Under the hood

How Velox is built

The tools behind every call — from hearing the caller, to understanding them, to answering in a natural voice.

What happens on a call

01

Listen

Speech-to-text turns what the caller says into words while they are still speaking.

02

Understand

A fast language model, grounded in your own documents, works out the right answer.

03

Reply

Text-to-speech answers in a natural voice, streamed so there is no awkward pause.

The stack

Each layer, and what it does

Speech-to-text

Hears the caller.

Language models

Decides what to say next.

GroqCerebrasNVIDIA NIMLlama 3MistralQwen

Text-to-speech

Speaks the reply.

Knowledge retrieval

Finds answers in your uploaded documents.

Backend

Runs live calls in real time.

FastAPIPythonUvicornWebSockets

Data & storage

Keeps agents, knowledge and media.

Frontend

This site and your dashboard.

Sign-in

Accounts and single sign-on.

NextAuth.jsGoogle OAuthGitHub OAuth

Hosting

Where it all runs.

Dev tools

What we build it with.

Planning

Where the ideas start.

Want to build on it?

Create a voice agent for your business and have it answering calls in minutes.

Get started