How I Build & Sell Voice AI Agents (For Passive Income)
A live 18-minute walkthrough of a seven-step system for building white-labeled AI phone receptionists using Trillet AI and Claude Code.
Posted
3 months ago
Duration
Format
Tutorial
educational
Views
257
9 likes
57 · 43
Big Idea
The argument in one line.
A white-labeled AI phone receptionist business becomes repeatable when Claude Code handles the configuration work, so the operator only needs a website URL to demo and an onboarding form to trigger the full build.
Who This Is For
Read if. Skip if.
READ IF YOU ARE…
You want to sell AI services to local businesses but find the per-client build time too long to be profitable at scale.
You are comfortable with Claude Code or AI coding tools and want a productized B2B workflow.
You are building an AI automation agency and need a recurring-revenue service with low ongoing maintenance.
You want to offer a service under your own brand without exposing the underlying platform to clients.
SKIP IF…
You need real-time calendar scheduling built in -- that feature is on higher tiers not shown here.
You are not comfortable working with APIs, MCP integrations, or phone number provisioning.
You are looking for a consumer product rather than a B2B service model.
TL;DR
The full version, fast.
Local businesses miss 20-60% of their incoming calls, and 85% of those callers never call back. The system deploys white-labeled AI phone receptionists under the agency brand, billed as a recurring monthly service. Trillet AI hosts and white-labels the voice agent with connected Stripe payments; Claude Code builds and configures the entire agent using only the client onboarding form responses and workspace ID. The full seven-step process runs from a five-minute demo to a live agent in under 30 minutes of human effort, targeting local service businesses at $95-$159/month.
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.
Opens with the $159/mo auto shop story; establishes that local businesses miss 20-60% of calls and 85% of those callers go to a competitor
02:01 – 02:55
02 · The Two-Tool Stack
Trillet AI (white-label hosting, Stripe) and Claude Code (automated build engine)
02:55 – 04:35
03 · Building the Demo in 5 Minutes
Live demo: paste business website URL into Trillet, generate agent, arrive at sales call with a working personalized prototype
04:35 – 06:08
04 · Onboarding, Sub-Accounts & Invoicing
Post-close steps: Trillet sub-account creation, client portal invite (view-only), onboarding form sent immediately, invoice setup
06:08 – 11:26
05 · Claude Code Agent Build
Paste onboarding responses plus workspace ID into Claude Code; Claude loads onboarding skill, smoke-tests Trillet API, scaffolds GitHub repo, writes system prompt, pauses for KB documents
11:26 – 16:05
06 · Reviewing What Claude Built
Walks through the completed agent: full system prompt, knowledge base, LLM/voice/transcription configuration, call limits, post-call email summaries, SMS booking link
16:05 – 17:15
07 · The ROI Sales Math
Missed calls times average customer value equals monthly revenue leaking; positions the monthly fee as trivial against that number
17:15 – 18:28
08 · The 7-Step System Summary
Slideshow recap of the full repeatable process from demo to go-live
Atomic Insights
Lines worth screenshotting.
Local businesses miss 20-60% of incoming calls, and 85% of those callers go to a competitor without calling back.
A working, personalized voice AI demo can be generated in under five minutes from just the business website URL.
White-labeling means clients only see the agency brand and recurring payments route directly to the seller Stripe account.
Claude Code connects to the Trillet API via MCP and handles the entire agent build: system prompt, knowledge base, tool calls, and GitHub repo scaffolding.
The onboarding form is sent the moment the deal closes -- the client responses are the only input Claude Code needs to build the agent.
Setting a 10-minute max call duration protects against runaway credit consumption from unusually long calls.
Post-call email summaries delivered to up to three recipients include the transcript, extracted variables, and a playback link.
The ROI math closes B2B deals: missed calls per month times average customer value reveals revenue leaking, making the monthly fee appear trivial.
Per-client GitHub repos keep each agent build isolated and version-controlled for future changes.
The starter tier at $95/month handles 24/7 call answering, call transfers, escalations, and SMS booking links.
Claude Code can resume a build from a new session by ingesting the previous session handoff note.
