26:03Brendan Jowett · Tutorial
Master ElevenLabs AI Voice Agents (Full 2026 Guide)
A 26-minute step-by-step walkthrough of the ElevenLabs agents platform: blank agent to live phone receptionist, with every major setting explained.
June 21stThe argument in one line.
Claude Code can assemble an entire voice-agent stack — phone number, speech-to-text, language model, and text-to-speech — from plain-English instructions, collapsing what used to be manual multi-vendor API integration into a single conversation.
Read if. Skip if.
READ IF YOU ARE…
- You build or resell voice agents for local businesses and want a faster way to wire up telephony, speech-to-text, and text-to-speech providers.
- You're comfortable directing Claude Code conversationally and want to see a real multi-vendor integration (Telnyx, LiveKit, Gladia, Eleven Labs) done end to end.
- You want to see how a live, API-backed function call gets added to a running voice agent without hand-writing the request logic.
SKIP IF…
- You're looking for a no-code voice-agent builder — this assumes you already have Claude Code, developer accounts, and are comfortable in a terminal.
- You want deep coverage of prompt engineering for the agent's persona — the video shows the setup conversation but not the prompt-tuning process.
The full version, fast.
A creator demonstrates building a phone-answering AI voice agent almost entirely through Claude Code conversation instead of manual API wiring. The stack: a Telnyx phone number connects over SIP to a LiveKit Cloud-hosted agent, which chains Gladia for speech-to-text, Claude Haiku 4.5 for the response logic, and Eleven Labs Flash v2.5 for text-to-speech. After a working demo call booking a bookkeeping appointment, he asks Claude Code to add a live function call that fetches the real-time gold price and has the agent read it back mid-call, without writing the request code by hand. He closes by arguing the real defensibility left for agencies isn't the build itself but negotiating access to a client's harder, undocumented internal APIs and CRMs.
Free for members
Chat with this breakdown — free.
Sign in and you get 23 free chat messages on us — ask for the hook, quote a framework, find the exact transcript moment, generate a markdown action plan. Bring your own key when you want unlimited.
Create a free account →Where the time goes.
00:00 – 00:58
01 · What I'm covering
Cold open: Claude Code is going to build and deploy a phone-answering voice agent from plain-English instructions.
00:58 – 01:33
02 · Tech stack overview
Names the four-vendor stack before the demo: Telnyx for the phone number, LiveKit for hosting, Gladia for speech-to-text, Claude for the model, Eleven Labs for voice.
01:33 – 02:27
03 · Demo call with the voice agent
Live phone demo: a caller books a bookkeeping appointment and asks for pricing; the agent quotes a $300/month starting rate and offers a free consult.
02:27 – 03:04
04 · Architecture walkthrough
Walks the Claude-drawn diagram: caller phone to Telnyx SIP trunk to LiveKit Cloud, which routes through Gladia STT, Claude Haiku, and Eleven Labs TTS.
03:04 – 03:57
05 · Gladia speech-to-text
Explains why Gladia was chosen: purpose-built for real-time phone conversations rather than general transcription, backed by an open benchmark.
03:57 – 04:17
06 · AI model selection
Claude Code picked Claude Haiku 4.5 by default; the builder notes he'd normally reach for GPT-4.1 or Gemini 2.5 Flash for voice-agent reliability.
04:17 – 04:30
07 · 11 Labs text-to-speech
Eleven Labs Flash v2.5 turns the model's text response into the human-sounding voice callers hear.
04:30 – 06:02
08 · Prompting Claude Code
Dictates the build spec out loud to Claude Code: Telnyx SIP number, LiveKit hosting, Gladia STT, Claude Haiku, Eleven Labs TTS — then lets it run.
06:02 – 07:22
09 · LiveKit Cloud setup
Shows the LiveKit Cloud dashboard and the 'receptionist' agent Claude Code created, and explains why LiveKit's API surface and self-host option won out over Retell or Eleven Labs' agent builder.
07:22 – 08:01
10 · Providing API keys
The only manual step in the whole build: copying LiveKit's project ID, project URL, SIP URL, and API key into the Claude Code conversation.
08:01 – 08:20
11 · Telnyx phone number setup
Same pattern on the telephony side — purchase and verify a number in Telnyx (1-2 day wait), then hand the API key to Claude Code.
08:20 – 08:51
12 · Agent build complete
Claude Code reports the build finished and hands back a live phone number, ready to call.
