all of that uh credentials in yourv file but uh there's levels to that. It can get a bit messy. So the bit warden what it can do here is your uh by installing that your Hermes can use one bootstrap token and pull those secrets more cleanly. So it's easier rotation less config mess uh and a better security uh hygiene. We also got the MCP catalog. We covered that yesterday. Uh and security. So this one I know it's another boring one but this is a very important one. So the Hermes team, the news research team, they had a look at a recent brainworm class uh prompt wear uh which is actually very scary. What it can what it does is it hijacks the agent using tool output, recalled memory or stored skills. So with this update, this sort of attack is now blocked at basically three uh choke points at those levels.
So uh tool output, recalled memory, and stored skills. They also added 15 new threat patterns and a security guidance plugin which matters a lot because the more capable and uh connected your agent becomes the more dangerous untrusted inputs become as well. So this release is clearly trying to harden the system as Hermes gets more powerful. Um you know as Hermes gets better now you really want to pay attention more to the security aspect as well. I know we just, you know, skipped through these ones because it sounds boring. It works under the hood. Nothing really uh we don't need to do anything about it, but it's nice to know what's going on behind the scene as well because hey, it's getting more popular. More and more attackers are trying to get uh take advantage of this and you know, you can be a victim of that if you're not careful, right?
Also, we got the skill bundles, which we covered a few days ago. You can check that one out. There's also a TUI session orchestrator that lets you switch between live sessions uh without leaving the interface. Honestly, I don't really use this. I just I want to stick to one TUI at a time. I don't want to have, you know, multiple. I'm not at that level yet. Okay. So, if you are like Tony Stark, then you can do this one. Okay.
So, that's all for the 0.15 update. Um a couple of hours later, they noticed some things broke after the update. Nothing big but the hot fix came very fast. So what happened here was they fixed the uh dashboard infinite reload loop in loop back and docker mode and they also patched the cananban worker uh behavior.
They also restored MCP bear command resolution in docker and brought back the full uh 19,000 entry skills catalog.
So, that quick follow-up is good to see because a release this huge usually shakes out a few rough edges, but you can see that they're following up on the broken stuff ASAP rather than having to wait a couple of days. All right, so I hope you find this video helpful. Smash up that like button if you do. Let us know any questions you have regarding the Hermes agent update. And I honestly want to know, how's your experience using Hermes Agent so far? Has it been smooth? Has it been faster? Because it most certainly has for me. All right, subscribe to the channel to follow for more updates and guides. My name is Ron.
Videos Relacionados
OpenHuman VS Hermes AI: Who Wins?
JulianGoldieSEO
285 views•2026-05-29
Long-Running Agents — Build an Agent That Never Forgets with Google ADK
suryakunju
142 views•2026-05-30
This computer is made from real human brain cells. And you can buy it.
Talktmsmedia
3K views•2026-05-28
BREAKING: Microsoft’s New Image Generating Model Beat Out GPT 1.5 and Nano Banana 2
aimmediahouse
122 views•2026-06-03
I Made the Same Anime Fight Scene in Every AI Video Generator
NobleGooseAnime
295 views•2026-05-30
Nvidia Bets Big On AI PCs | New Chip To Power Windows Laptops | Technology | AI Updates | N18S
cnnnews18
3K views•2026-06-01
I Tested NEW Opus 4.8 on Four Projects (Updated LLM Leaderboard)
AICodingDaily
298 views•2026-05-29
3D Platformer Update - NO CAPES
SolarLune
294 views•2026-05-30











