OpenClaw 3.22 Update: What's New (Release Notes Breakdown)

0:00Open-interpreter 3.22 just came out and it seems like it's been a whole year since we got the last update from 3.13 to where we're at now. And in this video, we're going to be catching you up to all the latest features. So, 3.22, this is a huge one. Like I said, it's been over a week since this last update, and it's not just one of those minor patch updates. You know, they fixed a couple bugs and they did much more. This is a massive release. There are features in here that I've been waiting for personally, and there are some breaking changes that you need to know about before you update so you don't break your setup. So, stick around cuz we're going to get right into it. So, I'm going to set the stage real quick. If you've been following the channel, you know I run my entire content operation on Open-interpreter. We have 13 agents

0:45running on the Mac mini 24/7 writing my whole content machine. Thumbnails, X posts, podcast outreach, newsletters, blogs, the whole thing. So, this is my business infrastructure. So, when a new version drops, I try to pay attention to the details on how it's going to affect me. So, I'm going to start with the headline feature. Claw Hub is now the default for skill installs. And if you don't know what that means, I'm going to break it down. Before this update, when you wanted to install a skill or a plugin, Open-interpreter would go to NPM first. NPM is where all the Node.js packages live. It's like the App Store for developers. But now, Open-interpreter checks Claw Hub first. Claw Hub is basically the community marketplace built for specifically for Open-interpreter skills, and they also

1:31added native commands for it. So, you can now type Open-interpreter skills search and then Open-interpreter skills install and then Open-interpreter skills update right from the command line. So, there's no more going through NPM, no more guessing which package is the right one. And why should this even matter for you? Because the ecosystem is growing so fast. People are building skills for everything. The weather, YouTube analytics, social media posting, web scraping, SEO, and now finding and installing those skills is just much easier. What I would recommend if you want to be a really cautious on the security standpoint is find a skill that you like, copy the link, give it to your agent, and then have it recreate that skill. If you're, you know, wanting to not download skills that you don't know

2:16who made them. You know, just as a precaution, that's what I do. So, you know, it's amazing that they integrated this and it's been a long time coming. You know, this is basically like going from searching for apps on the marketplace on the website and just opening the app store on your phone. That's the difference. That's how they integrated this. You know, it's plain and simple, very very easy for anyone that wants to get going start installing skills. Or if you want to be super secure, just go with what I suggested. This is going to be a huge one for anyone who wants to build a new capabilities to their agent without writing any code. But who does Who writes code anyways? I mean, that doesn't even exist anymore. So, the second biggest one is GPT-5.4 support. You know, OpenAI just dropped 5.4 a week

3:01and a half ago and Open-interpreter already supports it natively in this release. The default OpenAI model has been switched to GPT-5.4. They also support GPT-5.4 mini and nano. So, you've got the full range. The big model for heavy thinking, the mini for everyday tasks, and the nano for quick and cheap stuff. 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 Claude code course, a full Open-interpreter 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

3:47provide one-on-one coaching, so you could book a call with me, we could share screens, and I can help you get Claude code or Open-interpreter 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 over 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. So, here's what's really interesting. People on Reddit were already asking about this a few weeks ago. There is a whole GitHub issue tracking GPT-5.4 availability within Open-interpreter, and the community was saying, "Hey, you know, we can see the

4:32model upstream, but we can't select it." Now, you can. So, it's baked in, you know, forward compatible. If you're running your agents on OpenAI models, this is going to be a massive upgrade for you. 5.4 supports up to 1 million tokens of context. So, similar to Opus 4.6, 1 million tokens. That's a lot. That means your agent can hold a lot more information in its head when you're working on a task and skip the dementia part. So, if you're building with OpenAI through Open-interpreter, update today. Don't wait on it. Number three is per agent thinking and reasoning defaults. Okay, this one's very cool because before this update, if you set a thinking mode or a reasoning level, it applied to everything. You know, every agent, every task, same settings. But that doesn't make sense when you think

