OpenClaw 3.23 Update: OpenAI Token Fix, WhatsApp Runtimes, Qwen

By Beau Johnson··7 min read

Quick answer

OpenClaw 3.23 is a stability release that fixes what broke in 3.22: OpenAI tokens reverting to expired values, missing WhatsApp and Matrix runtimes, ClawHub auth on macOS, and false subagent timeouts. It also adds Qwen support through DashScope and a new Control UI. Update, then run OpenClaw doctor.

In this video I go through the OpenClaw 3.23 release. It came out roughly 14 to 18 hours after 3.22 and fixes most of what 3.22 broke. If you upgraded from 3.13 to 3.22 and your tokens, WhatsApp channel, dashboard or subagents started acting strange, this is for you. I also cover the handful of new features in this release.

I recorded this in March 2026. Tools like OpenClaw change fast, so check the current docs and release notes before you rely on any detail here.

What I cover in this video

  • 0:00 Why 3.23 shipped so soon after the big 3.22 release
  • 1:31 How 3.22 broke WhatsApp channels and caused 503 errors on dashboards
  • 2:17 Fix #1: OpenAI tokens reverting to expired values
  • 3:48 Fixes for bundled plugin runtimes (WhatsApp, Matrix) and ClawHub auth on macOS
  • 5:20 The subagent timeout fix and how it affected my Wu-Tang content pipeline
  • 6:05 Fixes for the ClawHub install version check, OpenRouter pricing, and the web search provider
  • 6:50 New in 3.23: Qwen through DashScope, the Control UI overhaul, CSP hardening
  • 12:51 How to update and clean up your config

Why 3.22 broke things

3.22 was a big release, with a new plugin SDK, new security features and a lot more. When a team ships that much at once, some of the changes conflict and things break. I updated that morning, made my video, and then saw problems once my agents had been running for a few hours. People in the community were reporting the same issues. Someone posted on dev.to that 3.22 had broken their WhatsApp channel and their dashboard was returning 503 errors. The WhatsApp runtime files weren't in the npm package, and the Control UI assets were missing too.

What I care about is how quickly the team fixed it. They found the issues, fixed them, tested and shipped 3.23 within the same day.

The key bug fixes in 3.23

  1. OpenAI token auth. You'd save your API key, and an hour later it had reverted to an expired value. That locked agents out of GPT. If you run cron jobs, you might not notice until you wake up to a pile of failed tasks. Tokens now stay saved.
  2. Bundled plugin runtimes. The runtime files for WhatsApp, Matrix and other plugins were in the repo but missing from the npm package, so a fresh global install had no working channels. WhatsApp and Matrix work again, and you don't need to downgrade to 3.13 anymore.
  3. ClawHub auth on macOS. ClawHub was quietly falling back to unauthenticated mode on Macs. That meant you could miss premium skills or get rate limited without knowing why. It's fixed for both the default path and the XDG path.
  4. Subagent timeouts. This one mattered most to me. My "Wu-Tang pipeline" is a set of cron jobs where subagents like RZA (script outlines), Inspectah Deck (X), Ghostface Killah (intel) and U-God (thumbnails) create content. In 3.22, workers that finished quickly were being reported as timed out. I was getting error notifications for jobs that had actually worked, and I stopped trusting my own pipeline. Fast subagents now report correctly.
  5. ClawHub install capability check. The checker was reading the wrong version. It told you that you needed 1.2.0 when you were already on 3.22, which blocked skill installs. It now checks your actual runtime version.
  6. OpenRouter pricing. The OpenRouter auto model was recursing infinitely during bootstrap, so cost tracking didn't work and you couldn't see what you were spending. That's fixed.
  7. Web search provider. You could set Brave as your search provider, but agents still used a stale default. They now use the provider you picked.

There are also smaller fixes. Mistral's max token defaults were set too high and caused 422 errors, so they've been lowered to safe values. Telegram threading now passes topic context in DM topics. Discord privileged commands return proper unauthorized errors. Chrome MCP on macOS now waits until tabs are actually usable before it reports ready.

What's new in 3.23

Qwen through Alibaba Cloud DashScope

Qwen support is now built in through DashScope. It's pay-as-you-go and works with both China and global Qwen API keys. Qwen performs well for a fraction of the cost of Anthropic or OpenAI models. I wouldn't move everything to it, and I wouldn't use it as your coordinating agent. It does work well for lower-priority agents that handle things like web scraping, script writing, reading simple blog posts and scraping Reddit.

