OpenClaw 5.20: New Update Just Dropped...

OpenClaw 5.20 just dropped, and this one fixes a lot of stuff. You can see, for example, Discord voice follows you, doctor catches plain text secrets, model status explains surprises, and Windows install gets in stock. So I'm going to explain exactly what this means, how it works, what just happened, et cetera. So basically, this one fixes some stuff that's been quietly breaking agents for months.

So Discord voice that follows you into voice channels. You've got a new version of doctor that catches plain text secrets for your API keys, XAI you can finally log into without a local host callback, and Windows installs that stop freezing. So I'm going to walk you through everything in this update, what it means for your business, and the one setup that makes all of it 10 times more powerful. So let's talk about this.

First of all, your agent walks into the voice channel with you. What does this mean? Let me start with the Discord voice update, because this is one that a lot of people were waiting for. Essentially, OpenClaw has always worked with Discord.

You could message your agent in the text channel. It does the work fine. But what was missing was when you moved into a voice channel, your agent stayed behind. It was disconnected.

The conversation broke. OpenClaw 5.20 fixes that. So your agent actually moves with you hands-free. So now your Discord voice sessions follow configured users into voice channels, and your agent moves with you.

What does this mean? Basically, you have multi-user handoff. And essentially, this means voice sessions pass between configured users in the same channel. There's allowed channel checks.

Your agent only joins channels you explicitly approve. Dave encryption recovery, which means sessions keep working when Discord rotates. Encryption keys mid-calls. And there's real-time context, which means it has the identity of the user and the sole profile context, which now ships with voice sessions by default.

So you can hop into a voice channel with a teammate. You can say, for example, agent, check the leads from this morning and summarize the top three. And your agent is right there in the call doing it hands-free for you. And for community managers, for creators, for anyone running Discord operations, this changes the shape of the day.

Next up, Doctr now warns you before a leak. And this is a big one. So what does this mean? Doctr now warns you when your OpenCloud.json file stores plain text, secrets like API keys or sensitive headers, right?

And basically, before this update, you could have an Anthropic key or an OpenRouter key sitting in plain text in your config. OpenCloud never told you. And you'd find out when somebody else found out, which is no bueno, right? Now you can run OpenCloud Doctr and it flags every plain text secret before it becomes an incident.

So what does the Doctr now catch? For example, model API key sitting in OpenCloud.json by encryption, sensitive provider headers, so bearer tokens, anything secret-shaped, sandbox tool policies, so conflicts, hiding, MCP server tools you actually configured, and stale provider model configs, right? That might fail validations after upgrade, right? When they get cleaned by the Doctr fix.

So this is the kind of fix most users don't appreciate until it saves them. But if you have provider keys sitting in your config right now, you can run the Doctr as soon as you update and that will help you out. Additionally, what we have here is we don't need to wonder which model is currently running, right? So status actually got smarter inside OpenCloud too.

This fix is something that's been quietly annoying everyone. So when a session is pinned to a different model, then you configured default. You used to have no idea why. Maybe a fallback kicked in, maybe an agent override hit.

The status said nothing. Now the status actually shows you everything in one place. So what does status now tell you? For example, configured default models, what your config says you should run, selected models, what's actually running right now, reasons for the difference.

So for example, it could be a fallback, it could be an override, it could be a manual pin, and clear hints of what to do if you want to switch back to default and direct docs links. These are clickable and actually take you to the relevant page. So you can run OpenCloud status and the system tells you exactly which model you're talking to and why, which means there's no more guessing why the report came back in the wrong voice. Now also logging XAI into a remote server just got easy as well.

So if you've ever tried to authorize XAI on a headless server, you know the pain. The old login needed a local host browser callback, right? Which doesn't actually exist on a remote machine, right? So you ended up SSH tunneling ports just to get logged in, which is a lot of technical work for 99% of people who are not actually technical.

OpenCloud 5.2.0 adds device code AWARF for XAI. This is a new workflow. So you can literally just start the login on the remote server, the terminal shows you a short code, then step two, you can open XAI on your phone or your laptop, paste the code into the login screen, and step number three is you authorize, right? So it's the same flow GitHub Copilot actually uses for this, and it's the same flow Anthropic CLI uses as well.

So if you run OpenCloud on a VPS, a modal sandbox, or a Daytona workspace, this single fix removes one of the most annoying setup blockers of the year. Additionally, Windows installs no longer freeze at the starting setup. And this one is really for every Windows user who tried to install OpenCloud and watch the wizard freeze on a screen for 10 minutes. The installer now launches an attached process so the wizard's terminal rendering stays clean instead of being eaten by a detached parent, right?

Which means no more visible freezes. So managed updates also use the gateway service node consistently. This just means it's easier to use OpenCloud on an install. Before this, if you had multiple node installations on your machine, OpenCloud would sometimes silently switch which node it ran between updates, and it would hide the source of breakage.

That should be gone now, and if Windows install has been the thing stopping you from actually setting up, OpenCloud today is the day, my friends. So let's talk about the five headline features and here's what's changed side by side, okay? So number one feature is Discord voice channels. Before 5.20, agents stayed in the text channel when you moved to voice.