Letting the business owner dial the demo number and talk to their own AI receptionist on the first sales call is the single highest-leverage closing move.
Takeaway
Seven steps turn one five-minute demo into monthly income.
HOW THE SYSTEM WORKS
The leverage is that Claude Code handles the configuration work that used to take hours, so the operator only needs to show up with a demo and ask good onboarding questions.
Trillet AI's white-label mode means clients only see the agency brand and recurring payments go directly to the seller's Stripe.
The demo requires only the client's website URL and takes under five minutes, giving the seller a working personalized agent for the first sales call.
The onboarding form is sent the moment the deal closes so the client responses can trigger the full Claude Code build without delay.
Claude Code connects directly to the Trillet API and handles system prompt writing, knowledge base creation, tool call configuration, and GitHub repo scaffolding automatically.
Setting a 10-minute max call duration (600 seconds) protects against runaway credit consumption from unusually long calls.
Post-call email summaries sent to up to three recipients include the transcript, extracted variables, and a playback link, removing manual follow-up overhead.
The ROI math closes B2B deals: missed calls per month multiplied by average customer value shows revenue leaking, making the monthly service fee appear trivial.
Glossary
Terms worth knowing.
Trillet AI
A voice AI hosting platform that lets agencies white-label AI phone agents under their own brand, connect their own Stripe for payments, and create isolated client sub-accounts.
White-label
A product sold under the reseller brand rather than the original platform name, so the end client never sees who built the underlying technology.
Onboarding skill
A Claude Code project containing documentation, templates, and API wrappers that Claude loads at the start of a new client build to scaffold the entire agent configuration.
Knowledge base
A structured document inside a Trillet agent containing business-specific facts -- hours, services, FAQs, intake questions -- that the AI draws on during live phone calls.
Post-call analysis variables
Structured data fields extracted from each call (caller name, issue type, etc.) that Trillet surfaces in the post-call email summary.
Sub-account
An isolated workspace inside Trillet created for each client, preventing any overlap between different clients agents, knowledge bases, and call logs.
“This is the part that used to take me hours. Now it's just a matter of minutes, and it's the same every single time. That's what makes this repeatable.”
Tight payoff line after long build demo→ TikTok hook↗ Tweet quote
18:13
“One demo, recurring revenue.”
Six-word encapsulation of the entire model→ IG reel cold open↗ Tweet quote
02:37
“Claude Code is the build engine. Instead of handwriting prompts and clicking through settings for an hour, two hours, three hours, Claude Code builds and configures the entire agent for you.”
Single-sentence core value prop→ newsletter pull-quote↗ Tweet quote
The Script
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.
17px
metaphoranalogy
This is the exact system that I use to build and sell voice AI agents for monthly passive income. I built a custom AI receptionist demo for an auto shop recently. It took me about five minutes.
They saw it. They signed up. And two days later, they were paying me $159 every month.
That's almost $2 a year. And from start to finish, I spent less than thirty minutes of my own time on the entire build. But most people trying to sell voice AI are using the wrong tools.
You can't put your name on it, you can't build a demo fast, and you keep losing clients, you should be winning. So in this video, I'm going to show you exactly how I do it using Trilit AI and Claude Code to build, demo, and sell voice AI agents that pay you every single month. Let's get into it.
Here's the problem that almost every local business has. They're missing a huge chunk of their phone calls. Look at this.
So most businesses miss anywhere from two to six out of 10 calls. That's 20 to 60%. And 85% of those people, they don't even call back.
They just call your competitor down the street. It varies quite a bit here, two to six, because it really just depends on the industry. Think about who's calling.
Nobody's just calling up a plumber to browse. They're calling because water is currently on the floor. They have a burst pipe.
Why do these businesses miss calls? I don't think it's out of laziness. The dentist is with a patient.
The plumber is under the sink. The shop owner is helping another walk in. The phone rings and no one's there to answer it.
What has the fix been until now is you hire a receptionist, but a receptionist is expensive. They call in six sometimes. They clock out at 05:00.
They don't work on weekends or holidays. Even a great receptionist can only take one call at a time. An AI receptionist, on the other hand, they answer every single call.
They work around the clock. They never sleep. They never take a sick day.
