An agency signs its fifth client for an AI receptionist. The first four went fine on a reseller plan. The fifth wants the agent to book into a custom CRM, the sixth wants a second language, and the seventh asks why their invoice is higher than the reseller's own list price. The agency is now running a software business on someone else's platform, with someone else's limits and someone else's margins.
That's the moment this decision usually gets made. Ideally, it gets made before then.
Path 1: Resell an existing white-label platform
Best for: agencies with a handful of clients, standard booking needs, and no interest in owning the product.
Reselling means putting your brand on someone else's platform and charging clients more than you pay. HighLevel is the obvious example for agencies already running client accounts there. Its AI is billed per client location on top of the agency's platform plan: pay-per-use, an AI Employee Growth plan at $50 a month with 100 voice AI minutes included, or AI Employee Unlimited at $97 a month. Voice AI usage runs at $0.045 per minute for the voice engine, plus phone charges. To rebill AI to clients at your own price, HighLevel requires its $497-a-month Agency Pro plan (prices checked October 2026).
This is genuinely the right answer for many agencies. Setup is fast, the vendor maintains the platform, and you can start selling this month.
The limits: you can only offer what the platform supports, your margin is whatever's left after the vendor's price, and a pricing change on their side changes your business overnight. That last risk is real. Synthflow, once a common choice for agencies, retired its self-serve plans in 2026 and now starts at $30,000 a year on enterprise contracts. Agencies reselling it had to rethink their offer.
Path 2: Build a branded layer on Vapi or Retell
Best for: agencies with an engineer who will own the integration.
Vapi and Retell are developer platforms. Agencies using them typically wrap the API in their own branded dashboard: client accounts, agent configuration, and reporting under the agency's name, with Vapi or Retell running the calls underneath. Vapi charges $0.05 per minute for hosting and Retell $0.055 per minute for voice infrastructure, with speech, model, and voice costs on top. A standard all-in minute lands around $0.11 to $0.15.
This gives you far more control than reselling. You decide the features, the branding, and the pricing.
The limits: someone has to build and maintain the dashboard, the per-client setup, and every agent's prompts and functions. And platform fees scale with every client's minutes, which matters more at 30 clients than at 3.
Path 3: Own a multi-tenant voice platform
Best for: agencies with enough clients that platform fees eat margins, custom requirements resellers can't meet, or a plan to turn voice AI into a product of their own.
Owning the platform means the orchestration layer is yours, typically built on LiveKit, with your choice of speech, model, and voice providers. Orchestration drops from about $0.05 per minute on Vapi or Retell to about $0.01 on LiveKit Cloud, and you can switch providers when prices or quality change.
The limits: this is a real software project with a real upfront cost, and you're responsible for keeping it running. It's the wrong choice for an agency that's still testing whether clients will buy voice AI at all.
What a white-label voice platform actually needs
An agent that sounds good on a demo call is the easy part. A platform that serves dozens of clients under your brand needs these pieces, and they're where most of the engineering goes.
Isolated client workspaces
Each client's agents, calendars, phone numbers, call recordings, and data must be completely separate. One client's agent should never be able to read or book another client's calendar. On Dynaris, this is enforced at the tool level: each workspace runs its own tool servers, so isolation doesn't depend on a prompt telling the agent to behave.
An agent builder your team or clients can use
Configuring a new client shouldn't require an engineer every time. Rawk.ai is a voice agent builder where KeenCraft led development, including a latency pass that cut end-to-end response time from about 900ms to about 320ms by owning the pipeline. That combination, easy configuration on top of a fast pipeline, is what a reseller platform has to deliver.
Per-client numbers and calendars
Each client gets its own phone numbers, its own business hours, and its own booking connection, whether that's Google Calendar, a CRM, or the client's own scheduling software. On VoiceCake, dental clients' agents book directly into each practice's own software.
Usage metering and rebilling
You need to know exactly how many minutes each client used, so you can bill them with a margin. Without per-client metering, you're guessing at invoices.
Branded client dashboards
Clients want to see their calls, bookings, and transcripts under your brand, not the underlying vendor's.
Monitoring across all clients
One place to see failed calls, slow responses, and booking errors across every client, so you find problems before clients do.
The margin math
Platform fees are small per client and large across many. Using orchestration costs alone, and assuming each client averages 2,000 call minutes a month:
| Clients | Monthly minutes | Vapi or Retell (~$0.055/min) | LiveKit Cloud (~$0.01/min) | Monthly difference |
|---|---|---|---|---|
| 5 | 10,000 | $550 | $100 | $450 |
| 25 | 50,000 | $2,750 | $500 | $2,250 |
| 50 | 100,000 | $5,500 | $1,000 | $4,500 |
That's orchestration only. Speech, model, voice, and phone costs apply on every path. At 5 clients, the difference won't justify a platform build. At 25 to 50 clients, it can cover the build within the first year or two, before counting the value of features resellers can't offer. The full per-minute breakdown is in what an AI voice agent actually costs in 2026.
How to choose
- How many clients will use voice agents within a year? Under 10, resell. Over 25, run the build math.
- Do clients need anything the reseller can't do? Custom CRM writes, specific languages, unusual booking rules, or data kept on your own systems.
- Do you have an engineer who will own a Vapi or Retell integration? If yes, path 2 is a good middle step.
- Is voice AI a service you sell, or a product you want to own? If it's a product, owning the platform makes it an asset, not a line item on someone else's invoice.
What it costs to build
KeenCraft builds white-label voice platforms in QA-gated milestones, so each piece is tested before it's paid. A multi-tenant platform with isolated workspaces, an agent builder, per-client metering, and branded dashboards is priced as a fixed Statement of Work after a strategy call. A smaller first phase, such as isolated workspaces and metering on top of Vapi or Retell, can start lower and grow into a fully owned platform later.
Book a call if you can say how many clients you serve today and what your current platform can't do for them. A proposal typically lands in 2 to 5 business days.