GPT-Live-1 Just Dropped. Voice Agent AGI is HERE! — Transcript
Full transcript
- 0:00Hey Sophie, we are looking for a new
- 0:02accountant.
- 0:02>> Okay, I'll check what next steps might
- 0:04be helpful here. And what kind of help
- 0:06would you need from the team?
- 0:08>> Um, really everything. Obviously the
- 0:09bookkeeping for our business, any sort
- 0:11of tax advisory would be good as well.
- 0:13>> Yep, we can help with your bookkeeping
- 0:15and both business and individual tax
- 0:17returns. And an accountant can review
- 0:19your situation. Could I get your name,
- 0:20please?
- 0:21>> That's OpenAI's new GPT live one model
- 0:24released this week in the API running on
- 0:26a real phone number. If you're not in
- 0:28the voice AI space, this might not mean
- 0:30much to you. But I spent three years
- 0:32selling AI phone agents to real
- 0:34businesses with receptionists and
- 0:36customer support systems for over
- 0:38$12,000 each. And this right here is an
- 0:41AGI moment for voice AI. To compare
- 0:44this, here's the previous best voice AI
- 0:46setup that we could get. Hey Alex, we
- 0:48just recently stopped working with our
- 0:50recent accountant and we are looking to
- 0:52get a new accountant.
- 0:53>> Yeah, we definitely help out with that.
- 0:55>> Would you be able to give me a quote at
- 0:56all for bookkeeping?
- 0:57>> The main thing is that first chat is a
- 0:59free console where they look at your
- 1:01books and give you a quote based on
- 1:03that.
- 1:03>> Would we be able to book an appointment
- 1:05to learn more?
- 1:05>> Yeah, we can set that up. Can I get your
- 1:07name?
- 1:08>> So there we have it. Doesn't sound too
- 1:10bad, responds relatively quick, but it
- 1:12still has that robotic quality. And that
- 1:14agent is three separate models glued
- 1:16together. Speech to text, which turns
- 1:18your voice into words, then an AI model
- 1:20that writes the reply, then text to
- 1:22speech, which reads it out aloud. Now,
- 1:24meanwhile, the big AI labs have been
- 1:26working on something called speech to
- 1:28speech. ChatGPT voice mode, which you
- 1:30may have already used, is already speech
- 1:32to speech, which is why it sounds so
- 1:34human and so good. The model itself is
- 1:36actually listening to your voice instead
- 1:38of reading a transcript. So, for
- 1:40example, when you laugh, cough, or
- 1:42pause, it knows exactly what's happening
- 1:44and just acts and responds like a real
- 1:46person would. The problem forever though
- 1:48is that speech to speech has never
- 1:49worked properly through the API, and
- 1:51especially not on a real phone number
- 1:53like we just saw. The latency was always
- 1:55really bad, calls kept breaking, and it
- 1:58was not something that you could ever
- 1:59put in front of a real customer. But now
- 2:01that has finally changed with the GPT
- 2:03live one API. Let's take another listen
- 2:06so you can directly compare it to that
- 2:07previous call. Hey Sophie, um what we're
- 2:09just looking for a new accountant.
- 2:11>> Great, thanks Brendan. What's the best
- 2:12number for the team to reach you on?
- 2:14>> Um can I just learn more about the
- 2:15services? You guys do do bookkeeping
- 2:17just before we lock anything in?
- 2:19>> Yeah, of course. We're a Melbourne
- 2:20accounting firm and we do offer
- 2:22bookkeeping as well as tax, BAS, and
- 2:26payroll.
- 2:27What sort of help would you like to
- 2:29start with?
- 2:29>> Okay, so there we have it. Today I'm
- 2:31going to go through step-by-step how to
- 2:33set up this brand new model yourself on
- 2:35a real phone number that you or anyone
- 2:37else can call and talk to. This is an
- 2:39incredible service to sell to other
- 2:41companies. These systems can take on
- 2:43complex receptionist roles, answer calls
- 2:4524/7, never miss a single call, and
- 2:47instantly push the data from every call
- 2:49into all of the systems that a company
- 2:51already uses. And the best part is that
- 2:52I'm going to use the new GPT-6 Astro
- 2:55model to build the entire voice system
- 2:57for us. So you don't need any technical
- 2:59knowledge at all to follow along and get
- 3:01this fully built. Let's get started.
