OpenClaw 3.28 Update: Grok X Search, Plugin Approvals, and Fixes
Prefer to watch? Watch on YouTube ยท Subscribe for new builds ยท Read the transcript
In this video I go through everything in OpenClaw 3.28: the new features, the bug fixes, and the breaking changes you need to know about before you update. It's for anyone building AI agents or automations on OpenClaw, especially if you run bots on Telegram, Discord, WhatsApp, Slack, or iMessage.
I recorded this in March 2026, and 3.28 was the fifth major OpenClaw update that week. Tools like this change fast, so check the current OpenClaw docs before you rely on any of the details below.
What I cover in this video
- 0:45 โ Native Grok X search, and how my 5:00 a.m. trend-scouting agent uses it
- 3:01 โ MiniMax image generation built into OpenClaw
- 3:46 โ The plugin approval system for security and control
- 5:17 โ ACP channel binds: Codex inside any chat
- 6:03 โ The Gemini CLI backend, OpenAI apply_patch, and Slack file uploads
- 7:37 โ MiniMax catalog cleanup and the start of the bug fixes
- 8:22 โ The WhatsApp infinite echo loop fix, plus the Telegram, Mistral, Discord, and iMessage fixes
- 11:28 โ Breaking changes: Qwen portal OAuth removed, old config migrations dropped
The headline features
Grok X search, built in
This is the one I'm most excited about. Grok, the xAI model, can now search X (Twitter) natively. You don't need a plugin or a workaround, and there's nothing extra to configure. You turn it on during onboarding, it's built into the responses API, and it's auto-enabled.
This matters for my setup because my first agent, Ghostface, fires off at 5:00 a.m. and uses Grok to scan X for trending topics in the AI space. It then feeds that data to the rest of the agents in my pipeline. Before this update that needed a plugin, which you had to set up by hand and which sometimes broke. Now it's first-class, so Ghostface got an upgrade without me doing anything. If you do trend analysis, content research, or any kind of intel gathering, I think X and Reddit are the best places to start, and now your agent can pull what people are talking about on X right now.
MiniMax image generation
OpenClaw now natively supports MiniMax's image-01 model. You can generate images, edit them, and control the aspect ratio. I use kie.ai, a marketplace for image and video models, for our thumbnails and UGC content, and it works great. A second option built right in means you can compare outputs and use different models for different jobs, without adding another API, service, or dashboard.
Plugin approval system
Before 3.28, when a plugin wanted to run a tool, it just ran. Now plugins can pause and ask for your approval first. You get a button in Telegram, an interaction in Discord, or you can use the slash approval command. If your agent has access to your email, your database, or your social media accounts, you probably want to see what it's about to do before it does it. You decide how much autonomy your agents get, which is how it should be.
Coding agent upgrades
ACP channel binds
You can now turn whatever chat you're in into a Codex backend workspace. In the video I describe the command like this:
/acp spawn codex bind
You don't need a child thread or a separate workspace. Codex works in your DMs, in your Discord channel, and in iMessage conversations. We use Codex for a lot of our coding work, and before this you had to spin up a separate thread to use it. Now you can bind it to the conversation you're already having, so there's no context switching.
Gemini CLI backend
OpenClaw now bundles Gemini CLI support next to Claude CLI and Codex CLI, so you get three coding agents out of the box and can pick the best one for each task. In my opinion Gemini is better for front-end design. If you're trying to choose between the other two, see Claude Code vs Codex.
apply_patch, Slack uploads, and MiniMax cleanup
- OpenAI apply_patch is now on by default for OpenAI and Codex models, so the edits Codex makes to your files should come out cleaner and more accurate.
- Slack file uploads now have an explicit upload file action for channels and DMs. If your agent had trouble sending files through Slack, that should be fixed.
- MiniMax model catalog has been cut down to M2.7 only. The legacy M2, M2.1, and M2.5 models are gone, so there's no more guessing which one to pick.
The bug fixes
Some of these fixes matter just as much as the new features:
- WhatsApp infinite echo loop: the bot was treating its own replies as new incoming messages, so it ended up talking to itself forever. I run Telegram myself, but people were complaining about this on X all week. It's fixed.
- Telegram message splitting: long messages were being cut mid-word. They now split at word boundaries.
- Mistral 422 errors: random 422 errors with empty bodies on the official Mistral API are fixed.
- Discord reconnects: stale gateway sockets and poisoned resume states were causing reconnect loops that never succeeded. Reconnects are clean now.
- iMessage reply tag leak: reply tags were showing up in the delivered text. Messages come through clean now.
