OpenClaw 5.18 Update Just Dropped...

追加: 2026-05-19

[00:00:00]Open 5.18 just went live and this one has a feature I've been waiting for. You can now talk to your AI agent out loud from the Android app with a real voice in real time. Your agent hears you, thinks, and talks back like a phone call with an assistant who has access to all of your tools and all of your data. On top of that, Grock OF is now more stable. If you have a super Grock subscription, your agent can use Grock without an API key. So, there's no extra cost there. You can just log in and your agent's browser just got smarter, too.

[00:00:34]It can now see pop-up windows and answer them instead of getting stuck. I'm going to walk you through everything in this update, what it means for your business, and the setup that makes all of this 10 times more powerful. Let's get into it.

[00:00:48]Let me start with Android talk mode because this changes how you interact with your agent when you're on the go.

[00:00:55]Open claw has always worked from your phone. You can message your agent through Telegram, WhatsApp, Discord, whatever you use. But talking to it out loud, that's new. Open Core 5.18 adds realtime voice to the Android app. You open the app, you tap talk mode, and you just start speaking. Your voice goes to the agent in real time. Your agent thinks and it talks back out loud through your phone speaker or headphones. And this isn't the old way where your voice gets converted to text sent to the AI and then the reply gets converted back to speech. That's slow.

[00:01:32]There's a gap and it feels robotic. This is real time. So your agent hears you as you speak. It starts responding before you've even finished sometimes. And even if you interrupt it, it stops and listens. Just like talking to a real person, your agent still has access to everything whilst you talk. It can use tools. It can search the web. It can check your data. It can run commands.

[00:01:52]And you see the transcript on screen as you go. So, you have a record of everything that was said. Think about what this means. You're in your car. You say, "What's on your schedule today?"

[00:02:03]The agent checks and reads it back to you. You say, "Cancel at 3 p.m. and message the clients. It's done.

[00:02:10]Handsree." Or let's say for example, you're walking between meetings. You say, "Summarize what my team discussed on Telegram today." Your agent pulls in context from the group chat and reads you the highlights whilst you walk. This turns your phone from a place where you type at your agent into a place where you talk to it. Like having a real assistant in your pocket. Now Grock, this was in the beta and now it's stable. If you have a super Grock subscription, you can just log in with your account and your agent uses Grock with no API key, no developer portal, no extra bills. So, it's one login that's done. Before this, you had to buy separate APIs from XAI on top of your subscription. Two payments for the say AI. Right now, it's just your subscription. So, your agent gets access to Grock's models, Grock image tools, and Grock's real-time information along with Grock speech and Grock video.

[00:03:03]That's all included. If you've been paying for Grock but only using it on the Xiaoa website, for example, in Twitter, or this saves you straight away. Connect it to OpenCore and your agent can use everything Grock offers because you're already subscribed to it.

[00:03:17]The browser just got smarter, too. And this one fixes a problem that's been quietly causing failures for anyone using browser automation. When your agent uses a browser to do things, for example, like fill out forms, click buttons, navigate websites, it sometimes hits a popup, a cookie consent box, for example, a login prompter, confirmation dialogue. Before this update, your agent actually couldn't see these pop-ups. It would just get stuck. The page wouldn't respond because it was waiting for someone to click a button on the dialogue, and your agent had no idea why. Open Claw 5.18 fixes that. So your agent can now see when a dialogue is blocking the page. The browser snapshot shows it and your agent can answer it.

[00:03:59]Click okay, dismiss it, type into it, whatever the dialogue needs. So workflows that used to fail silently because of a pop-up now work. And if you're using your agent to, for example, get data, fill out forms, book appointments, or do anything in a browser, this removes a whole category of random failures. Telegram got a bunch of reliability fixes. If you use forum topics, which are like organized threads inside a Telegram group, there was a problem where your agents replies would end up in the wrong place. Someone would ask a question in a topic and the agent would reply to the main group instead of the topic or generated images and videos would land in the base chat instead of the topic where they were requested.

[00:04:39]That's fixed. So everything stays in the right thread. Now, there's also a fix for scheduled messages. So, when your agent sends a message with a link through a scheduled announcement, the link would show up as raw code instead of a clickable link. People would see ugly code tags instead of a clean link.