They can handle multiple calls at the same time in different languages too, all for a tiny fraction of what a human costs. The entire system runs on two tools with a third one being optional.
That's it. Trilit.ai and claudecode.
Let me show you what each one does because the combination is what makes this fast and actually profitable. Trilit is where the agent lives and runs. It's the platform that hosts the voice agent and handles the actual phone calls.
The part that matters if you want this to be a business, Trilit lets you white label the whole thing. So your client never sees Trilit. They see your brand, your name, your logo, and you connect your own Stripe.
When a client pays their monthly fee, that money goes straight to you. Claude Code is the build engine. So instead of handwriting prompts and clicking through settings and configuring your agent for an hour, two hours, three hours, Claude Code builds and configures the entire agent for you.
It writes the system prompt. It sets up the tools and integrations. It creates the whole structure.
But enough talking about it. Let's actually build one right now so you can see how fast this really is. Remember what I said about that recent client that I just landed, the auto shop?
That lead came to me late one morning. I grabbed their website URL. That's the only thing I need, and I paste it in right here.
In my demo today, we're going to use this lucky business right here, TruePros Heating and Air. I go to their website and I copy their URL. Inside of Trilit, I go up here to the top right where it says new call flow and then create from business.
I just paste in that website URL right there and say look up business. Once it finds the business, I say continue to voice selection. I always default to Astra, and then I say generate flow.
This takes anywhere from three to four minutes. It's going to their website. It's scraping all the information.
It's building out a knowledge base, and then it's going to build out and configure the AI agent for me. This is the demo.
This is what I'm going to bring with me onto the sales call, onto that first consultation call. It's the number one thing that you can bring with you to the first call. It just finished.
This is the agent right here. A whole system prompt is built out for me. And if I go over to knowledge base, I can see that I have a knowledge base built out for me.
I can test it out really quick, see if there's any modifications that I need to make. Nine times out of 10, it's great. And I just show up to my sales call with the demo.
This is something that the business owner can get their hands on. Well, kinda. They can pick up their phone, dial the number that you have connected to this, and they can talk to it live with you on the call.
They can hear their own business name. They can ask questions about their business. It's hyper personalized to them.
They can start to see the vision. Now the first thing that I do once a client says yes is I set up their sub account inside of Trilut. Pay attention really quick.
It's super easy. You go over here to the top left hand corner, and then you select add workspace, and you give it a name.
That's it. I'm gonna name this TruePros Heating and Air, and then say add workspace. And it will take me to their workspace.
This is why I love TrueLit so much. It's the best option if you're actually wanting to run an agency. Trust me.
I've tried them all. I've tried 11 Labs, Retail, Vappy. Trilit is where you wanna be.
We invite our client to the portal. So just throwing in a demo email right here. We set the permissions to view only, set it to simplified, and we send the invitation.
We go back to our own workspace, the master workspace, and we go to invoices, and this is where we create that initial invoice. One thing that I totally forgot to mention is before I even create the subaccount and the invoice to send it over, the first thing that I do, because this is what takes the longest for my client to get back to me with, is the onboarding form.
So I send that over immediately. This takes about ten minutes to fill out. Now we can do this.
We can send it through email to my client, and then they can take their time and get back to me, fill it out. Or if they prefer, we can just hop on a quick call, and I can ask the questions in real time. All of my meetings are recorded and transcribed.
And so I can just take that meeting transcription and have it on file for Claude Code, which we'll touch on in just a sec. It's a really effective onboarding form. It gives me everything that I need to build out a working version one.
So at the end of it, I have all the responses and I have their knowledge based documents. So once those two things are taken care of, they're a paying client, it's time to actually build the agent. And this is where Cloud Code does most of the work for me.
Now we're inside of Cloud Code. I use anti gravity as my IDE, and we have it all hooked up and configured to the Trilet API.
That's how everything is automated from this point on. Once I have those onboarding form responses from my client, I can just paste this into Claude code and fire away. So I just say new client, onboarding form responses, and then paste everything.
Now there's two other things that I grab to send in that first message to Claude code. If you forget it, it's not that big of a deal because Claude will tell you, it'll get to a point, hey, looks like I need the workspace ID. Hey, do you have the knowledge based documents for me?