Control UI overhaul

The Control UI has a new black and red theme and now meets WCAG 2.1 AA accessibility standards. The config icons for the diagnostics, CLI, secrets, ACP and MCP sections are clearer. The roundness setting now uses fixed steps instead of a slider.

CSP security hardening

The Control UI now uses SHA-256 hashes for inline scripts and blocks inline script injection by default. If malicious code tries to inject itself into your dashboard, it gets blocked. This matters if you run your gateway on a network.

How to update

If you're on 3.22 and things have felt off, update. You can just tell your OpenClaw agent to update itself. Then run doctor so your config files are clean:

openclaw doctor

The doctor fix option repairs old configs automatically, including the Mistral token settings. If you downgraded to 3.13 because of the WhatsApp or dashboard problems, you can safely upgrade again.

I've run 13 agents on cron jobs on my Mac mini for 45 days. My main concern isn't having the newest features. It's whether the work is done when I wake up and whether my subagents report their status accurately. Trust and reliability are what this release is about.

Key takeaways

  • 3.23 is a stability release that fixes the main problems people hit after upgrading to 3.22.
  • If you run agents on cron jobs, the fixes for token reverts and false subagent timeouts matter most.
  • WhatsApp, Matrix and the dashboard work again on fresh npm installs, so you don't need to stay on 3.13.
  • Qwen through DashScope is a cheap option for lower-priority agents. Keep it out of the coordinator role.
  • After updating, run OpenClaw doctor to clean up your config.

FAQ

What does OpenClaw 3.23 fix?

It fixes OpenAI tokens reverting to expired values and WhatsApp and Matrix runtimes missing from the npm package. It also fixes ClawHub auth on macOS, false subagent timeouts, the ClawHub install version check, OpenRouter cost tracking, and the web search provider setting not sticking.

Is it safe to upgrade from OpenClaw 3.13 to 3.23?

Yes. If you went back to 3.13 because of the WhatsApp or dashboard problems in 3.22, those are fixed in 3.23 and you can update again. Run OpenClaw doctor afterward to make sure your config files are clean.

Why were my OpenClaw subagents showing timeouts when they finished?

In 3.22, workers that finished quickly were wrongly reported as timed out, even when the task had succeeded. 3.23 fixes this, so fast subagents now report the correct status.

Does OpenClaw support Qwen models?

Yes. As of 3.23, Qwen is built in through Alibaba Cloud DashScope on pay-as-you-go pricing, with China and global API keys. I recommend it for lower-priority agents, not as your main coordinating agent.

Transcript

Read the full transcript

0:00Open Claw 3.23 just came out and it's only been about 18 hours since the last drop. This is built in with a much more stability fixes, little feature enhancements, and if you're having a hard time going from a 3.13 to 3.22, we're going to talk about that in this video and what this new upgrade brings you. So, you know, it's not as fancy, not as feature rich, but it does cover some simple bug fixes that if you had a hard time getting it going, this is probably the reason why. So, we just dropped the video 3.22 earlier today, and that's going well, and it was a massive release. I mean, new plugin SDK, new security features, tons of new stuff. But, here's the thing that nobody talks about. When a team ships something that big, that fast,

0:46things break. And that's probably what you experienced this morning. You know, that's called how software works, basically. Welcome to software development. You ship 100 new features, and some of them step on each other's toes. Some break. That's just what happens. So, I updated to 3.22 just this morning and got the video out and was hyped out about doing it. And then I started running my agents throughout the day, and then I started noticing a few things that were different. So, my OpenAI tokens kept acting weird. I'd save my API key, come back an hour later, and it just reverted to some expired value. I was getting locked out of my own models, and I know I wasn't the only one because I was seeing people in the community talking about the same exact thing. It gets worse. There was a post on dev.to today. Someone wrote a whole article about how

1:313.22 broke their entire WhatsApp channel and their entire dashboard was throwing 503 errors. The WhatsApp runtime files weren't even included in the NPM package. So, if you did a fresh global install, your WhatsApp channel just vanished. Great. No messages in, no messages out, and the control UI assets were missing, too. So, your dashboard was just a 503 error page. Amazing. That is not a great look for the day of a big release. But, you know, I'm not throwing shade. It's just, you know, we we had gone like 10 days without a release, and I knew something big was happening. This is what happens. You know, you move fast, things break. And here's what separates a good one from a great one as far as teams, you know, is how fast they fix it. Within, what, has it been 14, 16