- 3:02Okay, so getting started in building out
- 3:04one of these phone agents, the first
- 3:05thing that I want you to download is
- 3:06going to be Codex. So Codex is OpenAI's
- 3:09coding agent and this is what we're
- 3:11going to be using to fully create
- 3:13everything to do with this voice agent.
- 3:14So once again, to reiterate, you don't
- 3:16have to be technical at all to follow
- 3:17along with this. We're not going to be
- 3:18doing any coding or technical setups
- 3:20ourselves. We are very simply going to
- 3:22be prompting this AI to build the entire
- 3:24system for us. It's going to be able to
- 3:26literally put the agent on the phone
- 3:27number and then all we need to do is
- 3:29call it and we'll have the entire thing
- 3:30set up for us. So just make sure that
- 3:32you've got Codex installed on your
- 3:33computer. It is a desktop application
- 3:35and as for the pricing of Codex, you can
- 3:37get started for completely free on their
- 3:39free plan, which will give you limited
- 3:41Codex access. But just be aware that you
- 3:43might hit the usage limits if you start
- 3:45talking to the agent too much and then
- 3:46you might be forced to upgrade to their
- 3:48go or plus accounts. And also keep in
- 3:50mind that these are Australian prices
- 3:52and not US dollars, so just keep that in
- 3:53mind. Next up, we now need to create an
- 3:55account with LiveKit. So, LiveKit is a
- 3:57voice agent development platform and
- 3:59it's what we're going to be using to
- 4:00connect this new OpenAI GPT Live One
- 4:03model two. And it's also going to help
- 4:04us to connect our phone number into the
- 4:06system as well. So, we are going to be
- 4:08using a separate platform for the phone
- 4:09number, but we do need a platform in the
- 4:11middle to connect our model and our
- 4:13phone number together and that's going
- 4:14to be happening on LiveKit. And this is
- 4:16also the exact same stack that OpenAI
- 4:18themselves are using. see that they are
- 4:20a customer of LiveKit and the
- 4:22speech-to-speech AI that is running on
- 4:24that ChatGPT app, which you may be
- 4:26familiar with, is a system running on
- 4:28LiveKit. So, we are getting the best
- 4:29system possible right here. And one
- 4:31really cool thing about LiveKit is that
- 4:33we can actually self-host it as well.
- 4:35So, because LiveKit is fully open
- 4:36source, we can self-host these voice
- 4:38agents and run them locally on our
- 4:40computer. But, of course, the downside
- 4:41of that is if you turn your computer
- 4:43off, then nobody's going to be able to
- 4:44call your agent. So, LiveKit does have a
- 4:46LiveKit cloud option as well. So, we can
- 4:48very easily use this and this is going
- 4:50to be my recommended way for doing this
- 4:51so that we can turn off our computers
- 4:53and these voice agents will still run
- 4:55continuously. And this is the setup that
- 4:57I would recommend if you are going to be
- 4:58building these out for other companies
- 5:00to have these systems run in more of a
- 5:01secure cloud that you can be confident
- 5:03that it's going to be running reliably
- 5:05as much as possible. Now, getting
- 5:06started with LiveKit, there are a few
- 5:08different plans. You can get started
- 5:09with the completely free plan and this
- 5:11allows you to build one agent and
- 5:13connect a phone number as well. So, you
- 5:15can get started in building out and
- 5:17following along with this entire video
- 5:18just by running on this free plan. But,
- 5:20of course, if you do scale this up and
- 5:22start to build other agents for other
- 5:23companies, but of course, they do have
- 5:25these other plans. So, if you do have
- 5:26multiple agents and multiple needs, you
- 5:28will be able to upgrade and handle all
- 5:30of that all on LiveKit's cloud. And once
- 5:32again, to be clear, these plans are for
- 5:34LiveKit cloud. Of course, if you are
- 5:36going to be self-hosting it, then you
- 5:37will not need to pay a monthly
- 5:39subscription for the cloud service
- 5:40because you're not going to be using it.