[00:04:59]That's fixed. So, the link looks normal now. And if you had requirement mention turned on, meaning your agent only responds when someone tags it, it was still trying to download images from messages that weren't meant for it and failing. And sometimes sending error messages to the group, which is fixed now. It ignores media from messages is not supposed to respond to. Now, if you want to get the most of these features, the voice mode, the Gro login, the browser fixes, check out our Aentic operating system inside the AR profit boardroom. It's a full operating system I've built that connects OpenClaw, Claude, and Hermes into one dashboard.

[00:05:37]So, your agents share one memory. They know your goals. They know your business. So, when you talk to your agent on Android already has full context for everything. When Grock answers a question, it pulls from your shared knowledge base. Everything compounds. Everything saves you more time the longer you use it. You get the full zip file, every prompt, the Obsidian memory setup, and coaching calls where we walk you through the whole thing. There's 3,000 business owners in there right now building with AI agents like OpenClaw, Hermes, and Claude. Link in the comment description or go to the arprofitballing.com to get it. Now, let me cover some more changes that make your setup more stable. The gateway starts up faster. They reorganized when things load during startup so your agent is ready sooner after a restart. Channel connections and plug-in services now overlap instead of waiting in line. If your agent takes a while to come back after a restart, this helps. Discord Voice got a fix for opening up real-time sessions. There was a problem where your agent would stop hearing follow-up messages in voice channels. Someone would say something, get a reply, then say something else, and nothing. The agent stopped listening after the first turn. That's fixed.

[00:06:44]Conversations keep flowing. Now, the codeex and opening integration got smoother. There were issues where image attachments sent through Discord wouldn't reach the AI model properly.

[00:06:56]Your agent would get the message but not the image. That is fixed. Images now get processed and sent to the model correctly. Sub agent handling got more reliable too. If you're running multiple agents that talk to each other, like a research agent that hands off results to a writing agent, the handoffs were sometimes dropping results. The parent agent will clean up before the child agent finished fixed. Results don't get lost anymore. Plug-in installs get more stable as well. Now, so there was a longstanding issue where installing or updating one plugin could break a different plugin that was already installed. The shared dependencies would get tangled. Fixed. Plug-in installs are now isolated better. The Mac app got a redesign on the settings pages. cleaner layouts, better navigation, and it now prefers direct connections over SSH tunnels when both are available, which makes the whole experience faster. Tool descriptions got shorter. The built-in tools, for example, like messaging, scheduling, web search, image tools, all of them used to have long, detailed descriptions that ate up your context window. Now, they're shorter. Same information, just fewer tokens being used, which means your agent has more space for your actual conversation and your actual work. And there's a quality of life fix too. When something goes wrong, error messages now tell you what happened and how to fix it. Instead of vague errors, you get specific guidance.

[00:08:20]Which command to run, which setting to check, which documents to read, small things, but it saves you a lot of time.

[00:08:26]Now, should you update this is a stable release. It went through multiple beta rounds before shipping. The community reaction so far has been positive.

[00:08:36]People are saying, for example, Groof actually works. The Android voice mode actually works. The browser dialogue fix is solving real problems. My advice hasn't changed. Right? If your setup works, backup first. Always run openclaw backup create before you touch anything.

[00:08:52]Know your current version. Test after updating. And if something breaks, roll back. The people who figure out AI agents. Now, whilst the tools are evolving fast, are going to be way ahead when everything settles. Every update you learn, every workflow you build, every problem you solve, it compounds.

[00:09:08]Now, if you want your open claw setup to actually save you time every day, not just be another tool you check sometimes, go grab the Aentic OS system inside the AR profitable boardroom. It turns openclaw, claude, and Hermes into one system with shared memory, shared context, and one dashboard you control.

[00:09:24]Your agents understand your business.

[00:09:26]They remember everything. And every new openclaw update like Android voice mode and Grog login makes the whole system more powerful automatically. I built it in one session and you get the zip file, the prompts, the obsidian memory setup and coaching calls where we set up together step by step. You got 3,000 members inside there so you can get help whenever you need to, daily tutorials, a 30-day road map for openclaw and a map to find people need. Link in the comments description or go to the arprofit.com to get access before you update. Run openclaw backup create backup first decide second. I'll see you in the next one.