2:17hours since the morning release? There's already a new release out now, and Open Claw just dropped 3.23, you know, and it fixes everything. So, let me walk you through the key fixes because some of these are going to save you a ton of headaches if you're building with Open Claw right now. The first one is the OpenAI token off fix. This was the one that was driving me crazy and everyone else. So, what was actually happening is when you saved your API token in the settings, the system was reverting back to it to an expired value. So, you'd configure it, everything works for a minute, and then suddenly agents can't talk to GPT anymore. And if you're running cron jobs like I am, you don't even notice it until you wake up and see a bunch of failed tasks. This connects to the whole OAuth flow that a lot of people have been struggling with. It's completely fixed now. Your tokens stay

3:03where you put them. 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 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. also provide 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 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 over 55

3:48members. 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. The second big one here is the bundled plugin runtimes. This is the WhatsApp and Matrix issue I was just talking about. The runtime files for WhatsApp, Matrix, and other plugins were literally missing from the NPM package. So, if you did an NPM install globally, those channels just didn't work. The code existed in the repository, but it wasn't packaged correctly. 3.23 fixes this. Your WhatsApp channel comes back. Your Matrix bridge works again. No more downgrading to 3.13 as a workaround. The third one is Claw Hub off on a Mac OS.

4:35This is a sneaky one because you wouldn't even know it was broken. What was happening is Claw Hub was silently failing back to unauthenticated mode on a Mac. So, when you were browsing skills or installing them, it wasn't actually using your login. It was treating you like a guest, which means you might have been missing out on premium skills or getting rate limited and not understanding why. But, they fixed for both of that on the default path and the XDG path. So, however your Mac is configured, it now actually works. The fourth one, and this is the biggest one for me personally, is the subagent timeout fix. So, I run what I call the Wu-Tang pipeline. It's a bunch of cron jobs that fire off at different times throughout the day to create content. The RZA writes the script outlines, and

5:20Inspectah Deck is over there on X handling things. Ghostface Killah gathers the intel. U-God makes the thumbnails and stuff. The whole thing runs on subagents. And what was happening in 3.22 is that workers that finished fast were being falsely reported as timed out. So, my agent would complete a task successfully, the script would be written perfectly, but the system would say it timed out. I was seeing error notification for jobs that were actually working. It was confusing, and it made me not trust my own pipeline. That's all fixed now. So, if a subagent finishes quickly, it reports correctly. No more phantom timeouts. The fifth one is the Claw Hub install capability. This one was blocking people from installing new skills. The plugin API capability checker was looking at

6:05the wrong version and kept saying you need 1.2.0 when you were already on a 3.22. So, every time out you tried to install a skill, it would actually block you. And now it checks against your actual runtime version, no more stale version gate. The sixth one is the Open Router pricing fix. If you're using Open Router, which a lot of us are because it gives you access to a ton of models through one API key, there was a bug where the Open Router/Auto Model was recursing infinitely during bootstrap. What does that mean in plain English? Your usage cost tracking was completely broken. You had no idea what you were spending, but that's fixed now. The seventh thing is web search provider. This is the one of those subtle bugs that makes you pull your hair out. You

6:50can go into settings, configure your Brave as your search provider, you save it, but when your agent actually runs a search, it's hitting the default provider instead. So, your configuration wasn't sticking for search. It was using some stale default. 3.23 fixes this, so your agents actually use the search provider you told them to use. Now, let me talk about the new stuff in 3.23 as well. Because it's not just bug fixes, it's more exciting than that. We got Qwen support through Alibaba's Cloud DashScope is now built in. This is pay-as-you-go, which is pretty cool. You can use China and global Qwen API keys. And here's why this matters. If you're looking for cheap model alternatives, Qwen gives you a really solid performance at a fraction of the cost,

7:35especially if you're working with Anthropic or OpenAI models, and especially for tasks that don't need the power of Opus or GPT. I'm not saying switch everything over, and definitely don't use it as a frontier model as your coordination agent. But, the lower priority agents like web scraping and script writing and reading simple blog posts to gather data, scraping Reddit, those are great. Qwen can save you a lot of money. The control UI got a overhaul as well, new black and red knot theme. Oh, baby, looking good. And more importantly, it's now WCAG 2.1 AA compliant for accessibility. Better config icons for diagnostics, CLI, secrets, ACP, MCP sections. The roundness controls are discrete in stops