So I go to Trilit, I go to their workspace, and I say copy workspace ID. I say here is the workspace ID, and I'm just going to leave it at that. I have the knowledge based files ready.
We're just gonna let Cloud Code do its thing, and then it's gonna get back to me eventually saying, hey. I need those knowledge base files. I'm using the new Opus 4.8.
It came out yesterday. I have not tried this out with Sonnet yet. I've just been using Opus 4.7 and now Opus 4.8, but let's just send it off and watch as it builds out the agent for me.
So it says that it's going to start by loading the onboarding skill, which I have created. That's what you're seeing over here.
All these files right here, it's my dialed in system. So it's loading the onboarding skill and getting oriented. So what's going on here is it's loading in all of the documentation, my templates, everything into context to scaffold the entire build.
Now remember that third optional tool that I mentioned kinda toward the beginning of this video? It's Airtable. So inside of Airtable, again, is optional.
I just document the workflows, the agents, the clients, and the full activity log. Looks like all of the prerequisites look good. So it's doing a smoke test with the trill it wrapper that we have created.
This is the full checklist that it's gonna go through. So at this point, what I can do is I can take care of other tasks. I can answer emails.
I can go respond to people inside of my community. I can make videos for my social media. I can go grab a bite to eat.
I can do my laundry, all while Cloud Code is doing everything, all the heavy lifting for me. Once it verifies all of the prerequisites, it gets building and checking off this list.
It's gonna create my client repo. That way the whole environment is isolated from all of my other clients. Then it's gonna build the whole AI receptionist.
It's gonna insert the knowledge base, which it doesn't have yet, but I'm going to provide it when it gets to that point. And for my starter tier, the $95 a month plan that I have, it's a very basic tier. So we don't have any back end workflows like inside of n8normake.com to be used as tools.
I I do this because tools are like giving your AI agents the muscles to actually do things. In this case, for our starter tier, we don't have any back end integrations. We're not integrating with their CRM or anything like that.
It answers questions. It answers calls twenty four seven. It escalates calls.
It it handles call transfers, and it sends a booking link through an SMS message. Hey, guys.
Quick one. Inside Automate What Academy, Alice spent an entire weekend stuck on one automation she couldn't get working. And within twenty minutes of joining the pro community, we jumped on one of our daily calls and got everything resolved.
You If want that kind of support, whenever you hit a wall building AI automations for yourself or your clients, Automate What Academy is exactly where you need to be. The link is below. Let's get back to it.
There's no real time calendar scheduling on this tier, but it's still really useful. At this point, this is what I actually do. It's a little bit new to my setup, so I kinda forgot about it.
I open up that TruePros project. So this is the project that Cloud Code just created. Remember that own its own isolated environment.
We just did that. I'm So gonna open it inside of a new session. So I click open, and then I go over here to Claude Code.
Now back in our old session, let's see what it wants from me. Intake was parsed to normalized. Airtable clients record exists.
My repo is live on GitHub. So this is kinda it acts as my version control. Very handy feature.
Now my move is to relaunch for that client scoped session. I have the new session open. Is there anything that I need to say specifically, or can I just tell the new session to pick up where we left off?
I'm just gonna ask this old session really quick to see if there's anything I need to take with me over there. I'm just gonna copy this. It told me to copy and paste this over there.
So we're gonna paste and send it off to continue the build. Is this the most ideal approach to separate the Claude sessions like this?
I'm not sure. I'm still experimenting. But either way, it doesn't matter because it's still saving me hours of my time.
Looks like it wrote up the system prompt for us. It looked up its reference documentation, and now it's creating the agent for us.
K. This is where it's stopping for now because it needs those knowledge based files from me. So this is something that I usually get from my clients.
I'm going to paste both of them right here inside of the KB folder. It's already created for me. So we have service area and hours.
We also have just services and frequently asked questions. I'm gonna say done. Continue.
What it's going to do now is it's going to take both of those knowledge base files or however many knowledge base files we give it, and it's going to build out a new knowledge base inside of the AI agent. I'll show you guys what that looks like in just a sec.
