OpenClaw 5.19 Update: Builds Reliable Agent Infrastructure

0:00OpenClaw 5.19 just dropped and this is more than just a regular release from the OpenClaw team. There is a ton of features that you need to be up to date with because we are past the era of when things used to break every time you update and agents failing and losing memory. 5.19 fixes a lot of common bugs that people just really got annoyed about and started switching over to Hermes. So in this video we're going to be breaking down all of those features. You know, this isn't the flashiest update but there are some key features here that you need to pay attention to. You know, this is also not one of those releases where you could say, you know, the agent can now control your whole computer with just one button and print a million dollars. We're not quite there yet and this is not a viral demo release at all either. This is infrastructure,

0:45making it more reliable, making it more compatible with your setup and that's exactly why this matters especially if you're looking to sell agents as a service. So the source here is the official GitHub release for 5.19 beta and it was published by Peter Steinberger just yesterday. I also checked the live GitHub API this morning and OpenClaw/OpenClaw was sitting at 373,000 stars. That's just incredible on where we came just a few months ago. And also 77,000 forks by the way. And it had been pushed here in the same morning. That's amazing just with this 5.19 update. So yes, Hermes is getting a ton of attention right now. I'm even using

1:31Hermes as well in my daily setup. I think we should be using both if you are looking to sell agents as a service or get into any kind of agentic workflow. There's no reason why you should be choosing one over the other. I think you should be using both. So Hermes, the hype is real. It's very easy to set up but with OpenClaw I feel like you can even go a little bit further taking the time to set it up correctly and also implementing it for your business. You know, and I've been saying that, I've been doing that for the last few months, and once you get it set up, it just works. So, this release shows the difference between the hype and the operator infrastructure. A lot of great fixes in this update. You know, where Hermes is winning, is winning in the attention, in the ease of setup, and it's winning in a lot of places, Frank.

2:16But, Open Claw is quietly building the rails, the infrastructure, the backend. And rails are what make agents useful after the demo is over, right? Here's the big angle from this release. Open Claw 5.19 is starting to look like the agent app store layer. Like, crazy. And you should really check out the release notes. I have it linked down below. And this isn't because there is a literal consumer app store with pretty little icons, it's because the plugin and skill system is getting more serious. This release adds define tool plugin, plus Open Claw plugins build. These validate and init for type simple tool plugins with generated manifests and metadata, optional tool declarations and context

3:01factories. I know this sentence sounds boring, but I get it. If you are just building any type of agent workflow, this is not boring at all. This is what's going to actually make your agents work. That means Open Claw is basically making it easier for you to turn on one-off tools into actual real plugins. A real plugin has shape. It has a manifest. It can be validated, and it can be built, right? And it can declare what tools it exposes. It could connect to runtime context. That's the difference between hacking something together at 2:00 a.m. and shipping something other people can actually use. And this is where the most of the agent platforms out there break. Everybody can make a demo. It's really easy. That's why people spend a lot of time making flashy little demo releases. Hence, the

3:47Google I/O drop yesterday. Real quick, before we keep going, if you're watching this and you want to actually build with some of these tools, not just watch videos about them, then you're going to want to check out our community down below, Shipping School. We have a full Cloud Code course, a full Open Claw course, and four live bootcamps every single week where we actually help you get set up from scratch. Like actually set this thing up, not just watch a tutorial and figure it out by yourself. And we also provide one-on-one coaching, so you could book a call with me, we could share screens, and I can help you get Cloud Code or Open Claw running on your machine. That's it, no fluff. I built this community because watching YouTube's only gets you so far. We launched it just 3 days ago and we have

4:32over 55 members. You need people around you who are actually building, people who hold you accountable, and coaches who can help you when you get stuck. I'll put the link in the description down below. Get in now before the price goes up. In my opinion, that was a complete fail. Everybody can wire a tool call as well. Everybody can get an agent to do something cool at least once. But, can you actually package that workflow and sell it? Or can you actually use it in your business in production? Can someone else even install it? Can the system even validate it? Can the agent understand what the plugin does without reading the whole brain and the whole context layer? Can you update it without breaking everything? That is the hard part here. And Open Claw 5.19 is pushing into that hard part. The second thing

5:18that jumped out is the global shared managed skills. The release says Open Claw skills install and Open Claw skills update can target shared managed skills with a global flag. So, this is again not a viral ask right here, but if you think about the actual builder use case, that's amazing. And if you are running one agent, local skills are just fine. And if you're running a team or a community or a content machine or a business stack, local only skills become pretty messy quick. One agent has the updated skill, another agent has the old one. One workflow works on one machine, another workflow fails on the second machine. This can be really annoying, and someone fixes a skill in one place