08:51 – 10:04
13 · Adding a function call
Asks Claude Code, in one sentence, to add a live gold-price lookup to the agent — it finds the API, wires the function, and updates the system prompt unassisted.
10:04 – 11:26
14 · Appointment booking workflows
Notes that deterministic booking flows would normally route through n8n or Make.com, but Claude Code could build calendar automation directly since everything else was built natively.
11:26 – 12:09
15 · Function call demo / live test
Calls the agent back and asks for gold's price; it answers with a live spot price that matches the API endpoint's actual returned value.
12:09 – 13:41
16 · Agency moat / integrations advice
Closes on where the real defensibility is left: not the agent build itself, but earning access to a client's harder, undocumented internal APIs and CRMs.
Lines worth screenshotting.
- Claude Code can wire together four separate voice-agent vendors — Telnyx, LiveKit, Gladia, and Eleven Labs — from a single spoken instruction instead of manual API integration.
- Adding a new capability to a live voice agent, like a real-time gold-price lookup, can be done by asking Claude Code to find the API and wire the function call itself.
- LiveKit was chosen over alternatives like Retell or Eleven Labs' own agent builder specifically because it exposes the most API surface for Claude Code to drive and supports self-hosting for lower latency.
- The demo agent defaulted to Claude Haiku 4.5 for the language model, even though the builder said he'd normally prefer GPT-4.1 or Gemini 2.5 Flash for voice-agent reliability.
- Gladia's speech-to-text is positioned specifically for real-time phone conversations rather than general transcription, which is why it was chosen over a general-purpose STT API.
- The entire manual portion of the build was copying three LiveKit credentials (project ID, project URL, SIP URL) and a Telnyx API key into the Claude Code conversation.
- Telnyx phone number verification can take one to two days depending on region, the one step in the pipeline Claude Code can't compress.
- The builder argues the actual competitive moat in voice-agent services isn't the build, which AI can now do in minutes, but negotiating access to a client's difficult, undocumented internal APIs and CRMs.
- A single natural-language request to Claude Code triggered API discovery, function wiring, and a system-prompt update without further input from the builder.
- The demo showed the agent quoting a live spot gold price on a phone call, verified against the API endpoint's actual returned value in the browser.
The voice-agent build is no longer the moat
STACK, THEN MOAT
Claude Code can wire a phone number, speech-to-text, model, and voice into a working agent from spoken instructions, which pushes the real competitive advantage toward earning access to a client's hardest internal APIs.
02Tech stack overview
- A production voice-agent stack has five swappable layers — telephony/SIP, agent hosting, speech-to-text, language model, and text-to-speech — and knowing the layers makes any vendor swap easier.
- A hosting platform is worth picking for its API surface, not just its dashboard, if an AI coding tool is going to be the one configuring it end to end.
03Demo call with the voice agent
- A voice agent that captures the caller's name, need, and preferred time before quoting a price handles the core job of a receptionist without extra scripting.
- Quoting a price range rather than a fixed number keeps a voice agent honest without needing per-client customization.
04Architecture walkthrough
- Mapping a voice pipeline as caller, SIP trunk, hosting, speech-to-text, model, text-to-speech, SIP reply makes it possible to debug or swap any single stage without touching the rest.
- Having an AI tool draw its own architecture diagram after a build gives a reviewable record of what was actually wired together.
05Gladia speech-to-text
- Choosing a speech-to-text vendor built specifically for live phone conversations, rather than general transcription, matters because real-time turn-taking is a harder case than a recorded file.
- Published benchmark numbers like word error rate and diarization error rate are a faster way to compare speech-to-text vendors than trusting marketing copy alone.
06AI model selection
- An AI tool's own default model choice for a task isn't automatically the best one; overriding it with a preference based on real-world reliability is still worth doing.
0711 Labs text-to-speech
- A separate text-to-speech step converting a model's text reply into audio is what makes a phone agent sound human rather than robotic — the model and the voice are two different jobs.
08Prompting Claude Code
- Specifying every vendor by name in one request up front avoids a back-and-forth clarification loop with an AI coding assistant.
- An AI coding tool can reuse already-connected accounts and skip re-asking for API keys it has seen before, speeding up repeat builds.
09LiveKit Cloud setup
- Checking a hosting dashboard for active sessions and agent status is still worth doing even when an AI tool handles the build, as a sanity check the agent is actually running.
