OpenClaw 5.9: Discord Voice AI Agents Are Here

OpenClaw 5.9 just dropped in pre-release, so OpenClaw has just dropped a free update that turns your AI agent into a voice agent and assistant inside Discord. Your agent can now join a voice channel, listen to people talk, and respond out loud in real time, and that's just one part of this update. OpenClaw 5.9 also adds WeChat support, fixes iMessage so your agent catches up on messages it missed whilst you can be yourself, and removes a feature that some of you are using right now which means if you update without reading this first or watching this first your iMessage setup could break. I'm going to walk you through everything that changed, what it means for your business, and the one thing you absolutely have to do before you touch this update.

Stick with me to the end and I promise your OpenClaw setup is going to be way more powerful and way more protected. Let's get into it. So OpenClaw 5.9 just showed up on GitHub as a pre-release beta. Peter Steinberger and the team pushed this change log and it's packed.

Discord voice, iMessage catch up, WeChat, new AR models, and one breaking change that could take down your iMessage setup if you're not careful. Now because this is a pre-release beta, not a stable version, don't update your live setup to this yet. I'll show you how to test it safely at the end and how to check it out before you do anything drastic. But I wanted to cover it now because the features in here are too big to wait on.

You need to see what's coming so you can get ready. Let's start with the headline. Discord voice just got a complete overhaul. Before this update, if you wanted your agent to talk in a Discord voice channel, you had two options.

You could use speech to text to text to speech, which means someone talks, he gets converted to text, the agent reads it, thinks, writes a reply, and that reply gets converted back to speech. It works but it's slow. This is delay. It doesn't feel like a conversation.

OpenClaw 5.9 adds a new mode called Agent Proxy and this is now the default. In Agent Proxy mode, the voice channel acts like a microphone and speaker connected directly to your agent. So your agent hears everything in real time, it thinks in real time, and it talks back in real time. This is using OpenAI's real-time voice models or 11Labs as well.

You can choose between them. So instead of that awkward pause whilst text gets converted back and forth, the conversation flows better. Someone can ask a question, the agent answers, someone interrupts, the agent stops talking and listens. It feels like talking to a person on a phone call except that person has access to all your tools, all your data, and all your workflows inside OpenClaw.

One thing to be careful of here is that these new features are new features so they don't always work smoothly but I'm talking you through the idea and what works and what the idea is behind this. And there's a smart detail here too. In rooms with Echo, like someone playing audio through speakers instead of headphones, the agent used to hear its own voice bouncing back and would get confused. It would stop talking mid-sentence because it thought someone was interrupting.

OpenClaw 5.9 fixes this. It ignores new audio whilst it's speaking and you can tune how sensitive the interruption detection is. So noisy rooms, for example, work better. If you're using 11 Labs voice, replies start coming back directly into Discord playback now, right?

The agent starts talking before it's finished generating the full response. So the first words actually come out faster. For anyone running a Discord community, a customer support server, or a team workspace, this is an upgrade. Your agent can sit in the voice channel and be available for live questions, like having a team and a team member who always is on call.

Now let me talk you through iMessage because this fix solves a problem that's been annoying a lot of people for months. If your Mac goes to sleep or your computer restarts or OpenClaw crashes, your agent misses every message that comes in whilst it's offline. When it comes back, those messages are just gone. People texted you, your agent didn't see it.

And those people just never get a reply if you're not doing this manually. OpenClaw 5.9 adds catch-up mode for iMessage. So when your agent restarts, it looks at what came in whilst it was off and processes those messages like nothing happened. You can turn it on with one setting and you can control how far back it looks, how many messages it processes per restart, and how many times it retries if something fails.

For anyone using OpenClaw as an iMessage agent assistant for client communication, for example, for lead replies, for personal usage, this means your agent doesn't lose messages anymore. It picks up where it left off. But here's the part you need to pay attention to. OpenClaw 5.9 removes blue bubbles completely.

Blue bubbles was the way some people connected iMessage to OpenClaw. If your iMessage setup uses blue bubbles, this update will break it. From what I can see, your agent would stop working on iMessage until you switch to the new method, which uses a direct connection through Apple's own system on your Mac. Migration isn't hard, but you need to do it before you update, or at least know that you need to do it later.

If you just hit update without reading the changelog or understanding this, your iMessage agent would go dark. Now here's what else is new. WeChat support, OpenClaw now has a Tencent WeChat channel, which is personal WeChat, the version most people use in China. You can install it during setup or add it later if you have customers or contacts on WeChat, your agent can now reach them there.

Mistral Medium 3.5 got added to the model catalog. This is Mistral's latest reasoning model. If you're using Mistral as your AI provider, you can switch to it right away. Google Gemini 3 Pro got retired if you're using it.