5:18about it. My research agent doesn't need the same reasoning depth as my coding agent. My X posting agent definitely doesn't need deep thinking mode to write a tweet. So, now each agent can have its own thinking in its own reasoning and fast mode settings. This means if you're running agents like I do, you could set your main brain agent to deep reasoning, set your social media agents to fast mode, and set your coding agents somewhere in the middle. It's like giving each employee on your team the right tools for their specific job instead of giving everyone the same toolbox. That's pretty cool. The fourth thing, and this one kind of blew my mind, is Claude, Codex, and Cursor bundle support. You can now install skills from Claude marketplace, from

6:04Codex, and from Cursor directly into Open-interpreter. So, cross-platform skill sharing. If you think about what that means, the skills that people are building for Claude, you can use them in Open-interpreter. The tools that Codex users are creating, you could pull them right over. The extensions from Cursor, same thing. Open-interpreter is becoming the universal platform that connects all of these ecosystems together. I don't want to overhype it, but this is the kind of feature that makes Open-interpreter a good hub for everything on your AI machine. So, the fifth thing is I covered this in a recent video, but the /btw command that you can now do in Claude code, you could do this now in Open-interpreter. You know, have you ever been in the middle of a coding session with your agent and you're deep in some complex task and then you want to ask a quick question like, "Hey,

6:50what's the weather outside?" or "Hey, what time is my next meeting?" But if you ask that question, it changes the whole context of your session and it takes up and fills up that context window. Your agent starts thinking about the weather instead of the code you were working on. Well, now you could type /btw and ask your side question. It doesn't change the session context at all. Your agent answers the question and goes back to what it was doing. This is one of those quality-of-life features that sound small, but if you're going to use it every single day, it's going to help you save more on context. And I've been wanting something like this since I started running my agents, so very nice. New web search providers as well with this update. You know, before the update, Brave Search was pretty much your only option for web search. You could use the Chrome DevTools MCP. There

7:37was some other options like Firecrawl, but now it's all built in with an Open-interpreter. You know, Brave is great, but now they bundled Exa, Tavily, and Firecrawl as search plugins right alongside Brave. So, more options for web research. Different providers have different strengths. Exa is great for finding specific topics on content. Tavily is built right in for AI agents specifically. Firecrawl is amazing for actually scraping and reading web pages. So, if you're building agents that do research, and let me tell you the most useful agents you need are the ones that do research. You know, having multiple search providers is a pretty big deal in that case. So, you're not locked into just one or two anymore. The seventh release or the seventh update within this release that really matters most is

8:23SSH SSH sandbox backend. So, this is one for more technical folks, but it is important because before this update, if you wanted to run a sandbox code, meaning code that runs in a safe, isolated environment so it can't mess up your actual machine, you basically needed Docker. Docker is great, but it's one of one of those more things to install and really kind of hard to wrap your head around when you're just starting out. And, you know, it's just one more thing that could break down the road. So, now you could run sandbox code on remote machines over SSH. You've got a full VPS somewhere, a spare Raspberry Pi, you could point Open-interpreter to it and run the code there. Not just Docker anymore. So, this opens up a lot of possibilities for people who want to

9:08run heavier tasks without loading up their main machine. The eighth item in this release is the Chrome DevTools MCP upgrade. The browser automation in Open-interpreter just got way better. It now supports Chrome, but Brave browser as well, and Edge, and basically any Chromium-based browser through user data directory profiles. So, if you're using Brave as your daily browser, which I am, and your agent can now interact with it properly. Before this, you had to jump through hoops. Now, it just works. The ninth installment of this upgrade is Anthropic Vertex AI support. You can now run Claude through Google Vertex. So, this is more of an enterprise thing. If you're a company that needs to run everything Google Cloud for compliance or whatever reason, you can now do that

9:54with Claude models in Open-interpreter. Not everyone needs this, but for the people who do, it is a pretty big deal. The next one is image generation just got standardized. You know, they removed the old Nano Banana skill wrapper and replaced it with a native image generation tool. You just set the image generation model in the config and then it works. No more installing a separate skill for that and no more dealing with third-party wrappers. It's clean, built-in image generation, which is pretty cool. Android upgrades also, if you're running in open claw with an Android node and a lot of people in our community are, you're getting a dark theme, call log search, SMS search. Your Android nodes are becoming way more capable. Dark theme alone is worth the update and I'm not going to lie about that. It just looks much better.