- 5:42And just to make things easier, I'll
- 5:43have CodeX and LiveKit linked it below
- 5:45in the description and pinned at the top
- 5:47of the comments down below. So, you do
- 5:48have easy access to these platforms.
- 5:50But, once you are on this page, the only
- 5:52thing that you need to do is essentially
- 5:53just create an account. So, you can
- 5:55click on this start building button, and
- 5:56then you'll just go through a typical
- 5:58sign-up process to create an account,
- 6:00and then you'll land on a dashboard like
- 6:01this. Now, don't get too overwhelmed.
- 6:03We're not actually going to be doing too
- 6:04much work in this interface itself,
- 6:06because we're going to be getting CodeX
- 6:07to essentially manage and do all of the
- 6:09work here for us. But, we'll come back
- 6:10to this in just a minute. The last thing
- 6:12that we do need is the actual phone
- 6:14number itself, and one of my preferred
- 6:16providers for this is Telnyx. They
- 6:18essentially just have a big database of
- 6:19a bunch of phone numbers in which we can
- 6:21purchase. And once we've purchased the
- 6:23phone number through Telnyx, we're going
- 6:24to be able to get CodeX, and finally, we
- 6:26should be able to call this Telnyx phone
- 6:28number, and it will connect directly to
- 6:30our LiveKit Build voice agent. I also
- 6:32just wanted to quickly add that if
- 6:33you're just looking to get a US phone
- 6:35number, you can actually purchase that
- 6:36directly through LiveKit itself. If you
- 6:38want to get any international numbers
- 6:40outside of the US, then I would
- 6:42recommend using Telnyx. In my case, I
- 6:44bought an Australian phone number, and
- 6:45so that's why I'm using Telnyx in this
- 6:47use case. But, if you're just looking to
- 6:49get a US phone number, I would say it's
- 6:51probably going to be the easiest just to
- 6:52purchase it directly through LiveKit,
- 6:54and you can skip the entire Telnyx
- 6:56setup. So, to get started on Telnyx, you
- 6:58will just need to hit the login button,
- 6:59sign up for an account, and we'll come
- 7:01back in just a minute on actually
- 7:02getting this all connected to CodeX so
- 7:04it does everything for us. Okay, so we
- 7:05are now in the CodeX desktop app. Make
- 7:07sure that you have signed in to your
- 7:09account, and make sure you are in the
- 7:11CodeX option. If you look at the top
- 7:12right, make sure you're in CodeX and not
- 7:14the chat GPT
- 7:15mode. Once we are in this CodeX mode, we
- 7:17are now going to create a folder in
- 7:19which we're going to be doing all of
- 7:20this work in. Now, it should prompt you
- 7:21on this main part of the screen to
- 7:23create a folder, and if it doesn't, you
- 7:25should be able to click into this text
- 7:26and select a new folder or create a new
- 7:28project, in which will automatically
- 7:29create a new folder. In my case, I've
- 7:31named the folder LiveKit Demo. And now,
- 7:33as simple as it sounds, I am literally
- 7:36just going to talk to CodeX with my
- 7:37voice and tell it exactly what I'm
- 7:39looking for. I'm going to tell it that I
- 7:40want a voice agent. I'm going to give it
- 7:42the use case for the voice agent to be a
- 7:43receptionist, maybe for an accounting
- 7:45company or something else. And I'm going
- 7:46to tell it that I want the stack to be
- 7:48an agent that's built on LiveKit. I want
- 7:50that agent to be using the GPT Live 1
- 7:52API, and I want to get a phone number
- 7:54from Telnyx plugged into this as well.
- 7:57And literally by me speaking to the
- 7:58computer and telling it to build that,
- 7:59it's going to figure out how to build
- 8:00everything for us, and I can sit back,
- 8:02relax, and just watch it work. Hey
- 8:03Codex, I would like you to build a voice
- 8:05agent. I would like to build a
- 8:07receptionist system for a plumbing
- 8:09company. If you could build this system
- 8:11out using LiveKit, using the new GPT
- 8:14Live 1 API to power this, that would be
- 8:16great. I would also like you to connect
- 8:18into Telnyx. I've got a phone number on
- 8:19Telnyx. Uh if you can connect into that
- 8:21account as well, uh to build the entire
- 8:23setup. And so, literally from that
- 8:25incredibly basic prompt, laying out step
- 8:27by step what I want it to build, Codex
- 8:29is going to be able to go on the
- 8:30internet, find LiveKit, it's going to be
- 8:32able to look through that documentation
- 8:33to look at all of the LiveKit APIs, and
- 8:35it's going to be able to look at that
- 8:36code from LiveKit and understand exactly
- 8:38what it needs to do to actually build
- 8:40this voice agent on my LiveKit account.