OpenClaw automatically moves you to Gemini 3.1 Pro preview. You don't need to do anything, it handles the switch, but it's worth knowing in case your model name change, you know, if your models change inside OpenClaw and you wonder what happened there. Amazon Bedrock got service tier support, so if you're running models through AWS Bedrock, you can now choose between default, flex, priority, and reserved tiers. This lets you control the cost and the speed depending on your workload.

The command line got smarter as well, so when something goes wrong, your settings are broken, a channel won't connect, an add-on fails to load, for example. OpenClaw now tells you what happened and gives you the exact command to fix it. Before, you'd get a vague error message and have to figure out the repair yourself. Now it points you to the answer.

Now, if you want help setting up OpenClaw, if you want to understand how to use this for your business, how to automate tasks, you can join the AI Profit Boarding. We have some powerful roadmaps in there, we have all of our best tutorials inside there, you can connect with me personally, and we have a lot of members inside the AI Profit Boarding community using OpenClaw and AI agents like Hermes too. So, if you want to know the best ways to automate your business, save time, get more leads with OpenClaw and Hermes agents, link in the comments description, or just go to the AIProfitBoarding.com. Now, let me cover some of the most important fixes, because this beta also cleans up a lot of stuff that's been broken.

Telegram got several improvements. Long messages that were getting cut off now continue in a new message instead of just stopping. The thinking level picker now shows you what level you're currently on before you change it, and the network recovery also got smarter. If Telegram servers hiccup, your agent falls back to a backup connection, it now checks periodically.

If the main connection is back, instead of staying on the backup forever. Slack threads got fixed. If someone messages your agent in your channel and then replies in a thread, those should stay as one conversation. Before, they could split into separate sessions.

Now they stay together. Discord got a fix for group chats. Your agent now knows when a message is addressed to someone else to stay quiet instead of jumping in, and it wraps links properly so they don't create those ugly preview boxes that take over the chat. The gateway starts faster and uses less memory too, so they trimmed a bunch of work that was happening on startup.

Things that don't need to happen until they're actually needed. If your agent felt sluggish when starting up, this should help. Backups also got more reliable. The backup command was sometimes copying files that were still being written to, like active conversation logs, which could cause errors.

Now it skips those and focuses on the stuff that matters. Windows also got some love, so updates used to hang after they finished because the terminal wouldn't close properly. Fixed, and a window would flash on screen every time the gateway started, like a ghost terminal popping up for a split second. That should be gone now.

MacOS also got a bunch of fixes, so if you're on a Mac with Apple Silicon and using Homebrew to install Node, the gateway restart command sometimes couldn't find Node after a restart. That should be fixed now. The stop command used to fully disable the gateway, meaning it crashed later. It wouldn't auto-recover.

Now it just stops it and lets it recover on its own if something goes wrong, and the repair tool no longer restarts the gateway that's already running perfectly fine. Docker containers now handle long running processes properly. If background tasks spawned by your agent weren't getting cleaned up, they'd all pile up and slow things down. Now the container cleans them up automatically.

There's also a bunch of security fixes as well, so that should help a lot more. There's over a hundred more fixes covering like codecs, provider memory, dashboard, and more. This is a big, big changelog that OpenCore shipped, and one of the biggest updates that's come for them. Now the most important part of this is how to actually handle it.

Remember, this showed up on GitHub as a pre-release beta. It's not stable yet. Just be careful. If you do update or you're thinking about updating, I would wait.

And the reason that I say that is because you've got to be careful with these updates. Sometimes they can break your setup, and I would wait for the community response, which you can check on X and elsewhere, even the Discord, etc., and just check if it's actually working, if it's actually stable, because the most recent releases from OpenCore have not been that stable, and you've seen that with the trajectory of Hermes agent versus OpenCore, right? Hermes has grown, whereas, for example, OpenCore has actually reduced in demand on OpenRouter, and Hermes has actually become the most popular AI agent app in the world right now. I'd also create a backup before you update to anything, and one interesting way of setting this up is you can actually run this, for example, on a separate setup, even on like a different profile on your Mac, for example, and just test it.

See if it actually works. If it doesn't work, or it doesn't feel like the update that you want, then don't update to it on your main setup, because you don't want to break anything. I mean, if you're happy with your setup, just stick to that, right? There's plenty of people using OpenCore and happy with it, but typically they're on like the older versions, not the less stable versions that seem to be updated every day and seem to be way less stable.

Bear in mind as well, this is in pre-release, so it should be dropping fully as a stable version later and beta later, but yeah, that's all the details from the latest update. So if you want to learn how to use OpenCore properly with all of my best roadmaps on how to generate more leads, save time, get more customers with OpenCore and AI agents like Hermes too, and connect with me personally, plus get daily tutorials, new updates, all my best guides on this sort of stuff, feel free to check out the link in the comments description, or go to the aiprofitball.com. Hope to see you inside there. Cheers for watching!