- Self-hosting an agent layer instead of using a managed platform trades a small latency improvement for meaningfully more setup complexity, usually not worth it unless latency is the real bottleneck.
10Providing API keys
- In an AI-assisted build, the human's remaining job narrows to copying credentials from a dashboard into the chat; the integration logic itself gets delegated.
11Telnyx phone number setup
- Phone number verification can take one to two business days depending on region, a hard external dependency no amount of AI assistance shortens.
12Agent build complete
- A working end-to-end voice agent can go from a single spoken spec to a live, callable phone number in one sitting.
13Adding a function call
- A single plain-English request can trigger an AI coding tool to find the right API, wire a new function, and update the system prompt without further direction.
- Manually building a voice-agent function call, finding the endpoint, picking the HTTP method, assembling the request body, is traditionally one of the more tedious and error-prone parts of the build, which is exactly what got automated here.
14Appointment booking workflows
- Deterministic multi-step workflows like calendar booking are still often better served by dedicated automation tools than folding everything into the agent's own logic.
- Because the rest of a stack was built natively through a coding assistant, extending it to write directly to a calendar API is a natural next step rather than a separate integration project.
15Function call demo / live test
- Verifying a live agent's spoken answer against the actual API response it pulled from is a simple way to confirm a function call is really working.
16Agency moat / integrations advice
- As AI coding tools make the mechanical build of a voice agent trivial, the competitive advantage shifts to whoever can negotiate access to a client's harder, undocumented internal APIs and CRMs.
- The businesses hardest to serve with a fast AI build, those with messy or gated internal systems, are also the ones with the least competition once integration access is secured.
Terms worth knowing.
- SIP (Session Initiation Protocol)
- The telephony protocol used to hand an incoming phone call from a carrier like Telnyx into a hosted voice agent, in place of a traditional phone switch.
- LiveKit
- A real-time agent-hosting platform used here to run the phone-answering AI agent and connect it to the speech-to-text, language model, and text-to-speech providers.
- Gladia
- A speech-to-text API built specifically for real-time phone conversations, used to convert a caller's speech into text for the language model to process.
- Telnyx
- A telephony provider used to purchase and host the phone number that routes incoming calls into the voice agent via SIP.
- Function call (in an AI agent)
- A capability the language model can invoke mid-conversation to fetch outside data or take an action, such as calling an API to look up a live price.
- Turn detector
- A component that decides when a caller has finished speaking so the agent knows when to start responding, shown in the architecture diagram next to the text-to-speech step.
Things they pointed at.
06:02toolLiveKit Cloud
08:01toolTelnyx
Lines you could clip.
08:51
“Can you add a function call to our voice agent now that is gonna be able to retrieve the live price of gold on the call.”
12:44
“There definitely still is quite a big moat when it comes to building out these systems, and that does mostly come down to integrations.”
Word for word.
Read-along
Don't just watch it. Burn it in.
See every word as it's spoken — crank it to 2× and still catch all of it. The same dual-channel trick behind Amazon's Kindle + Audible.
metaphoranalogy
Today, I'm gonna go through one of the fastest ways to get a voice agent set up and deployed on a phone number. This method is gonna be using ClaudeCode to do all of these setup for us in plain English. We're gonna tell it to build out this voice agent.
We're gonna tell it to add in function calls, and I'm gonna let ClaudeCode handle essentially everything to do with the development and setup and deployment of that voice agent. So if you are currently manually setting up tool calls to book appointments for your voice agents or you're manually writing your system prompts still, then this is a development practice that I highly recommend getting upskilled on because it is just that much more efficient to get these voice agents deployed out there in the real world.
As always, it does require the human touch to come in and fine tune things and take feedback from real live calls to make these agents better. But from an initial development perspective, this method for setting up these voice agents is no doubt about it the smartest way to get up and running quickly. And because of how quick it is to set up these voice agents, it allows us to spend the effort to continue to refine that agent to get it to that production level as quickly as possible.
Diving right into it, if you have yet to watch any of my other Claude videos, I would recommend understanding a little bit about how Claude and Claude code works. Ultimately, what you're looking at right now is Visual Studio Code, this is what I use to interact with Claude code. And at the moment, I am inside my , which is essentially a folder in my computer in which has all the context about me and my business to be able to make much more informed decisions on what to build and how to build it.