6:03and then forgets to update the rest. That is how automation becomes chaotic. Global managed skills are part of the thing for turning agents from toys into shared infrastructure. That matters a lot for us in shipping school inside our community. We're kind of over the whole open claw breaking every time we update it, and it's really nice to see the whole community and team pushing for these updates. So, and it just matters a ton for people building businesses on top of open claw in general, you know? And that matters basically for anyone trying to teach normal builders how to set up repeatable workflows, because repeatability is the product, right? We want it to be reliable. We want it to actually work. We're spending a lot of our own time trying to get this to even do one thing, right? You know, if I teach someone a workflow in our

6:49community, and it only works on my machine, that's not a business. That's just a magic trick or a cool little demo, and who cares, right? The third thing in this update is browser reliability. Open Claw 5.19 adds browser modal dialog handling inside of snapshots. So, it could basically service pending recently handled modal dialogues, return blocked by dialog when an action opens a modal, and answer pending dialogues by dialog ID. So, this is one of those things that only sounds boring if you have never built browser agents. And if you have, you know this is huge. Browser agents die on the stupid stuff, like a cookie pop-up or a confirm dialog, a permissions modal, a login prompt, basically anything like a

7:35warning that blocks the page. Agent thinks it clicked it, but it didn't, and the page is blocked, and the next action fails, and then the whole task drifts. And if the agent does not know it got blocked by a dialogue, it starts guessing, which is the worst part. And that's when things get really dangerous. So, when Openclaw adds explicit dialogue awareness, that is not a small UX fix. That is trust infrastructure. The browser can now say, basically, "Hey, I'm blocked by a dialogue." The agent can now treat that as a state. A human can approve or reject it if it wants. But the workflow can continue is the important part instead of pretending that nothing happened. That is how you make agents basically less dumb, right? And not by giving them a bigger model

8:20every week, which is amazing, but by giving them a cleaner state out the box. The release also adds browser evaluate timeout controls. So, long-running page functions can extend request timeout budgets. So, the same idea here. Real websites are messy in general. Pages can hang from time to time. Scripts can take time. Workflows need more than a default timeout. So, if your browser agent is going to do real work, it needs controls like this. Come on. The fourth thing is the Codex plugin management from chat. Did you know you can plug in your Openclaw into Codex and basically have an amazing coding agent? Not a lot of people know that. Openclaw 5.19 adds commands to the list. Enable and disable configured native Codex plugins from the

9:05chat without editing config by hand. Who does that anymore? This matters because Codex is becoming a core part of the builder workflow. We are seeing this everywhere. Codex fixed builds. Codex watches workflows. Codex reviews the code. Codex runs cloud agents. Codex is not just an autocomplete anymore. So, like, if you haven't used the Codex app, what are you doing? You need to get on board. But the usage limits that you get on the $20 to $100 a month plan are absolutely crazy. I've never once hit my limits, and I'm using this every single day. You need to get it plugged in with your open claw. So, if open claw basically can manage Codex plugin surfaces just from your chat, that lowers your friction. You do not want

9:50normal builders editing config files every time they need to turn something on or off. What are you doing? They will break it. Just a matter of time. You know, you will get scared. It's intimidating, or you may just quit. The easier path is chat native control with clear boundaries. List what exists and name all the thing, disable the thing, and then just keep moving. That is how real adoption happens. The fifth thing, which is an important one, is the Mac app settings redesign. There is another people This is a just another one where people will just skip over, but it's super important, especially if you're using a Mac on your open claw. So, this does matter a lot. The release says basically the Mac app settings page got consistent card layouts, cache navigation, cleaner per missions, voice

10:35skills, crons, and exec and debug panes, plus better spacing around the native sidebar. This is not just cosmetics. Agent tools are super intimidating. Permissions are scary. Cron is confusing in general. Skills are confusing. Exec is super scary for normal people because they know it can touch their machine. Debug panes can also make people feel like they broke something even though you didn't. So, better settings are not fluff here. They are on board. And they are trust. They are the difference between a builder trying open claw for a week and a builder actually keeping it installed. This is where the Hermes agent has been putting pressure on open claw. Hermes has momentum because people feel like it is becoming easier to just

11:22get going. And it is. That's the crazy part. If you're struggling to keep up with content, well, I'm about to save you about 40 days worth of work. [music] I built something called content machine. It's 10 AI agents that run on the open claw orchestration and they handle everything. Scripts, [music] thumbnails, X posts, blogs, outreach, clips, newsletters, [music] all of it. So, I went from 1,000 subscribers to 4,000 subscribers on YouTube in 7 days using this exact system. Every single morning I wake up [music] and the content's already done. I spend maybe 15-20 minutes reviewing and approving them and I move on with my day. It works for any niche, fitness, finance, real estate, marketing, whatever you are

12:07building and it is 100% completely customizable to your [music] use case. So, you get the mission control dashboard, all of the cron jobs, everything I've built over the last 40 days [music] helping me gain more and more people to subscribe and join the community. So, you plug in your own thing and it molds it to you, it learns how you talk and it [music] writes so it doesn't sound like AI slop. $97 one time, it's not a subscription. I'll put the link down below and you'll thank me later. It's like how many of you guys have used and switched to Hermes just because it's easier to use. I mean, come on, it is. So, open claw really has to keep improving here and I think it makes open claw better overall, right? It puts the pressure on the open claw community. So, setting skills and daily operator experience are all getting better, not