- Image analysis restored: the generic fallback for providers like OpenRouter and MiniMax works again.
Breaking changes to check before you update
- Qwen portal OAuth has been removed. You need to move to a Model Studio API key.
- Config migrations older than two months have been dropped. If you still use old legacy config keys, they'll fail validation, so update them.
Key takeaways
- Grok can now search X natively in OpenClaw with no plugin, which makes real-time trend scouting agents much easier to set up.
- The plugin approval system lets you review what a plugin is about to do before it runs, which matters most for AI agents that touch real systems.
- You now get three coding agents out of the box (Claude, Codex, and Gemini), and ACP channel binds put Codex inside the chat you're already in.
- If you run WhatsApp, Discord, Telegram, or iMessage bots, this update fixes several annoying bugs.
- Before you update, check for Qwen portal OAuth and old config keys.
FAQ
What's new in OpenClaw 3.28?
OpenClaw 3.28 adds native Grok X search, MiniMax image generation, a plugin approval system, ACP channel binds for Codex, a bundled Gemini CLI backend, apply_patch on by default for OpenAI and Codex models, and Slack file uploads. It also fixes the WhatsApp echo loop and several Telegram, Mistral, Discord, iMessage, and image analysis bugs.
Do I need a plugin to search X with Grok in OpenClaw?
Not as of 3.28. Grok X search is built into the responses API and auto-enabled. You turn it on during onboarding.
How does the OpenClaw plugin approval system work?
Plugins can now pause before running a tool and ask you to approve it. You approve with a button in Telegram, an interaction in Discord, or a slash approval command.
What are the breaking changes in OpenClaw 3.28?
Qwen portal OAuth has been removed, so you need to move to a Model Studio API key. Config migrations older than two months have also been dropped, so old legacy config keys will fail validation until you update them.
Transcript
Read the full transcript
0:00Openclaw 3.28 just came out and here is everything you need to know. I need to tell you something before we even get into the features. This is the fifth major update this week. That's five updates in 7 days and I don't know any other platform on the planet that is shipping this fast right now. It's one of the reasons I keep going all in on Openclaw building my entire content pipeline through it because when a team ships like this, you know, they're not slowing down. So if you're building with AI agents, if you're running automations, or if you're trying to get your business plugged into Openclaw, you need to pay attention to these updates. This is why I'm quick to release them, break them down as far as features and how it can help you take
0:45the next step in building your first AI automation because every single one of these updates makes the platform better and 3.28 is no exception. There's some really big stuff to cover here. I want to touch on the key points. So let me just run through what we have. The headline feature for me that sticks out the most and the one I'm most excited about is Grok X search. So native integration, built right in. So what does this mean for you and I? It means Grok, the X AI model, can now search X, search Twitter natively. No plugin setup, no workaround, no extra configuration. You just turn it on during the onboarding and it works. And here's why this matters so much, especially to me and you. We use Grok
1:31for trend scouting every single morning. Me, my first agent that fires off at 5:00 a.m. uses the Grok X API to search trends. So our Ghostface agent starts at 5:00 a.m. scans X for trending topics in the AI space. So it pulls that data and it feeds it to the rest of the pipeline and the rest of the agents. But before this update, that required a plugin. You had to set up the search tool manually, make sure it was connected, make sure it didn't break, which happens, and now it's first class. This happens already during the onboarding. So it's built into the responses API. It's auto enabled. So our Ghostface agent just got an upgrade without me doing any other work. If you're doing any kind of intel
2:16gathering, I think X and Reddit are the best places to start. You know, if you're doing any kind of trend analysis or any kind of content research, this is a massive upgrade for you. You can now have your agent search on X in real time and pull back what people are actually talking about. Not yesterday, not stuff from a month ago, but like literally right now and it works right out the box. So that's the feature number one that I'm fired up the most about and makes a huge impact on my setup. Honestly, that alone would have been enough for pretty solid upgrade since the last upgrade just literally came out 3 days ago, but they kept on going. Feature number two is Minimax image generation. So Minimax just dropped their image 01 model and Openclaw now supports it natively. You can generate
3:01images, you can edit images, you can control the aspect ratio. This is another option besides kai.ai for image generation, which is just a marketplace for video and image models, which is what I use. Um it is pretty cool, you know, now that you have choices, you don't have to go down that route. You know, I've been using kai.ai for our thumbnails and our UGC content and it works great. But having another model in the mix means you can compare the outputs. You can pick the best one. You can use different models for different use cases. More options is always better when you're building content pipelines and the fact that it's built right into Openclaw means you don't need another API, another service, another dashboard and it's just there. I like that. Feature number three and this one's a