Now assuming that you do have Claude code set up and running, I'm now just gonna go through a diagram that I've got Claude to create just walking through the tech stack that I'm currently using for this voice agent and that Claude built all for me. Now before I dive in, I'm gonna do a quick demo call so you can see exactly how this Claude created AI voice agent sounds.
Hello. How can I help you today? Hey.
I was looking to see if I could book an appointment if possible, uh, sometime this week.
Certainly, I can help you with that. May I please have your name and the reason for your appointment?
Uh, Yeah. I was just looking to see if I could, uh, get some bookkeeping services. Is that something that you guys do?
Yes. We do offer bookkeeping services. We provide monthly and quarterly bookkeeping to help businesses stay organized.
What day and time would be best for your appointment this week?
Do you have a rough price estimate before I book an appointment?
Yes. I can give you an illustrative estimate. Our small business bookkeeping packages typically start at $300 per month.
However, pricing can vary based on the complexity of your specific needs.
Would you like to schedule a free consultation to get a more precise quote? Okay. So there we go.
Sounds really good. We are using Eleven Labs, their voice to be able to power this. So we are still taking advantage of that real realistic sound and speed.
So taking a look at the right hand side, this is sort of the setup of the voice agent that we have at the moment. Pretty much how it works is that we call in, of course, we are hitting this phone number, and we are hosting this phone number through Telnyx. So LiveKit is what we're gonna be using to host the actual agent development part of it.
So we've got a lot of different providers in terms of our speech to text, our text to speech, 11 labs, AI models, all that stuff just gets combined through LiveKit. So we can see here it goes down to the agent receptionist, and then we've got some providers connected all in here. The first provider that we've set up is Gladia, and this is our speech to text tool.
So pretty much my speech speaking into the phone now needs to get converted into text so we can send that off to an AI model. So this right here is Gladia. You can sign up for an account, and you can then get your API key.
Provide us a Claude, and it's gonna be able to start to set up everything through their API documentation. This product is pretty much built exactly for this use case. You can have a look at their products and all of their speech to text models that they have.
Essentially, the reason that I'm using this is that this is purpose built for AI voice agents. So actual phone conversations, this is purpose built for.
And they've got pretty good APIs, I can tell Claude Code, for example, to set this up and use this alongside our LiveKit agent, and it's gonna be able to do that without me needing to really do too much. If you did wanna sign up for this and get your API key to get Claude to set this all up for you, I will have my own link to gladiator at the top of the description and pinned at the top of the comments.
I do have my own link for gladiator, so if you could use that link, that really does just help to support the channel. Next up, we've got the AI model in which it's gonna be powering the intelligence or the voice agent. At the moment, it has selected the Claude Haiku 4.5 model.
To be honest, I typically don't use a Claude model when it comes to voice agents. Claude decided it's gonna use a Claude model, of course. I would probably typically use a GPT 4.1 model or I might use a Gemini flash two and a half model.
They're a bit more reliable. That then gets sent off to the Eleven Labs flash v 2.5 model, and this is just taking that text response that the AI has generated and then turning it into that human like speech. So that really is the entire stack.
That's the whole setup that Claude is gonna be building for us. So in order to get this set up, it honestly is incredibly easy thanks to the power of Claude code, of course. All I really need to do is just tell it exactly what I'm looking at here on the right hand side of the page.
I'm gonna go ahead and just detail all of the platforms that I wanna use in order to set up this voice agent, and then honestly, I can sit back and relax and watch Claude go ahead and try to build everything that I'm looking for. You will obviously need to detail the use case of your agent as well in terms of the system prompt and what it's gonna be built for, and of course, it is gonna run you through some of the instructions in order to find API keys and give it access to everything it needs.
Hey, Claude. I need you to build a brand new voice agent for a demo that I'm doing for a YouTube video. I need you to set this agent up via a Telenix SIP, so we're gonna set up the phone number via Telenix.
You can use one if we already have one existing. Then I want the voice agent itself to be set up on LiveKit, so we're gonna have to connect Telenix to LiveKit to host that agent. And then when it comes to the actual voice agent itself being hosted on LiveKit, I wanna use the Gla I then wanna use the Claude Haiku four and a half model to power the intelligence, and then I would like to use the Eleven Labs Flash v 2.5 model in order to get L speech.
Okay. So as easy as that was to speak out to Claude, I can now send this off to Claude to get working on. Now because I have connected and set up a couple of these accounts, it is probably gonna be able to get away with not using some API keys or I don't have to provide those API keys because I have previously.