Looks like Cloud Code just wrapped up. Now this is the part that used to take me hours. Now it's just a matter of minutes, and it's the same every single time.
That's what makes this repeatable. Let's go see what it built out for us. Hitting refresh, looks like it got the folders right.
So I have an active folder and testing folder. Currently, this guy is in testing. So let's click into it and see what it configured.
So we have the entire system prompt built out right here for us. And it even integrated the tool call for sending that SMS booking link. Can you see that right here?
So booking link SMS tool, it even configured that for me. It inserted the customer phone number, dynamic variable right there, and we also have the date and time information that we're giving to the AI agent so that it doesn't hallucinate. Everything here that you're seeing in the system prompt alone would have taken me hours, let alone all the other configurations that I need to do over here in the middle section.
Everything is pretty standardized over on my end. We have the welcome message. We have the knowledge base.
So let's take a look at the knowledge base that it spun up for us. So we have the business type. It's a residential heating, cooling, and plumbing company.
Here's the business description. Here's the phone number. We have hours and scheduling.
We have service and offerings. We have pricing and payment. We have intake questions.
These are the questions that are asked to the customer over the phone. Like, what's your name? What's the service address or at least the city?
Is this for heating, cooling, or is it a plumbing issue? We also have 14 frequently asked questions right here. And any other details, we just pasted that in right here.
So the agent can pull from this knowledge base as factual context. We have the LLM configured properly. I also have the fallback configured properly.
So my fallback that I use currently is Gemini 3.1 flashlight preview. The voice that I default to is Astra, and the fallback is this sitcom girl from Hume AI, and also the transcription model. The one I default to is Deepgram Flux, and the fallback is Deepgram Nova three.
As for the voice, I have all of this configured to my liking. We have call timing and limits properly set. So my max call duration that I set for clients just to start off is ten minutes, so six hundred seconds.
And then the max call message, I just say, hey, I appreciate the conversation, but we've reached our limit. That way it doesn't get abused and we eat up all of the credits and it ends up costing, right, my client a lot of money. I've got my post call analysis variables that we extract at the end of every single call that gets sent to my client via email at the end of every single call.
Post call actions, I actually told Claude Code not to do this right now. So email summary, what you can do, this is a really cool thing in Trilet, is you can actually send post call email summaries to three maximum emails. So so like the business owner, the general manager, maybe the on call technician.
I don't know. But you can just add those emails right here. It's really simple.
At the end of every single call, they get a beautiful call summary in their inbox where they can click a button and listen to the call. They can view the transcript, any extracted variables, all of that stuff.
And this is also configured properly. It's the SMS booking link. So I think this is just a fictional booking URL.
But essentially, when a customer does opt in like, yeah, I need to get on the calendar, cool. I just sent you over a booking link to your phone number. Go ahead and take a look and you can schedule there.
You can find the next available time. So that's for my starter tier. On all the higher tiers, I offer real time booking on the call.
Now this is version one. If my client ever wants to change anything or update their process, update their knowledge base, I just hop into Cloud Code, let Cloud Code know what changes need to be made, and then it has all of the documentation there. It pulls everything from the GitHub repo, and just like that, it gets to work and implements all of the necessary changes for me.
Now the last thing that I have to do, this is manual unfortunately, is we have to go acquire a phone number. It's really easy.
You just go over here to the left hand side, go to phone numbers, and then you can either bring your own phone number from Twilio, let's say, or you can just buy a number. And then depending on which plan you're on with Trilit, they offer complimentary numbers. The area code that I want to see for available numbers is 385.
It's here in Utah. I'm gonna say search numbers, and these are all of the numbers that I can claim. So I can acquire this first one, 4236422.
So I'm gonna say claim for free, and now the number shows up for us. All I have to do is go over here to the three dots, configure, I'll give it a label, TruePros AI receptionist, and then we select an agent.
So this is the one that we have. This is our version one. And then hit save configuration.
Now every single call that comes into this phone number right here gets hooked up to the AI receptionist. One thing I forgot to mention that's really helpful on sales calls is being able to do the math with the business owner or whoever's on that call, the general manager. Do the math with them on the call.