3:46big one for anyone who cares about security and control. Plugin approval system. So here's what this means. Before this update, when a plugin wanted to execute a tool, it just executed. No questions asked. Now, plugins can pause, they can stop and say, "Hey, I'm about to do this thing. Do you approve?" And you get a button in Telegram and you get an interaction in Discord or you can use the slash approval command. That's a pretty nice feature for people running AI agents that have real access to real systems. Like for example, you know, if your agent has access to your email or like your database or social media accounts, you probably want to check in place just in case. You probably want to see what it's about to do before it
4:31actually does it and now you can. It's built right into the platform. You get to decide what level autonomy your agents have, which is actually how it should be. 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 Openclaw 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 can book a call with me, we can share screens and I can help you get Claude code or Openclaw running on your machine. That's it, no fluff. I built this community because
5:17watching 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. the link in the description down below. Get in now before the price goes up. Feature number four is ACP channel binds and this one is one of those features where when you see it, you're like, "Why didn't this exist before?" It's like common sense. But you know, basically you can now turn any current chat into a Codex backend workspace. So you could just type slash ACP spawn Codex bind and that's it. No child thread needed, no separate workspace. Codex right in your DMs, right in your Discord channel and right in iMessage
6:03conversations. We've been using Codex for a lot of our coding work and previously you had to spin up a separate thread for workspace to use it. Now you can just bind it to whatever conversation you're already in. So if you're already chatting with your team on Discord, you don't need to spin up a quick coding agent. You just do it right there. No context switching, no extra steps, which I think is pretty cool. Then the fifth feature is the Gemini CLI backend. So now Openclaw bundles Gemini CLI support alongside Claude CLI and Codex CLI. That means you have three coding agents available right out the box. Claude, Codex and now Gemini and now you can pick whichever one works the best for your task at hand. I think this is really cool because it gives you more flexibility. Some tasks handles better,
6:50some tasks Codex handles better and now Gemini's in the mix too, which I think is better for front-end design. That's just my opinion. Competition between models is good for all of us. It makes everyone better and having all three available without any extra setup, that's the Openclaw way, right? Feature number six is OpenAI apply {underscore} patch is now enabled by default for OpenAI and Codex models. This means better code editing out of the box. If you've been using Codex for coding tasks, the patches it applies to your files are going to be cleaner, more accurate and better than ever. It's one of those small changes that makes a big difference in your day-to-day workflow. The seventh feature is a Slack file uploads. So this is a fix that a lot of people have been waiting for. Explicit
7:37upload file action for Slack channels and direct messages. Files just work now. If you've been trying to have your agent send files through Slack and running into issues, that's over with. It's fixed and now it actually works. And feature number eight on the new addition side is Minimax model catalog cleanup. They trimmed it down to M2.7 only, removed the legacy M2, 2.1 and 2.5 models gone. Cleaner catalog, less confusion. You don't have to wonder about which Minimax model to pick anymore. It's just default 2.7, the latest one. Clean, fresh out the kitchen. Let's talk about the fixes because honestly, some of these fixes are just as important as the new features. The first fix is the WhatsApp infinite echo loop. Oh my goodness. So
8:22here's what's been happening. If you've had a WhatsApp bot running on Openclaw, the bot's own replies were being reprocessed as new incoming messages, so annoying. So it would reply, then see its own reply and then reply to that and then see that reply and then it just keeps on going forever and you're thinking you're going crazy. Infinite loop. So now your WhatsApp bot would just sit there talking to itself over and over. Not good. Now, we're not running WhatsApp, me personally, I'm using Telegram, but I know people were complaining about this on X all day, every day this week and now that's all fixed. And I know a lot of you are running WhatsApp through Openclaw. So that should be good to go. So if you've hit this bug, that must have been really frustrating. I'm sorry for you, but now your bot is now fixed and that nightmare
9:09is over. So completely fixed, check it out, download the update. Telegram message splitting is fixed number two. So when Openclaw sent long messages through Telegram, it was splitting them mid-word. So like it would just chop a word right in half and put the second half in the next message. Not amazing, but now it splits at word boundaries properly. So it's a small thing, but it makes a big difference in how your bot's messages look and feel, more professional and you don't have to wait for the next message to come through. The third fix is Mistral 422 errors. If you were using the official Mistral API, you might have been getting these random 422 errors with empty body messages. That's all fixed. It runs and works properly now. No more mysterious failures. Fix number four is Discord