- 8:42One thing that I will quickly add to my
- 8:43prompt is that I want it to be using my
- 8:45LiveKit cloud account. We don't want it
- 8:46to accidentally build it and run it
- 8:48locally on my computer. So, I will just
- 8:50tell it that I want it to build the
- 8:51agent into my LiveKit cloud account. And
- 8:53just to make sure, can you put this
- 8:55agent in my LiveKit cloud account? I
- 8:57don't want this to be running locally. I
- 8:59want you to connect into my LiveKit
- 9:01cloud account and build the agent there.
- 9:02>> Just another quick note, in your prompt,
- 9:04I'd actually recommend specifically
- 9:06typing LiveKit CLI, just telling it in
- 9:09the prompt to install the LiveKit CLI
- 9:11specifically. In my prompt, I just said
- 9:13LiveKit, but if you want it to run as
- 9:15smoothly as possible, just type LiveKit
- 9:17CLI. That will then pick up on exactly
- 9:20what you're looking to install. In my
- 9:22case, mine was already sort of
- 9:23configured, um so it was able to figure
- 9:25that out. But to ensure that it runs as
- 9:27smoothly as possible, just type LiveKit
- 9:29CLI. It's going to be able to install
- 9:32and understand exactly what we need to
- 9:33do just by doing that.
- 9:34>> Okay, I have now sent that message off
- 9:36to Codex. We're going to about to sit
- 9:37back, relax, watch it work, and it is
- 9:40just going to tell us step-by-step
- 9:41everything that we need to do. And if we
- 9:43ever run into any issues, if you are
- 9:45following along with this, you can just
- 9:46tell CodeX what the issue that you're
- 9:48having is. You can just ask it how to
- 9:50find something that it told you to find.
- 9:52We can see now that CodeX is asking us
- 9:53about the plumbing company that we're
- 9:55building the voice agent for. You can
- 9:56just make something up. This is just for
- 9:58a demo. Okay, now just quickly, we can
- 9:59see that the agent actually found a
- 10:01LiveKit project from one of my other
- 10:03folders that it was able to connect
- 10:04into. So, in your case, obviously, if
- 10:06you're setting up brand new, this is not
- 10:07going to happen for you. But in your
- 10:08case, what should happen is that it
- 10:10should tell you that it requires your
- 10:12account to be authenticated. And it is
- 10:14more than likely going to ask you to
- 10:16provide some kind of an API key from
- 10:18LiveKit. And we can find these keys by
- 10:20going into the LiveKit web dashboard,
- 10:22heading into the settings section, which
- 10:24is just if we click out of here at the
- 10:26bottom in settings, and then we click on
- 10:28API keys. This is where we can create a
- 10:30key in the top right. And we can create
- 10:31this key in the top right with the
- 10:33create key button. We can name this
- 10:35whatever and once it's created, I can
- 10:37click into this one that I just created.
- 10:39We will then get back quite a few
- 10:40details, and it's more than likely that
- 10:42CodeX will ask us for some of these
- 10:44details. And you just need to follow
- 10:45along, take the API keys and the secret
- 10:48and all the variables that it might be
- 10:49asking for, and provide it directly to
- 10:51the agent. Now, I don't recommend
- 10:53directly copying and pasting these API
- 10:55keys into the chat itself, and CodeX
- 10:58will more than likely not recommend you
- 10:59to do that either. But it will give you
- 11:01a step-by-step on how to do it properly.
- 11:03Essentially, we just have to paste these
- 11:05keys into a secrets file that CodeX has
- 11:07created for us. This way that the keys
- 11:09don't actually go through any AI model,
- 11:11and we're just putting the keys in where
- 11:12they should go. And once we've done
- 11:13this, you should get that authentication
- 11:15into your account, and you should see it
- 11:17come up with your account name as well
- 11:19just to confirm that it is properly
- 11:21connected. Now, another thing to note,
- 11:22whilst this is building, you can see
- 11:24here it says your OpenAI key has access
- 11:26to GPT live one. Now, CodeX might
- 11:29already have access to your API key if
- 11:31you've used it in any previous projects.
