OpenClaw 4.24: New AI Voice + Browser Updates

OpenClaw 4.24 just dropped and your AI agents can now join Google Meet calls. They can make and handle phone calls using your full agent, not some stripped down version. DeepSeek V4 Flash and V4 Pro are now built in and the browser automation got a serious upgrade. This is a packed release.

Let me show you what changed and why it matters for you. Let's start with the one that blew my mind. Google Meet is now a built-in plugin inside OpenClaw 4.24. So your AI agent can now join a Google Meet call.

It connects with your Google account. It can join meetings through Chrome or through Twilio or for audio, right? And whilst it's in the meeting, it can pull in the full power of your OpenClaw AI agent. So all your tools, all your context, everything.

But it goes further than just sitting in the call. Your AI agent can export meeting artifacts. So recordings, transcripts, smart notes, attendance lists, it outputs everything. It's clean files that you can actually use.

And if you need to look up past meetings, there's a history scan that pulls conference records. Here's why this matters. If you're running a business, think about how many meetings you sit through every week. Now imagine your AI agent tracks those calls, joins them, takes notes, and tracks who said what and gives you a clean summary afterwards.

Or imagine you can't make a meeting and your agent joins for you, listens, and then reports back for you. For agencies running client calls, this could save hours every single week. And there's a recovery system built in as well. If your agent already has a Meet tab open and something goes wrong, a timeout, a disconnect, et cetera, it can recover that tab instead of opening up a new one.

It can detect login screens and permission blockers and tell you exactly what needs fixing. So it doesn't just silently fail. It tells you what's in the way. Now here's the other big voice update.

Voice calls got a major upgrade in OpenClaw 4.24. Before this, when your agent was on a phone call, it was running a lighter version of itself. It could talk, it could listen, but it couldn't access all the tools. It couldn't dig into its full knowledge base.

It was limited. That changes now. OpenClaw 4.24 adds something called Agent Console. So when your agent is on a live phone call and hits a question it needs some help with, it can pause, ask the full OpenClaw agent with all its tools and context, and come back with a proper answer.

Same thing works in Google Meet and in the talk feature through your browser. So if a customer calls and asks something that requires, for example, looking up data, or checking a file, or running a tool, your agent can handle it mid-call. It doesn't have to say, I'll get back to you. It gets the answer right there.

And there's a new Gemini Live voice provider. Google's Gemini can now power the voice side of your calls with two-way audio and tool support. So you've got options. OpenAI Realtime for browser-based voice, Twilio for phone calls, Gemini Live for another voice path, and more options meaning you can pick what works best for your setup and your budget.

For the voice call setup itself, there's a Smoke Test command. Before you make a live call, you run a dry test to check if your Twilio connection is actually working, if your provider is ready, if everything is wired up, and so you don't have to find out something's broken when a real customer calls. You can test for it, which is smart. Now let's talk about DeepSeek v4.

Both v4 Flash and v4 Pro are now built into OpenClaw 4.24, and here's the key detail. v4 Flash is now the default model when you're setting up DeepSeek for the first time. That's a big deal. DeepSeek has been one of the most cost-effective models out there.

v4 Flash is their fastest option. So if you're just getting started with OpenClaw and you don't want to pay for an expensive API call whilst you learn the ropes, for example, the default model is now one that's fast and cheap. You can always switch to something more powerful later, but it just got way more accessible. v4 Pro is there too when you need a bit more muscle and the thinking system works properly now.

Before, if you were mid-conversation and switched to a DeepSeek v4 model, the history replay could break because the model expected certain data that wasn't there. That's fixed. You can switch models mid-session without errors. If you want help picking the right model for your business, DeepSeek v4 Flash for speed and cost, GPT 5.5 for power, and Claw for personality, right?

That's what you want to focus on. Now if you need more help on this sort of stuff, you've got the AI Profit Boarding. You've got four coaching calls every week where members bring their OpenClaw setups and we go through this sort of stuff together. Which model for which task, how to set up voice agents that handle customer calls, how to get your AI into meetings.

We've got a 30-day OpenClaw roadmap that walks you from just installing OpenClaw to actually bringing in leads and customers. You get daily tutorials on it with every new feature, including DeepSeek v4 and the voice call upgrades in OpenClaw 4.24. There's 2,900 business owners in there and a member map so you can find users near you. Link in the comments description or go to the aiprofitboarding.com to get access.

All right, browser automation. This got some serious upgrades in OpenClaw 4.24 too. First of all, coordinate clicks. So your agent can now click on exact spots on a web page using pixel coordinates.

Before it had to find elements, buy their labels or roles, which works most of the time but breaks on weird websites that don't label things properly. Now it can just say, click our position 450, 330 and hit exactly what it needs. If you're using browser automation to help you, for example, automate a lot of your tasks or interact with web apps for business, etc, this makes your agent way more reliable. Second, the default time budget for browser actions went up to 60 seconds.

Before, some actions would just fail because the page was slow to load. Not because something was wrong, it was just slow. Now your agent has more patience and you can set custom timeouts per profile, which means one browser setup can be fast and snappy whilst another can take its time with heavy pages. And third, tab recovery got better.

If your agent's browser crashes or if you restart OpenClaw, stale browser locks from the old session used to block the new one from launching. OpenClaw 4.24 detects those locks, clears them and then retries so you don't have to manually clean up after a crash. And there's a new browser doctor command. If something's not working with your browser automation, you can run diagnostics that tell you exactly what's wrong.