But in your case, if you are gonna be building this from scratch, of course, it will just go through some extra instructions in order to get some API keys from Gladia, for example. Now whilst our voice agent is building in the background on Claude code, so this right here is the LiveKit cloud sort of login. You will need to create an account for LiveKit and provide your agent with a couple of instructions with a couple of keys in order to better access your account and create agents in your account.
You can see at the bottom here, we can see your agents. I do have one receptionist that I created earlier, and this one was specifically all created through ClaudeCode itself. So we really don't need to be even in this interface if we don't want to be, but we can click into this agent if we do wanna view any of these statistics as to how well it's performing.
But the whole intention here is not to even need to look at this dashboard so we can really manage it all just through Claude Co. Now at the moment, I am using LiveKit. If you've watched some of my other videos around voice agents, you're probably familiar with platforms like Retul or Eleven Labs.
These platforms are also perfectly still fine to use alongside ClaudeCode. I would say that LiveKit has the most amount of API and most amount of documentation to be able to build everything as much as possible with the most amount of flexibility through something like ClaudeCode, as well as the ability to self host or locally host your voice agents is something that is sort of unique to LiveKit.
So if that is something that you would want to do, doing it through LiveKit does just give you that extra little bit of flexibility. And the reason that you might want to do that is if you wanted to host it on a more local server to help reduce the latency of the voice agent. But I do think it is quite a small improvement, I do still recommend using platforms like Retail or Level Labs and getting Claude to build in there as well.
So whilst this agent is still building, the only things that you're really going to need to be doing on your end manually is going ahead and copying all of your keys. So at the moment, I'm in the LiveKit settings. I've scrolled down a bit because I don't wanna reveal what my keys are, but essentially, you'll get a project ID, a project URL, as well as a SIP URL.
So if you just give those three things off to your ClaudeCode agent, it's likely just gonna ask you for them anyways. It will then have full access to your entire LiveKit setup. You will, of course, also need to provide it with the API key.
You can see on the left here API keys. We'll obviously need the API key to authenticate that. But once that's done, that is all the manual work you need to do, and ClaudeCode will take control from there and just set everything up.
And it's pretty much the same setup for your phone number on Telenix. You will need to purchase a phone number and verify your business if you are gonna be getting a phone number through Telenix. Depending on where you're located, it could take about one to two days to get that phone number verified, so just note that.
But once it is verified, you can then take the API keys from Telenix, hand it over to your Claude agent, and it is once again gonna be able to handle everything for you. Okay. So my Claude agent has now completed, and we can see right here it's telling me to call the number.
I told it to set it up for an accounting use case, and now it has told me that it is now set it up in the way that it's gonna be for an accounting use case. And it's gone ahead and set up everything with all our providers with Claudia, Claude, and Eleven Labs as well. And honestly, that is all I needed to do.
This is a fully working voice agent now. We are gonna be able to call this number. It's gonna be able to hit all of the endpoints that it's connected to in terms of LiveKit, and it should be able to plug into all of these different providers as well.
And we've now got a working voice agent. As for making any updates, it's as easy as just telling it to make an update. And And if we wanna see the prompt, I can ask it to provide me with the prompt to verify or look at any information in there.
I can also come directly here and tell it to set up a function call for me. Hey, Claude. Thanks for setting up the voice agent.
Can you add a function call to our voice agent now that is gonna be able to retrieve the live price of gold on the call. And so I'll send that off, and what should happen is it's gonna go through the entire process of finding the API endpoints to use to be able to find the live price of gold. It's gonna be able to connect it into the voice agent, set it up as a function call, and then go into the prompt itself to obviously make sure that that API and that function is called upon if somebody's asking for the price of gold, for example.
And if you're familiar with building voice agents in the past, this process can be a bit tedious having to set up your own API endpoints, having to figure out what post and get requests are, making sure you're using the right one, then setting up the entire JSON body to send the right request off has probably been the hardest part or the most trickiest part of getting one of these voice agents set up reliably.
And right now, we're looking at the Claude agent literally doing everything for us in terms of finding that API. We can see here it's found something. It's now literally setting up the agent with that function call, and I am doing literally nothing right now.
Obviously, finding the price of gold is probably not a very common use case for voice agents at all. Maybe it is if it's gonna be for sort of an investment thing, but booking appointments is something that we could do through this as well. Typically, for appointment booking workflows, you might use something like nnn or make.com to have a sort of deterministic flow of what's happening.