How many calls do you miss on average every single day or every single week? What's the value of your average customer? Is it $300?
Is it $500? Maybe it's a thousand. Then And you just crunch the numbers.
Help them see the return on investment. We're selling b to b here. So what you're telling me, John, is you miss a handful of calls every single day.
Let's say two or three. And you're saying that the average customer value is $300. And then you do the simple equation.
Two missed calls on weekdays, that's 10 per week. Four weeks in a month, that's about 40 missed calls every single month.
Multiply that by $300, that's the amount of revenue you could be missing. Please let me help you capture more of that.
I have the solution for you. It's ready to go. I can have it live in x days.
I can have it live in your business in forty eight hours or whatever your time frame is. Now let me zoom out and show you the entire system start to finish. Because once you see it as a repeatable process, you realize that you can run this over and over.
So here are the seven steps. Number one is you build a demo in five minutes. I showed you guys how to do that.
It's very easy. You bring that with you to the discovery call, to the initial consultation so that you can show it to them.
They can get their hands on it. They can hear it. They can talk to it.
And then you close the deal. Send them the onboarding form immediately. Keep that momentum going.
Create their Trilet sub account and you set up invoicing. Number five, you feed the answers to Claude code and it builds the agent for you. Number six, you review it, you attach the phone number.
And number seven, you give them the phone number, say, hey, here's version one. Go ahead, take it for a spin. Your client tests it out.
They get back to you with initial feedback. You make some modifications because it's not going to be absolutely perfect right off the bat. And then when they're happy, you go live and set up call forwarding.
That's the whole thing. One demo, recurring revenue. The same system that I used to turn one five minute demo with an auto shop into 159 a month on autopilot.
Now you can go build your first one. Everything I showed you today, that's the kind of stuff I post here. Real builds that you can copy.
If that's useful to you, subscribe and stick around. I'll catch you on the next one.
The Hook
The bait, then the rug-pull.
One auto shop. Five minutes to build the demo. Two days later, $159 landing every month. The bait is the time -- and the payoff is that Claude Code handles the build while the operator grabs lunch.
Frameworks
Named ideas worth stealing.
17:15list
The 7-Step Repeatable System
Build a demo in 5 minutes
Showcase on discovery call; close the deal
Send onboarding form immediately
Create Trillet sub-account and set up invoicing
Feed onboarding answers to Claude Code; it builds the agent
Review the build; attach phone number
Give client the number, collect feedback, go live with call forwarding
End-to-end process from first contact to recurring revenue
Steal forAny productized service sold to local businesses
16:05model
Missed Call ROI Math
Ask how many calls the business misses per week and their average customer value, then: (missed calls/week times 4 weeks) times average customer value equals monthly revenue leaking
Steal forAny B2B sales call where you need to justify a recurring fee
CTA Breakdown
How they asked for the click.
VERBAL ASK
18:15subscribe
“Everything I showed you today, that's the kind of stuff I post here. Real builds that you can copy. If that's useful to you, subscribe and stick around.”
Clean, low-pressure close. Sponsor (Automate What Academy Skool community) placed mid-video at roughly 9:07.
Add Modern Creator as a preferred source and Google shows you more of our breakdowns in Search, Top Stories, and AI Overviews. It only changes what you see, and you can undo it in your Google settings anytime.
Add to Preferred SourcesOpens your Google source preferences with us pre-loaded. Tick the box and you're done.
A creator rebuilds his voice-controlled AI assistant on Claude Fable 5.1 and shows it browsing a 3D map of his notes, drafting an invoice from memory, and placing a real phone call to a tire shop.
A $200/month Claude Max power user runs the real numbers on his own token usage, then rebuilds his workflow around open-weight models and tiered model routing instead of switching to a different single vendor.
An ex-Rolls-Royce ML engineer wires Claude Code straight into BytePlus's Seedream and Seedance APIs, skips the credit-metered ad platforms entirely, and live-builds a JBL speaker ad for $6.64.
A walkthrough of a free GitHub tool that reroutes Claude Code through free-tier API keys from NVIDIA and other providers once the paid usage limit runs out.