8:20now instead of a slider. Oh my goodness, which just feels more intentional. It's a cleaner experience. CSP security hardening. This is the kind of thing you don't think about until bad things happen. They added SHA-256 hash line interface, scripts in the control UI. What this does is block inline script injection by default. So, if you somehow get malicious code that tried to inject itself in your dashboard, it gets blocked. This is a real security improvement for anyone running their gateway on a network. The Mistral model fixes are worth mentioning, too. The max token defaults were set way too high, causing 422 errors. If you've been trying to use Mistral models and getting weird errors, that's probably why. They lowered the defaults to safe values, and

9:05you can now run Open Claw doctor fix to repair old configs automatically. And then there are some specific fixes. Telegram threading now properly passes topic context in DM topics. So, Discord privilege commands return to proper unauthorized errors instead of some misleading generic message. And the browser Chrome MCP on Mac OS was timing out when attaching to existing sessions. It now waits for the tabs to actually become usable before reporting ready. So, here's what I want you to take away from this video, and I know it's not feature rich like the last update. And I think it's really important to know that not every release will be feature rich. It's all about shipping fast, and then iterating, fixing, debugging, and the loop goes on. You know, not every update

9:50needs to be flashy. Not every release needs to have some massive headline feature that makes you go, "Holy moly, guacamole." Right? Sometimes, the most valuable update is the one that makes everything you already have just work. I mean, come on. If you've been with doing Open Claw, just getting the thing working is fulfilling, right? It works. Great. Now, don't touch it. So, work working more reliably, that's what we want, right? We want it to work the way it was supposed to work in the first place. So, I've been running these AI agents on the Mac mini for 45 days now, and 13 agents running on cron jobs. We got them generating content, managing operation. And the thing is that keeps me up at night is isn't whether we have the latest features. I just hope that they do the work, and I wake up and everything's ready to go. If you're

10:35struggling to keep up with content, well, 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, [music] 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, [music] fitness, finance, real estate, marketing, whatever you're building, and it is 100% completely customizable to

11:21[music] 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 [music] 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. So, everything kicks off 5:00 a.m. By the time I'm at the computer 7:00 a.m., I'd expect for everything to be done. And it has been lately, you know? And that's my main concern is like whether my sub agents are actually going to report their status correctly so I could just get in to see what's going on. And that's what 3.23 is all about, you know, trust and reliability. And within 14 hours, the boring stuff that

12:06leads you to actually building something and lets you actually build a business on top of this platform, and not have it break. The fact that Open Claw team shipped this in less than 12 to 14 hours, I mean, that's pretty impressive. 3.22 just dropped like this morning, and things broke as they do within a single day. They identified every issue, they fixed it, they tested and shipped a whole new release. And that's the kind of team you want building the platform you're running your business on. I see a lot of tools in this space that ship fast but break things and then take weeks to fix them. Open Claw's shipping fast and fixing fast. That's what matters when your livelihood depends on it. So, if you're running 3.22 right now and things have been a little weird, just go update it. Just tell your Open Claw agent, update it. We're on a new version. Come on. Get with it. Run Open

12:51Claw doctor after to make sure your configure to configure files are clean. And if you were one of the people who downgraded to 3.13 because of the WhatsApp or dashboard issues, you can safely come back to life now. It's okay. You can download the newest update. Everything's been patched. So, if you're just getting started with AI agents or you're running them for a while and you want to level up, we have a community called Shipping School, where we now have 128 members in the last 12 days. Real people working on real projects. We're not just a bunch of AI avatars talking back each other. We have a we're community focused. We do six live calls every single week and build sessions, roast sessions, technical deep dives, helping you take the next step in your building journey. I'll put the link to it down

13:36below. And, you know, we're growing like crazy, and I love it because we're all like-minded people trying to build the same thing, working on building the same thing. I do have courses in there with Claude code and Open Claw if you want to check that out and learn as you go, but we're really focusing on the community and building in public and sharing each other's feedback. So, what I say is just if you're having a hard time with Open Claw right now and you're on an older version, just go update. Go update it all. Make sure your agents are running clean, and remember the flashy features get the views, but the stability updates are actually what keeps this thing working and alive. Comment down below, what did you think about this latest update? And if you haven't subscribed already, we really release like four to six videos every single day covering all aspects of AI. Subscribe, hit the notification bell. We'll probably see

14:21you in about 30 minutes with another update. Be blessed.

Ready to start building with AI?

Join the Shipping Skool community free. Paid members get live bootcamp calls every week and weekly-updated courses.

Join Shipping Skool

Keep reading