Now, after 5.20, agent follows you into voice channels with full context. Next up, API keys in your config. Before, sometimes they'd be sitting in plain text with no warning, and you would only find out when somebody else does. Now the doctor actually catches them before they become an incident.

Next up, when the wrong model runs, before it'd be silent and you'd just have to guess, oh, why has something changed? Why has it stopped responding the same way? Now the status shows why with a hint and a documents link. Additionally, XAI login on a remote server.

Before, SSH tunnel was required for the local host browser callback. Now it has device code off. You can just paste a short code in and you're in. And the fifth one is Windows install experience.

So before, the wizard was freezing at the starting setup on a black screen. Now it launches and it should be a much cleaner install. By the way, if you want all of this in one connected system with all of your AI agents, OpenCLAW is powerful on its own, but it's 10 times more powerful when it shares memory with CLAWD and Hermes and everything else. So this is what our agent operating system actually does for OpenCLAW.

OpenCLAW over here, we've got CLAWD, we've got Hermes. Antigravity is plugged in as well. We can connect to Notebook LM. We've got a full studio for SEO.

We can generate videos, images, text to speech, et cetera with Grok. This is an operating system, right? And so if you want to get the most out of these new OpenCLAW features, you can get it all inside the full operating system I built that connects OpenCLAW, CLAWD, and Hermes into one dashboard. So your agents share one memory inside OpenCLAW, as you can see right here.

So it's actually plugged into our Obsidian memory vault, which is pretty cool. And if you want the agent operating system zip file or 30-day roadmap for coaching calls a week on OpenCLAW, CLAWD, and Hermes, daily tutorials, 3,000 business owners you can connect with about AI automation, and remember maps you can connect with people in your city, join the AI Profitable Boarding, link in the comments description, or go to the AIProfitableBoarding.com. Now, there are a bunch more changes as well, just focused on making sure OpenCLAW is a lot more stable. So there are actually 84 commits in this release.

The headlines were five. The other 79 fixes, nobody is going to talk about, but these are the ones that quietly just help it make OpenCLAW more stable. Eight of the most important ones are sub-agent handoffs. So these no longer drop.

For example, the researcher agent, writer agent results are preserved. There's no more lost work. Scheduled tasks survive upgrades better now, and old format files are no longer wiped during an update. Anthropic routing now respects CLAWD CLI auth.

You've got codex encrypted reasoning, replay. So basically, stale transcripts no longer crash your sub-agents. There's also a memory search system. So this properly closes local embedding providers when searches time out, which means you don't have any old models not working that are just eating up RAM.

We have plugin hooks, which get a 30 second default timeout. So if a plugin isn't working, it's no longer going to block context compaction. There's some credential improvements. And also the gateway is now exposed in status JSON, right?

So scripts can finally read the running version programmatically. And by the way, that's just eight out of 79. Every single one came from a real failure. Someone hit in production.

Now, some people are going to say, should you back up, right? Here's what I would recommend. My advice hasn't changed. So always back up first and decide second.

This is a stable release. It went through validation. The community reaction in the first three hours are pretty positive when I've seen it on X. And the replies of people saying the Discord voice mode actually works.

The doctor is catching real secrets. Windows install actually runs. But still, if your setup works, back it up first so you can run open CLAWD backup create inside your terminal before you touch anything. Know your current version as well.

Test after updating. And if something breaks, just roll it back, right? These are the commands you can actually run with open CLAWD to create a backup, to update, to run the doctor and also the status. And in 10 minutes, you can update.

It's pretty current. Now, the thing that I would say is the people who figure out AI agents right now, whilst the tools are evolving fast, are going to be way ahead when everything settles, right? So every update you learn, every workflow you build, every problem you solve, it all improves and compounds into your skillset, right? The better you can manage these agents and the more you learn, the better you're going to be in the future too.

And if you want open CLAWD to actually save you time every day, open CLAWD is great, but open CLAWD inside the agent operating system with CLAWD and Hermes sharing memory is a different category of thing completely. So you can plug open CLAWD into the agent operating system. And if you want your open CLAWD set up to actually save you time, not just be another tool you check, grab the agent operating system inside the AI Profitable. It turns open CLAWD and Hermes into one system with a shared memory, with shared notes.

And every new open CLAWD update, like the Discord voice mode and the XAI OAuth that just shipped, they make the whole system more powerful automatically. So you don't need to rebuild, you just plug in the new update into open CLAWD. And if you're wondering what else you get inside the AI Profitable, you actually get the agent operating system zip file built in one session. So you get every prompt, every config, the Obsidian memory set up.

You get a 30 day open CLAWD roadmap walking you through open CLAWD, CLAWD and Hermes step-by-step. Coaching calls every week where we can set up with you live and you can bring your questions. Daily tutorials walking through every open CLAWD releases it ships. 3,000 members already running the stack.

So someone is always online 24 seven to help. A member map seen connect with people locally near you and a prompt library built around open CLAWD workflows, including for example, like how to connect it to Discord, how to build customer service agents, how to create content with this, how to generate leads. And as I mentioned before, run open CLAWD backup create first, decide second. I'll see you in the next one.

Cheers for watching.