Hermes Voice Agent + GPT-6 Astra Is Next Level

Julian Goldie (0:00)

Hey, Greg, all right, great question. So Greg is asking about the badge feature with Hermes. This is a new feature that dropped in version 0.2 of Hermes updates. So it's the one before last 0.2. And with Hermes desktop, you can basically use Hermes desktop, it gets it to basically reply to you, and then you can orchestrate it directly, right? And the way that you can do this is, well, I'll come on to how to set up in a second, but if you want to see an example of this in action, so we can make sure we start a voice conversation here, inside a new conversation inside Hermes, and then it can talk back to us. So it's basically a voice agent built into Hermes desktop that can reply directly to us. By the way, if you're watching this, you may not hear the sound when it replies, but it does speak back to you, which is fantastic. But just bear in mind, like sometimes on the replays of these, it doesn't sound that good, so that's probably why you can't hear about. So let's test this out now. Once you're going to stop, you can see the Astra has configured this for me, so I'm going to ask it to stop right there. So we'll stop that right there. And then if we go back to Hermes here, we can use GPT-6 Astra. So that's what I'm using to reply. And if we go inside the stark voice conversation here, hey, how's it going? So you can see now that it's replying to me. And also if I speak up at the top of it, it's going to interrupt it. So it's going to use a barge feature to reply to us. Are you ready? So it can speak back to us. We say, are you ready? It will say, yep, ready. What's up? Can you just write a quick headline for an article about AISO? So this is the voice agent. We can end that conversation right here. It reads it back to us. So you might not have heard it on the recording, but yeah, it replies back to us with the voice. So it can reply in real time. And we can also use a barge feature, which means we can interrupt it halfway through. So for example, if it's thinking and it takes a few seconds to think and reply, if it's thinking, then we can interrupt it halfway through using Hermes barge feature. And that just means we can basically like, you know, interrupt it halfway through like a live voice agent. So that's how you can use this, Greg. Great question.

So just to recap it.

And actually, let me show you exactly how I configured it. So you can see here, I actually tried to use it first of all, and then it said it failed. So you can see here it says, the voice has failed. And I just took a screenshot. So I took a screenshot of the desktop app. And then in 45 seconds, Astra configured it to reply back to me.

And it used computer use to do this. So here's the way that I fixed it. If it's not working for you currently, or if Claud is having a hard time, and I found like Claud has had a hard time with everything recently, like I've literally switched everything over to GPT now. Basically, the steps that I followed were, you take a screenshot of Hermes desktop, you plug that into Astra 6 inside Codex, and then you ask Codex to fix it. It took about 45 seconds for Astra 6 to configure it with computers inside Hermes desktop, so the voice transcription actually works. And I'll show you a screenshot example of this. Also, one thing to note is that you can use the barge feature just to interrupt it at any time. So this basically means you can interrupt it when it's speaking out. I also think that if you just wanted, for example, oh, actually, I noticed that you say here, Greg, that it types out larger answers instead of speaking. For me personally, I want to have a look at this. Let's have a look. Yeah, I think you could go inside Chat Cheapity and ask Chat Cheapity to make sure Hermes types out the long answers as well. So if you find, for example, any problems in the way it's configured, I would just use Astra 6 to configure it for you. I definitely wouldn't use Claude at this point. I find Claude is like just not direct at fixing things, and often it struggles to give you a straight answer. So if you're struggling with Claude, use GPT-6 instead. And they're basically the steps. So just to recap, if anyone's watching this and they want to configure the voice agent inside Hermes, get Codex and Astra 6 to do it for you. You might also wonder, okay, which API or which model should you use as a brain? I'm using Astra 6 for this. Honestly, that's probably going to use a lot of tokens if you're using it constantly. So you could switch to any of these, but I think personally, I would go with GPT, because I think it tends to be quick. And you could use, for example, like GPT-6, sorry, GPT-56 or 5.5, you don't want to use too many tokens. And that way, you already have the subscription plugged in, and you don't need to pay for an API to use that as well. And then just to recap, you take a screenshot of Hermes' desktop, you plug that into Astra 6 inside Codex, you ask Codex to fix it, and then boomshack lacquer in 45 seconds, you're good to go. And it works pretty nicely.