OpenClaw 5.24: New Update Just Dropped…

OpenClaw 5.24 just showed up on GitHub as a pre-release beta and there's a feature in here that could save you hours every single week. Your AI agent should now be able to listen and join a Discord voice call, listen to the whole conversation and write down everything that was said with automatic meeting notes, no typing, no extra tools. Your agent just listens and captures it all. On top of that, you should be able to approve your agent's actions from iMessage with a single thumbs up emoji.

Voice calls are supposed to get way more interactive. You'll be able to steer your agent in real time whilst it's still talking and there's an image upgrade that should save you tokens on every single conversation. I'm going to walk you through everything coming in this update, what it can mean for your business and the one setup that ties all of this together. Stick with me to the end, let's get into it.

Now, before we dig in, this is a pre-release beta. It's on GitHub but it's not stable yet. Given how buggy some recent updates have been, I would not put this on your live setup just yet but I'll show you how to test it late, but I'll show you how to test it safely at the end and the features coming in this are worth understanding now so you're ready when it does get stable. Let me start with the meeting notes because if this works well, it's a game changer.

Right now, when you're on a Discord voice call, for example, a team meeting, a client call, community sessions, everything that gets said kind of disappears the moment the call ends, unless someone took notes and nobody takes good news, not nudes, notes during a call. You're busy talking, right? So OpenCore 5.24 adds a meeting notes plugin. Your agent should be able to join the Discord voice call and record what everyone says.

It captures who said what, writes it all down and you can pull up the full transcript when the command line is ready whenever you need it. You can set it to auto start as well so every time a voice call begins, your agent should just start capturing. You don't have to remember anything, you don't have to click anything after the call, you've got a full record. You can ask your agent to summarize the key points, pull out action items and find what someone said about a specific topic all from the transcript.

For anyone who runs team meetings, client calls or community sessions, that could save you hours. No more trying to remember what was discussed, no more can someone send me the notes, your agent should already have them. Now iMessage approvals, simple but powerful feature here as well. So when your agent wants to do something that needs your permission, for example run a command, make a change, access something sensitive, it asks you first.

Before this update you had to type out a response or use a command to approve it. OpenCore 5.24 adds thumbs up approval. So thumbs up means yes, thumbs down means no. One tap from your phone whilst you're doing something else.

If your agent sends you a dozen approval requests a day, that's a dozen 30 second interruptions turned into one second taps. WhatsApp already had this, now iMessage should too. And also voice calls got more interactive. When you're talking to your agent through discord voice or the web dashboard and your agent is working on something, you should now be able to check its status, cancel what it's doing, redirect it or queue up the next task.

Or whilst it's talking to you, before you had to wait for the agent to finish, before you could interact again, now you should be able to steer it in real time like talking to a real assistant. Discord voice also added wake name gating. So your agent should only respond when someone says its name first. So in a busy voice channel with lots of people talking, your agent stays quiet until someone addresses it directly, which means no more random replies to conversations that weren't even meant for the agent.

Images also got smarter, so your agent should now adjust image quality based on what AR model you're using. Some models handle detailed images well, others waste tokens on detail that can't process. OpenClaw 5.24 should pick the right compression automatically so that you save tokens, your agent works faster and you can choose between token efficient, balanced or high detail if you want to override it. The gateway also starts up faster now, so they've reorganized how things load, so your agent should be ready sooner after a restart.

Again, I want to say should because at this point we've seen so many updates where things are promised, but they don't work that smoothly in reality. So that's why I always apply should instead of does, just so you know. Add-on data, channel info and settings that don't change between restarts are now saved and reused instead of being rebuilt from scratch every time. And if your agent has been slow to come back after a restart, this should actually help.

The chat search got better in the dashboard too, so you can now search through your conversation history and load older chats without the whole list slowing down. If you've had your agent running for months and have hundreds of sessions, finding an old conversation should be much easier now. Grok web search should work before your subscription login detail now. So basically what this means is you can log in with OAuth.