- 11:33But, if it doesn't have access to that
- 11:35API key yet, or it just seems like it
- 11:37hasn't found that API key and is at any
- 11:38point asking you for that OpenAI API
- 11:41key, you will, of course, just need to
- 11:43head to OpenAI's API console. So, you
- 11:45will, of course, need an account with
- 11:46OpenAI, the same one that you're using
- 11:48for Codex is fine. So, on the openai.com
- 11:51page, you can just go to login and then
- 11:52click into the API platform. And then,
- 11:54you should see at the bottom left of
- 11:55this dashboard, there's an API keys
- 11:57section. Inside of here, you can then
- 11:59very easily create the API key, copy it,
- 12:01and then provide it to the project where
- 12:03needed. Okay, so our Codex agent is
- 12:05almost complete at building out our
- 12:07entire voice agent. And once it's done,
- 12:09I'm going to give it a live call, so we
- 12:10can listen to exactly how good it
- 12:11sounds. But, before we do that, if you
- 12:13could just quickly check that you are
- 12:14subscribed by scrolling down below and
- 12:16clicking on the subscribe button. I
- 12:18recently discovered that 90% of you
- 12:19watching right now are not subscribed to
- 12:21the channel at all, which is pretty
- 12:23insane, and I would really appreciate it
- 12:25if you were subscribed. And make sure to
- 12:26keep watching as we're now going to
- 12:27finish up building this voice agent.
- 12:29Then, I'm going to give it the live call
- 12:30and really stress test it, so we can see
- 12:32exactly how well it sounds from the very
- 12:34little work that we've put into actually
- 12:36building it. And then, after this, I'm
- 12:37also going to get Codex to build an
- 12:38actual tool call into our voice agent,
- 12:41so it isn't just a conversation back and
- 12:43forth, but we can get our voice agent to
- 12:44actually do something for us. So, that
- 12:47might be in the form of booking
- 12:48appointments or sending information to
- 12:50some kind of a CRM system. Okay, so
- 12:52Codex is almost complete at building out
- 12:54the voice agent. The one The last thing
- 12:55that we do need to do is give Codex
- 12:57access to our Telnyx account. So, once
- 12:59again, Telnyx is just used for our phone
- 13:01numbers. You will just need to click the
- 13:02login button at the top right and create
- 13:04an account if you don't have one
- 13:05already. Once you've created that
- 13:07account, you should land on a dashboard
- 13:08that looks similar to this. All that we
- 13:10need to do is go to left-hand side and
- 13:12click on the voice suite. Inside of this
- 13:14section, underneath the numbers tab on
- 13:15the left-hand side, we can then click
- 13:17into phone numbers. Now, this is where
- 13:19all of our existing phone numbers are
- 13:21going to be stored. You can see one of
- 13:22my Australian phone numbers, in which is
- 13:25on my dashboard. And so, all you need to
- 13:27do is to click on buy number in the top
- 13:29right corner. And then, you're going to
- 13:30be able to buy any phone number that you
- 13:32want through Telnyx, and it will just
- 13:34store itself on our account. And then,
- 13:36we're just going to connect Code X to
- 13:37our Telnyx account, and it's going to
- 13:38set up everything for us. So, if you are
- 13:40looking for a US phone number, you can
- 13:42just click on search numbers, and just
- 13:44make sure that the US is selected. And
- 13:46you'll see that there are hundreds of
- 13:48phone numbers that we can get,
- 13:49especially the US ones. There is a lot
- 13:51of them, and they're about a dollar a
- 13:52month to use. Once you do have your
- 13:54phone number, there isn't anything else
- 13:56to do. Just make sure to follow through
- 13:57with any instructions that it gives you
- 13:59to set up that number. But otherwise,
- 14:00just go to the top right corner in this
- 14:02sort of profile icon. You will, of
- 14:04course, need to add some balance into
- 14:06your account. It does cost money to make
- 14:08these calls, of course. And so, you can
- 14:10add in 5, 10, 15 dollars just to make
- 14:12sure that you can do some demo calls.