For example, like a missing chrome install or wrong permissions or bad profile setup. Instead of just guessing, you actually get a clear answer. Now let me hit the channel fixes because there are several good ones here. Telegram had a phantom error that's been bugging people.

After your agent already sent a reply, Telegram would sometimes show a message saying agent couldn't generate a response, even though it just did. OpenClaw 4.24 deletes that phantom message. If a reply already went through, the error doesn't show. Telegram also got better at showing tool progress.

When your agent is working on something and sends status updates, like for example searching or thinking, those used to sometimes trigger weird markdown formatting or discord style mentions that made no sense in Telegram. That's cleaned up. Slack got a bunch of fixes too. The big one was message ordering.

So when your agent sends multiple messages quickly in Slack, they used to arrive out of order sometimes. That made conversations confusing. OpenClaw 4.24 now sends one message at a time in sequence, so they show up in the right order every time. Slack thread handling got tighter too.

If your agent is replying inside a thread, it stays in that thread. Before, sometimes parts of the reply would leak out into the main message. That is fixed. And if someone uses also send to channel on a thread reply, your agent now probably sees that message before it would ignore it.

WhatsApp got voice note transcription. So when someone sends your agent a voice message on WhatsApp, OpenClaw 4.24 now transcribes it before passing it to your agent. So your agent reads the text version of what the person said. It doesn't need to handle raw audio.

That's huge for customer facing WhatsApp agents, for example. People love sending voice notes. Now your agent can actually understand them. WhatsApp also got better at delivering media from tools.

So if your agent generates an image or a file as part of its response, that media gets delivered properly through WhatsApp. Before, tool generated media could get dropped fixed. On the memory side, there's a useful improvement. So when your agent searches his memory, it now shows you how much of the result came from the text search versus the meaning-based search.

Why does that matter? Because it helps you understand if your agent is finding things by exact words or by understanding what you meant. If you're tuning your agent's memory for better accuracy, that breakdown is gold. And for anyone running OpenClaw on a machine without powerful cloud credentials, the local system embedding system no longer requires a heavy AI package to be installed by default.

It only loads that package if you specifically set up local embedding. So OpenClaw starts faster and uses less space out of the box. The memory dreaming system, which processes and organizes your agent's memories over time, got decoupled even further. It now runs as its own thing, completely separate from the heartbeat system.

And there's a fix for a real problem. And before, heartbeat prompts were accidentally leaking into normal conversations. Your agents would sometimes act like a user message was a heartbeat check and respond with a silent acknowledgement instead of actually replying. That's gone.

Heartbeat stays in its lane. Your conversations stay normal. MCP, which connects your agent to external tools and services, got better cleanup. One-shot MCP connections used to stick around after they were done eating up resources.

Now they shut down when the job is finished. And if an MCP session gets stuck, there's an idle timeout that cleans it up automatically. The session system got more stable too. If OpenClaw restarts mid-conversation, it now picks up where it left off more reliably.

The restart continuation system hands off the work to a delivery queue before cleaning up. So even if the restart is messy, your conversation doesn't get lost. And there's a compaction fix worth mentioning. When your agent's conversation gets long and needs to be compressed, the summary system was building summaries on top of old summaries, like a game of telephone, where the message gets more distorted each time.

OpenClaw 4.24 now recreates summaries from the actual conversation instead of stacking them. So compressed conversations stay accurate. There are over 139 contributors on this release. 139 people shipping code for OpenClaw 4.24.

This is the largest contributor count I've seen on a single release. The community behind the tool is massive and it's moving fast. And look what's happening here. Your AI agent can now join your Google Meet calls.

It can handle phone calls with full tool access. It can browse the web and click on exact spots. It understands WhatsApp voice notes. It works with DeepSeek v4 out of the box for almost no cost.

Every single update lowers the barrier and raises what's possible. A year ago, if you wanted the AI that can join meetings, handle calls, browse the web, manage messages across five platforms, and remember everything, you'd need a team of developers and a serious budget. Today, one person with OpenClaw can set all of that up over a weekend. And it keeps getting easier.

The people who start now are the ones who will have a six-month head start on everyone that waits. And that head start compounds. The agent gets smarter. Your workflows get tighter.

Your business runs faster. Come join us in the AI platform. This week's coaching calls are going deep on stuff like voice calls, OpenClaw 4.24, how to set up voice agents, how to get your AI into these sort of meetings, and how to use DeepSeek v4 Flash to keep costs low whilst you scale. We've got the full 30-day OpenClaw roadmap, daily step-by-step tutorials on every feature as it ships, a prompt library built around OpenClaw AI agent workflows.

2,900 members selling this sort of stuff up right now on a member map, so you can find local people using AI agents like OpenClaw, Hermes, Agent Zero, etc. So you can get help, share playbooks, meet up, etc. Link in the comments description or go to the AIprofitboarding.com. If you want to update, just type in update inside the chat, or you can click the update if you're using the gateway on OpenClaw.

Bear in mind, you should always make sure you create a backup before you update to the new version. And bear in mind as well, these are new features, so they're bound to be buggy. But this is the way things are going, things are improving, and it's looking really cool as a project. Thanks for watching.