12:52just the deep architecture. This sixth thing here is diagnostics. Open claw 5.19 adds gateway start startup probe, config, runtime, and resource count costs in restart traces. It also overlaps startup logging and plugin service startup with channel sidecars to reduce restart ready latency while preserving readiness gating. The translation here is when open claw restarts, it is getting better at telling you what had happened and where the time went. That matters because agents run like little businesses. If the gateway is slow, you need to know why. If plug-in startup is dragging, you also need to know why. If a channel sidecar is blocking readiness, you need

13:38to know. You cannot run a serious agent system just on vibes. Even though it's vibe coding, we need to actually know what's going on and why it's happening. You need observability. You need traces and you need proof. Otherwise, every failure becomes a mystery and then you don't understand how your workflow works. And that mystery will kill the trust with Open Claw. That is the whole lesson of this release. Open Claw 5.19 is not trying to win the internet with one flashy trick. It is trying to remove mystery from the system. Plug-ins got clearer. Skills got easier to share. Browser state gets clearer. Codex plug-ins get easier to manage. Settings get easier to use and restart behavior gets easier to inspect. And that is how agent platforms mature. We are on like month three or four here on this Open

14:23Claw. And it from when it first launched to where it is now, it's a completely different platform. I'm going to talk about the Hermes comparison because that is the question for everyone and everyone's thinking it right now. Does this make Open Claw safer from Hermes? I think yes, but not automatically. Hermes is still moving fast. It's one of the fastest growing projects out there. Hermes has real builder energy and Hermes has real use cases. Hermes has a very simple story for people to understand. The agent that grows with you and that is clean and I love that. Open Claw has a deeper operator story. But deeper does not always win. Clear wins, easy wins, repeatable wins. You know, I get I get that. But Open Claw 5.19 is a good move because it makes

15:08repeatability stronger. But Open Claw still has to package the story better for normal builders. Because most people will not read a release. They won't release the note. Like they won't read it and they won't understand why define tool plugins even matter. They need to see the actual outcome. So, here is the outcome. You can build clear plugins, you can manage shared skills more easily, and you can actually run browser workflows that handle pop-ups and models instead of say silently failing. You can actually manage your Codex plugin surfaces from the chat, which is a huge upgrade. And then you can also debug restarts with better traces. You can make the Mac app actually feel in less intimidating for normal people, and that's the story of this release.

15:53OpenRCA is becoming less of a hacker toy and more of an operating system for AI agents. And yes, I know people say that phrase way too often, but in this case, that's literally the direction they're heading. Operating systems are not just the flashy app. They are permissions, settings, plugins, debugging, and drivers with all shared resources and process management and crash recovery. That is what OpenRCA is building, and that is why I think this release is worth covering. If you are a builder, here is what I would do with this information. Do not just update right yet because the version number just changed. Look at your workflow and ask where these new pieces matter inside of your work. If you build custom tools, study the plugin chase. If you teach skills or reuse workflows across machines, study the global management

16:38skills. If you run browser automation, test the modal handling. If you reuse or use Codex inside of OpenRCA, look at the new chat controls for plugins. If you are helping normal people install OpenRCA, look at the Mac app settings cleanup. If your gateway restarts feel real random, look at the restart traces. That is how you turn a release note into a business advantage. Not by saying a new update, cool, that's amazing, but by finding the one piece in this update that removes friction from your actual workflow. For me, inside of Shipping School, this is also the kind of update that makes teaching easier because the hardest part of teaching AI agents is not showing a cool demo. The hard part is getting someone from zero to a

17:23workflow that they could trust on repeat. And that means fewer scary settings, cleaner plugin installs, better browser behavior, better logs, clearer skills, and less mystery when things break. That is what normal builders need. They do not need another thread about AGI. They need the whole thing to work after dinner when they sit down to build. So, my take here is simple. Open Claw 5.19 is not the loudest release, but it is one of those releases that tells you where this market is going. The future is not smarter models anymore. The future is agent infrastructure that normal builders can install, extend, debug, and trust. Hermes is pushing the market with speed and attention. Open Claw's answering with rails and infrastructure. And the winner, you know, at this point,

18:08can't we just have two winners? It's not one loses and one wins. I think they're both just building for two completely builders. Different builders, you know? And it's it's building at a rate where it's fast enough to where we're all excited about it. It's also getting stable enough for us to now trust, simple enough to us for to teach it, and powerful enough to build businesses on. That's the lane right here. 5.19 is a very clear step in that direction. If you need help setting up an AI agent, we have unlimited 101 calls in our community. If you want to book a thousand calls, go ahead. We'll help you get your agents up and running. We also do seven live calls a week in a group like setting. If you need help with your product, help with Hermes, Open Claw, Claude code, we're here to help. I'll put the link to it down below. You should check it out. And we'll see you

18:53in the next one. Have a blessed day.