But because we are doing this all natively through ClaudeCode, I could pretty much tell it to build the entire work automation to go directly to Google Calendar and connect that to our voice agent. Now we would want to host that live somewhere so we could deploy that onto Railway, for example, which is just a cloud hosting provider, but that is just a really big benefit of building these voice agents directly through Claude code as opposed to doing it through some sort of a visual interface.
And then there's obviously all the benefits of building this inside an AI operating system like I am right now. If I was to build a voice agent for my own business, it's got all the context about everything you could possibly need. And if I was to build this out for maybe a client, if I've got client information in my folder, we could get Claude to reference that client information to be able to maybe look at a call we have with the client to understand exactly what they're looking for and build it custom to that specification.
And then maybe a client sends me an email telling me that something is broken, We can then tell Claude that we got this email. It's gonna be able to read the email, find the email to be able to read, and then make all the updates in the voice agent itself based on the email that I got. So this really is the future of voice agent development and just development in general.
So I think it is fair to say that I am a pretty big fan when it comes to building out voice agents and building out really anything through ClaudeCode just because it is just that much more productive. Okay. We should now be completed.
It's telling us that it has gone ahead and set up this function call, and it should be able to return us some info. Hey. I was looking to see if you could find the live price of gold for me if possible.
The current spot price of gold is around $4,028 per troy ounce. Okay.
So there we go. It said that it was $4,028 per troy ounce.
So And if I just quickly scroll up in our Claude build right here, we can actually have a look at the URL endpoint that it's using to find this data. And if I open this up, we can essentially have a look at the API endpoint in which is being used to retrieve this information. And we can see that the price at the moment is at 4,028, and that's exactly what we were told over the phone.
So, of course, that is exactly what it was able to call and pull back as data live on the call as I was speaking to it. So we've now got a voice agent. We've now got the ability to add functions to that voice agent as well just like this.
And so there really is nothing that we can't do right now. We can set everything up just through our Claude code conversation. And so one thing that I will say if you are an agency owner, it might seem a bit scary that you can create these voice agents so easily, but there definitely still is when it comes to building out these systems, and that does mostly come down to integrations.
So a lot of the businesses that I've spoken to that were looking to get voice agents developed, there was a very good chunk that just had very difficult APIs and CRMs and systems to integrate with. And these were all completely solvable problems, but just required a little bit extra effort in order to be able to get that set up.
And those are the types of industries and areas where there is still just a massive amount of opportunity because, obviously, this AI ClaudeCurrent is not able to develop a relationship with an API or a company for that matter. But if you are able to establish a relationship with one of these providers to give yourself access to that integration, that is obviously gonna be a pretty big moat if you can get access to those systems for that particular type of industry.
You can then give that access to your own Claude agent to then build out these agents that nobody else can because you've put in the effort to get that integration. Anyways, I won't go too much on a tangent. Hopefully, you've learned something about speeding up your voice agent development workflow.
Once again, if you are interested in getting Gladia set up for your voice agent speech to text specifically, I will have my link for Gladia at the top of the description and pinned at the top of the comments. And if you're interested in building out your own AI meeting notetaker all through Claude to help save you some money on subscription costs, I'd highly recommend checking this video out right here where I did exactly that.
The Hook
The bait, then the rug-pull.
The pitch fits in one line: hand Claude Code a list of providers and a use case, and it wires the phone number, speech-to-text, language model, and text-to-speech into a working receptionist before you finish watching the setup.
Named ideas worth stealing.
02:27list
The voice-agent stack
- Telnyx (phone number/SIP)
- LiveKit Cloud (agent hosting)
- Gladia (speech-to-text)
- Claude Haiku 4.5 (language model)
- Eleven Labs Flash v2.5 (text-to-speech)
The five-vendor chain Claude Code wires together to answer a phone call: SIP in, transcribe, think, speak, SIP out.
Steal forany Claude-Code-built phone agent that needs a documented reference stack to start from
How they asked for the click.
VERBAL ASK
13:23next-video
“if you're interested in building out your own AI meeting notetaker all through Claude to help save you some money on subscription costs, I'd highly recommend checking this video out right here”
soft CTA to another video, paired with a repeated Gladia affiliate-link mention in the description and pinned comment.
MENTIONED ON CAMERA
FROM THE DESCRIPTION
PRIMARY CTAWhere the creator wants you to go next.
OTHER LINKSAlso linked in the description.
Visual structure at a glance.