9:55reconnect fix. This one was so annoying. Stale gateway sockets and poisoned resume states were causing reconnect loops. So, your Discord bot would lose connection and then it would just keep trying to reconnect over and over and it would never actually succeed. That is all fixed. Clean reconnects now. So, fix number five is iMessage reply lag leak. So, basically those bracket to bracket reply tags were leaking into the actual develop delivered text. So, the person on the other end would see this ugly technical tag in their iMessage and it's not a good look, but that's all fixed now. There's clean messages in iMessage. Fix number six is image analysis has all been restored. The generic fallback for
10:41providers like Open Router and MiniMax is working again. So, if you were having trouble with image tools, with analyzing images, with generating responses based on images, that's back online. If you're struggling to keep up with content, 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, 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 [music] and the content's already done. I spend maybe 15-20 minutes reviewing and approving them and
11:28I move on with my day. [music] It works for any niche, fitness, finance, 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 [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. Now, I want to touch on some breaking changes real quick because this one is super important. If you're using Kwen or the portal OAuth, that has been removed. You need to migrate to a Model Studio API
12:14key and config migrations older than two months have been dropped. So, if you've been really on the old legacy configuration keys, they're going to fail validation. You need to update them. Here's what I want you to take away from all of this update. Five updates in seven days. Monday we got 3.22, Monday night 3.23, Wednesday 3.24, and now Saturday 3.28. The Open Claw team is shipping at a pace that I've never seen from any platform in the space. Anthropic is a close second, but I think they're just copying everything Open Claw does at this point. I mean, come on now. They are. And I've been watching it closely every single update. I mean, come on. It's like they ship a feature, Anthropic ships a feature. It's hard to keep up. Every update on Open Claw makes the platform more powerful,
13:00more stable, more capable. And the thing that really gets me excited is the direction that it is all going. Native search integration with X through Grok. I mean, come on. Image generation is already built in. Plugin approval for security. Three coding agents out of the box. These aren't random features, y'all. This is a team that knows what builders need and they're delivering it fast. If you're building with AI and you're not on Open Claw yet, what are you doing? I genuinely don't know what you are waiting for. God help us. It is one of the most fascinating times to be alive, to be building AI agent startups, AI agent services, to be able to automate your life, buy back your time. You got to get in on the action. Open Claw's free and this is where the action
13:45is. This is where the tools are. This is where things are moving the fastest. Speaking of building, you know, if you want to learn how to set all this up. Maybe you're on the sidelines. You're like, I don't know where to No idea where to start. Well, if you want to learn how to build your first AI agent and your first pipeline and how to actually ship products using these tools, that's what we do inside the community. We've had 168 members sign up in the last 16 days and we have six live calls a week. We have a community of over 160 members and we're all doing this together. We're all learning from each other. You know, if that sounds like something you need in your life and you want to be held accountable and to, you know, kind of take the next step in your AI journey, I'll put the link to it down below. Whether you join or not, it doesn't matter. I just wanted to break this down for you to be helpful to you
14:31to understand what this update means for you as a builder and I hope it gives you a clear picture of what just shipped and released right now and giving you more clarity on how to use and take advantage of these features. Because at the end of the day, it's not about knowing about what the features are. It's about actually using them to build something useful. So, if you haven't subscribed already, please hit the subscribe button. Hit the notification bell. We are usually release four to six videos a day, but I've had family come into town, so I've been kind of taking it easy. So, we'll see you in the next one. You guys have a blessed night. Bye-bye.
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 SkoolKeep reading
OpenClaw 4.23 Update: GPT Image 2 via Codex OAuth and Context Forking
OpenClaw 4.23 lets GPT Image 2 run through your existing Codex OAuth login, so you no longer need a separate OpenAI API key. It also adds context forking for sub-agents, per-call timeouts on media tools, more than 15 security fixes, and a catalog fix so GPT-5.5 works.
OpenClaw 5.2 Update: Cron Jobs, Context Safety, and Security Fixes
The OpenClaw 5.2 update is about reliability, not flashy features. Cron jobs run in their own lane, Codex and GPT-5 runs compact before they hit the context window, and approvals and secret checks are tighter. It also adds Discord voice guardrails and better model routing.
OpenClaw 5.4 Update: Google Meet Voice Agents and Reliability Fixes
OpenClaw 5.4 (2026.5.4) is mostly about voice and reliability. Google Meet and Twilio calls can now run through a realtime Gemini voice bridge with paced audio and proper barge-in handling. The release also fixes problems on Windows and Docker, in Codex transcription, plugin loading, channels, cron diagnostics, command safety, and memory.