I've been testing this, actually built it into the agent operating system we have this morning, and basically you can generate videos, you can generate images, you can do Grok and Twitter search, and you can do text-to-speech very, very quickly with Grok 4.3. It seems to work the smoothest out of all the OAuths I've seen, and I honestly prefer this versus something like Chat Cheapity 5.5 when you do the OAuth with OpenClaw. So before, even if you had Grok OAuth set up before, web search would still need a separate API key. Now it should use the same login for everything, which means models, images, and search.

The install package got smaller as well, so they removed documentation images from the download, so you have the same tools, the same features, just a faster download. Now if you're on a slow connection or installing on a small server, that helps. And if you want to get the most out of OpenClaw, the meeting notes, the approvals, the voice controls, all of it, check out our agent operating system inside the AirProfit Boarding. It actually connects OpenClaw, Claude, and Hermes into one dashboard with shared memory.

So your meeting notes feed into the same system all your agents use. Action items from calls become tasks your agents can work on. Context and voice conversations make every agent smarter. Everything connects.

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 building with AI agents right now inside there. Link in the comments description or go to the airprofitboarding.com.

Now let me cover some of the fixes as well because there are some improvements on stability in here too. So telegram format topics should work better. There was a problem before where one slow topic would block all the other topics in the same group. So your agent will get stuck on one conversation and everything else would wait.

Now each topic should run on its own lane. One slow reply basically doesn't hold up the rest. There's also a fix where follow-up messages in a topic would fail because they inherited a cancel signal from a previous message. So you'd send a message, a second message in the topic, and it would silently fail because the system thought it was cancelled.

That should be fixed now. And Anthropx Claude models have an issue where image support would disappear. So if your local settings had stale data about what Claude can do, it would mark Claude as text only as a model. Even though you can integrate voice image models inside there alongside Claude.

So that should be fixed. DeepSeq had a problem where certain tool setups would get rejected. If your tools used a specific kind of complex setting structure, DeepSeq would refuse them. OpenClaude now cleans those up before sending them.

Should work now. Memory search should be faster and less disruptive too. So before a big memory search could freeze your whole system for several seconds. Now it runs in small batches and lets other things happen between batches.

So your agent should be more responsive even during heavy searches. Skills had a refresh as well when you change them. So before if you updated a skill whilst your agent was running, the agent wouldn't notice until you started a completely new session. Now it should pick up changes right away and the update system should handle freshly published packages better too.

There was an issue where running OpenClaude update right after a new version dropped would fail because the package registry hadn't cleared it up yet. That should be bypassed now for OpenClaude managed installs. Now to handle this. This is a pre-release beta.

Don't put this on your live setup. We've seen too many updates break things recently to rush into a beta. Now if you want to test it you can backup first, run OpenClaude backup create. Then you can run OpenClaude update dash dash channel beta.

Test your channels, test your voice, test your messages. If something breaks just roll back to the stable version. Know your current version as well so just write it down. If things go wrong you know exactly where to go back and wait for the stable release before running this for anything important.

Let other people find the bugs first because you don't want to waste your time. If you want your OpenClaude setup to actually save you time every day, check out the agent operating system inside the iProfile bottom. It turns OpenClaude, Claude and Hermes into one system with shared memory and one dashboard. So you can feed in meeting notes from calls into your agents.

Approvals happen faster. You've got voice conversations to give context to everything else. Every update makes the whole system more powerful and you get the zip file, the prompts, the obsidian memory setup. You also got a studio built in.

Now we've OpenClaude for generating videos, images, twitter search, and voice text-to-speech generation now too. Along with I've set up a workspace on there so you can preview everything you've previously created. So like all the apps and the tools and if you've built any games or anything like that, you can actually preview and use them inside the agent operating system. On top of that, we have coaching calls where you can build stuff like that together with us.

We've got 3,000 members, daily tutorials, a 30-day roadmap, and a map to find people near you. Link in the comments description or go to theairprofilebottom.com to get assets. Before you touch anything, run OpenClaude Backup Create. Backup first, decide second, and I'll see you in the next one.

Thanks for watching. Bye-bye.