10:39Telegram upgrades also in this latest installment is auto rename DM topics. You can now organize the topics a bit better, custom bot API endpoints for people who self-host Telegram bots and silent error replies. That last one is nice because before, if your agent hit an error during a Telegram conversation, it would just send back an ugly error message. Now it just handles it quietly. The 13th biggest upgrade here is MiniMax M2.7. If you're a MiniMax user, the default model is now M2.7 and they added fast mode with the high speed flag. So that is really cool for the MiniMax users. The 14th installment here is the control UI improvements. Expand to canvas on

11:25chat bubbles, a roundness slider for themes, which I thought was a funny addition, but people were asking for it, and better usage views, so you can actually see where your agents are spending. Here's the part that you really need to pay attention to. Breaking changes. If you update without reading this part, things will break, so please listen. The old Claude bot _ and mult bot _ environment variable names are removed, gone forever. So if you still have those in your config, they won't work anymore. You need to switch everything to open claw _. If you set this up a while ago and haven't touched it, go check your environment variable files right now. The old mult bot state directory is no longer auto detected as well. If your

12:11state files are still on the old .mult bot, you need to move them to open claw. Otherwise, open claw will not find them. If you're struggling to keep up with content, I'm about to save you about 40 days worth of work. I built something called content machine. It's 10 AI agents that run on the open claw orchestration and they handle everything. Scripts, thumbnails, X posts, blogs, outreach, clips, newsletters, 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 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,

12:59finance, real estate, marketing, whatever you are building and it is 100% completely customizable to your use case. So you get the mission control dashboard, all of the cron jobs, everything I've built over the last 40 days 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 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. The third thing is Nano Banana Pro skill wrapper is completely gone like I mentioned earlier. You need to use the new image generation model config instead. The plugin SDK changed as well. If you're a plugin developer, you need to migrate. Check the docs. The fifth

13:45thing is the Chrome extension relay path was completely removed. If you were using that, run open claw doctor with the fix flag and it'll sort it out for you. Now real quick on the security because this matters more than you even think. They blocked JVM injection attacks through the environment variables like Maven ops. They hardened the voice call webhooks against unauthenticated callers and they blocked built tool environment variable exploitation. If you're running agents that execute code, which you should be running in a sandbox anyway, these security fixes are keeping you much more safe. So my take on this whole release, this is the update where open claw stopped just being an AI tool and started becoming the operating system for AI agents. This was one of the

14:30longest times we had to wait for the latest update and I knew Peter was in the kitchen cooking it up and all the other people working on this project. So this was a massive update and I needed to make this video to keep you up to date on what is going on. Claw hub as the default marketplace, that is probably one of the coolest probably upgrades in this whole thing. Cross-platform skill sharing from Claude, Codex and Cursor, GPT-4 baked in, per agent reasoning controls. These aren't small features. This is a big infrastructure overhaul. I've been running open claw since the early days. We're almost approaching day 50 now since I got my Mac mini and I remember when it was way harder to set up. This was hours just to get it working and I remember when you couldn't even do half the stuff that comes built in now. And every release, they just keep making it

15:15better and better, just much easier for people that are not technical to get this thing going. The pace of development on this project is in the exponential phase. So if you're thinking about getting into open claw or even sitting on the sidelines watching people like me build with it, this is the version to jump in on. The learning curve is getting smaller, the skills marketplace makes it easier to add capabilities and with GPT-4 support, you've got the latest and one of the greatest models to go right out the box. So if you want to learn how to set this up and actually build something with it, that's literally what we teach inside of our community. We have 121 members in the last 13 days and they're all building. We're sharing what we're building. We're focused on community first and working in group-like settings. Yes, there's courses in there

16:01on Claude code and open claw, but we are driven to actually work with people live, human-to-human interaction and helping you take the next step in your AI journey. So I'll put the link down below and you could check it out. Make sure you hit the subscribe button too because we do release about four to six videos every day and if you like AI news, check it out. Hit the notification bell. You're going to see everything from like updates to real builds and what I'm doing with open claw to just AI breaking news. You'll see it all and I hope you guys have a blessed rest of the day. We'll see you in the next one. Bye-bye.