- 14:14But for reference, it will cost a couple
- 14:16of cents for every single call that you
- 14:18make. So, this will last quite a while.
- 14:20And also, in this section, just make
- 14:21sure to click into API keys. And then,
- 14:24we just need to create a brand new API
- 14:25key. And once again, Code X should be
- 14:27giving us the step-by-step instructions
- 14:29on where to place this API key in the
- 14:31files it's given us. And once you've
- 14:33done that, that is literally all that we
- 14:34need to do. Code X will now have full
- 14:36access to our Telnyx account, and it's
- 14:38going to be able to get that phone
- 14:40number and fully connect it to Live Kit
- 14:42without us having to do any of the
- 14:43technical work ourselves. Okay, so this
- 14:45is now done. We can see built and
- 14:48deployed Harbor Plumbing's AI
- 14:49receptionist to your Live Kit Cloud
- 14:51account powered by GPT Live 1. So, this
- 14:54is all ready to go, and it's even
- 14:56telling us that this is the phone
- 14:57number, in which is the one that we just
- 14:59saw on my Telnyx account. And it is
- 15:01supposedly all set up and ready to go.
- 15:03>> Hi there. This is Mia, Harbor Plumbing's
- 15:05AI demo receptionist. This call may be
- 15:07recorded. How can I help with your
- 15:09plumbing issue today?
- 15:11>> Hey Mia. Um, we've got actually a
- 15:13leaking shower tap, uh, leaking shower
- 15:15head. I was wondering if that's
- 15:17something that you guys, uh, do any
- 15:19services for, and if you could provide
- 15:20us with a rough price estimate for
- 15:22something like that, that would be good
- 15:24as well.
- 15:27>> Yep, a leaking shower head is something
- 15:29we help with. We cover leaking taps,
- 15:32shower heads, and pipe leaks. Pricing
- 15:34can vary depending on the job and what's
- 15:36found on site though. So, I can't
- 15:38promise a quote. What's happening with
- 15:40the leak? Like is it steady dripping or
- 15:42more of a constant flow?
- 15:44>> Yeah, [snorts] it's more of a constant
- 15:45flow. There is sort of a flood that's
- 15:47building up at the bottom, so it is a
- 15:48bit concerning.
- 15:52>> Mhm, I see. Thanks for telling me.
- 15:55Is there any active flooding spreading
- 15:56beyond the shower area or any sewage
- 15:58coming up?
- 16:00>> Uh yeah, it's sort of flowing out of the
- 16:02bathroom now and almost down the stairs.
- 16:04So, we do need to get this checked on
- 16:05pretty quickly.
- 16:10>> Yeah, that's understandable. Thanks for
- 16:12letting me know. I'm just checking on
- 16:14the next step.
- 16:15Please move everyone away from the water
- 16:17and stairs and contact local emergency
- 16:19services on triple zero if there's any
- 16:21immediate danger. Don't enter unsafe
- 16:23areas. Once you're safe, I can take down
- 16:26the details for this demo inquiry. Is
- 16:28anyone in immediate danger right now?
- 16:31>> [snorts]
- 16:31>> No, everyone's safe. Uh we just wanted
- 16:32to get somebody to uh come out and help
- 16:34us.
