AI · Telecom·Florida, USA
Voxtell AI
White-label AI voice for telecom partners
Voxtell sells AI voice agents through telecom resellers, which means the product has to carry someone else's brand and survive real phone calls. Fluid Web migrated the voice pipeline between providers in three weeks, replaced the authentication layer, and built the white-label theming that lets partners sell it as their own.
What we walked into
Voxtell's go-to-market runs through telecom partners and resellers rather than direct to small businesses. That changes the engineering requirements substantially: the product has to be deeply rebrandable, and it has to be reliable enough that a partner will put their own name on it.
Fluid Web was brought in as full-stack engineering on the AI voice side, working on the calling pipeline, telephony, transcription, messaging and the white-label layer.
The problem
The platform was built on Retell, which was working, but was behind on the AI capabilities the product roadmap depended on. Staying put meant the roadmap was capped by someone else's release schedule.
Separately, the authentication layer was causing friction that a library choice should not cause — a small problem that quietly taxes every feature built on top of it.
Migrating the voice pipeline in three weeks
We moved the production voice pipeline from Retell to Vapi over roughly three weeks. The honest framing matters here: call reliability was comparable on both, so this was not a rescue from a failing provider. It was a capability decision — Vapi had the AI features the roadmap needed and Retell did not have them at the time, and the commercial terms were favourable.
Vapi sitting on top of Twilio meant the telephony integration stayed coherent through the move rather than requiring a parallel rebuild of the call path.
- Production voice pipeline migrated in approximately three weeks
- Driven by AI capability access, not by reliability problems
- Telephony continuity maintained through Twilio underneath
- Live calling, transcription and post-call messaging preserved across the move
Replacing the authentication layer
We also migrated authentication from NextAuth to Better Auth. This is the kind of work that never appears on a roadmap and pays for itself repeatedly — the previous library was not serving the product well, and every feature built on top of a difficult auth layer inherits that difficulty.
Replacing it was a contained piece of work with a compounding return across everything built afterwards.
White-label deep enough for partners to sell
Reseller businesses fail on shallow white-labelling. If a partner's customers can tell they are looking at someone else's product, the partner will not sell it — and a logo swap is not sufficient to hide that.
The theming layer goes considerably deeper, so partners present a product that reads as theirs, with multi-tenant patterns underneath built for reseller scale rather than for a handful of accounts.
- Theming deep enough that partners present the product as their own
- Multi-tenant structure designed for reseller distribution
- Partner-facing surfaces alongside the end-customer product
- Branding as tenant configuration rather than per-partner builds
Voice, transcription and messaging on one thread
An AI receptionist that only handles the call is half a product. Calls produce transcripts, transcripts produce follow-ups, and follow-ups happen over messaging — all of which needs to resolve to the same customer rather than three disconnected systems.
We built the calling, transcription and post-call messaging paths to share that context.
The hard parts
Migrating a live voice product
Voice is the least forgiving surface to migrate. Calls are synchronous and real people are on them, so there is no window where a degraded state goes unnoticed.
Choosing to move for capability, not crisis
Migrating a working system on the basis of future capability is a harder argument to make than fixing something broken, and it requires being honest that reliability was not the reason.
White-label depth versus maintainability
Every additional layer of brand customisation is another dimension of configuration to keep working. Going deep without letting partner-specific behaviour fragment the codebase is the real constraint.
How it ran
Embedded as full-stack engineering
Joined on the AI voice side covering the calling pipeline, telephony, transcription and messaging.
Provider migration
Production pipeline moved from Retell to Vapi in roughly three weeks, retaining Twilio telephony underneath.
Authentication replacement
NextAuth replaced with Better Auth, removing friction that was taxing every feature above it.
White-label depth
Theming and multi-tenant patterns extended for partner distribution.
Results
Voxtell's voice pipeline moved to a provider whose AI capabilities matched the roadmap, in approximately three weeks, without losing live calling, transcription or messaging.
The authentication layer was replaced with one that stopped taxing feature work, and the white-label theming went deep enough for partners to sell the product under their own brand.
Partners can offer AI receptionists without building voice infrastructure, on a multi-tenant foundation designed for reseller scale.
Built with
Scope of our work
Our work covers the AI voice pipeline, telephony, transcription, messaging, authentication and white-label theming, with partial involvement in the partner portal. We did not build the NetSapiens integration or the marketing site, and we do not have visibility into Voxtell's call volume or partner numbers.
Got something similar?
Tell us where delivery is stuck. We will give you an honest read on scope and shape before anyone talks contracts.
Book a technical review