- 16:36Okay, so there we have it. Obviously, it
- 16:37now works and it's swapped over to that
- 16:39plumbing type receptionist system. Of
- 16:40course, there will always be refinements
- 16:42when it comes to the actual
- 16:43conversational aspect in terms of in
- 16:45terms of what the agent is or isn't
- 16:47capable of doing. So, you may have heard
- 16:49that it was telling me that it was a
- 16:50demo agent and it wasn't able to do
- 16:51certain things. So, Kore AI has
- 16:53obviously prompted it to be a demo
- 16:54agent. What I could do is just tell it
- 16:56to remove the demo agent part of it,
- 16:57obviously. But, hopefully, you get the
- 16:59idea. In just very little work, we've
- 17:00been able to build that full-on voice
- 17:02system. And if we spend some actual time
- 17:04into this, giving it some proper
- 17:06requirements for a legitimate business's
- 17:08series of call reasons, we're going to
- 17:09be able to build the system pretty good
- 17:11to be able to handle a wide range of
- 17:13calls. But, of course, these agents are
- 17:14useless if we are just talking back and
- 17:16forth with them and we're not doing any
- 17:18actual work. And so this is where tool
- 17:20calls come into place. So a tool call
- 17:22might be booking an appointment or
- 17:23sending some information into a CRM
- 17:25system. This could also be pulling
- 17:27information out of a CRM system into the
- 17:29voice agent as well. And so I'll quickly
- 17:31showcase exactly how we can set up tool
- 17:33calls very, very easily through this as
- 17:35well. It is as simple as me telling
- 17:37Codex to set up a function call within
- 17:39this voice agent. We don't have to use
- 17:41any secondary or third-party platforms
- 17:43to build any automation workflows
- 17:45anymore. We just need to tell Codex to
- 17:46literally build out the entire workflow.
- 17:48It's going to be able to connect into
- 17:50any external apps that have APIs that
- 17:52are readily available and build
- 17:53everything out for us. Hey Codex, I
- 17:55would now like you to build a function
- 17:57call into this voice agent in which is
- 17:59able to look up the live price of gold.
- 18:02So if you could if somebody is asking
- 18:03for the live price of gold, this should
- 18:05trigger the function call to hit a
- 18:07public API endpoint for the live price
- 18:09of gold, pull that price back into the
- 18:12conversation and read it out to the user
- 18:14so that they are getting the real
- 18:15up-to-date price and it's not some
- 18:17outdated price. And now I'm going to
- 18:18send that off and that's all I need to
- 18:20do. Okay, Codex is now telling us that
- 18:22this is now complete. The tool call has
- 18:23been added to our voice agent. I'm going
- 18:25to give it a call and see if this works.
- 18:27>> Hi there. You're through to Mia, Harbor
- 18:29Plumbing's AI demo receptionist. This
- 18:31call may be recorded. How can I help
- 18:33today?
- 18:34>> Hey Mia, I was just wondering if you'd
- 18:36be able to give me the live price of
- 18:38gold at the moment.
- 18:42>> Sure thing. I'll check the latest gold
- 18:44price for you.
- 18:48Okay, the latest spot quote from Gold
- 18:50API is 4,314
- 18:53US dollars and 50 cents per troy ounce,
- 18:56updated about 30 seconds ago.
- 18:58Is there anything else I can help with
- 18:59today?
- 19:00>> Perfect, that's it. Okay, and there we
- 19:02go. You can see that it was able to
- 19:03obviously do that function call to get
- 19:05the live price of gold. It actually said
- 19:07the name of the API on there. Obviously
- 19:08you'd fine tune that. It's not really a
- 19:10use case for a plumbing business. But
- 19:11just to prove the point, we can do tool
- 19:13calls on this speech-to-speech model, as
- 19:15well. I hope you've enjoyed the video.
- 19:16Let me know down below in the comments
- 19:17what you think of this new voice model.
- 19:20And, if you want to push GPT-6 Astra
- 19:22even further, I did a full-on video
- 19:24building out websites with this new
- 19:26model. And, honestly, the results have
- 19:27blown my mind as to how good it is at
- 19:29design now. So, I'd highly recommend
- 19:31checking this video out.
About this transcript
This page contains the full transcript of GPT-Live-1 Just Dropped. Voice Agent AGI is HERE! by Brendan Jowett, generated from the public captions YouTube serves with the video. The transcript has 4,715 words across 653 segments, with the original timestamps preserved so you can click any line to jump to that moment in the embedded player.
What you can do with it
Use the transcript to take notes, quote the speaker, build a study guide, generate a summary with ChatGPT or Claude via the YouTube Summary tool, or export it as a timed subtitle file with YouTube to SRT. You can also re-open it in the transcriber to translate the transcript into 100+ languages.
Free YouTube transcript tool
YouTube2Text is a free YouTube transcript generator — no signup, no daily limit. Paste any YouTube link and get the full transcript instantly, with timestamps, click-to-jump, translation to 100+ languages, AI prompts for ChatGPT, Claude, and Gemini, and exports to TXT, SRT, VTT, or Markdown.