Brighten Voice Assistant

Agent Platform Features

Representing five years of work, and tens of thousands of hours of iteration and testing


Agents always respond within 250 milliseconds, through a combination of unheard of concurrency, dynamic thought, and asynchronous dialog management


Huge models, unlimited custom entities, contextual speech, rapid training all contribute to our industry leading accuracy

robustly available

Our agents pushes as much of its brain as it can to local devices, so it can be available 24/7, rain or snow, network or no network. 

partial recognition

Using their intelligent contextual recognition, our agents can understand partial sentences, and state fully and dynamically form expectation of what happens next

Conversational Repair

When our agents sense a misunderstanding, they will ask for clarification

Live learning

Almost unbelievably, our agents learn many things from every interaction.  Something as profound as individual local dialect variance, something as important as the name of the new pizza place.

Agent Architecture

Great experiences from a world class, patented system 

World-Class Experiences

 All of these features work in concert to create graceful, highly productive experiences for your customers.

Imagine instant voice search for your product that learns what people are asking for automatically, and let’s you know.

Imagine a voice navigational system that senses when people need to stop to eat, and is always ready to show them the way home, so your customers are never lost, even when they don’t have an internet connection.

Imagine the magic of having all of your devices learning and blending in with local dialects, across the planet.

All of these things are possible when you start the journey with us

Radically powerful on device 

Studio quality audio

Noise filtering, level matching, and voice activity detection give our agents good ears for listening

streaming recognition

Our agents listen even when thinking about the last thing you said 


Stateful dialog manager including turn taking, support for interruption and barge-in 

World State

Agents continually listen to the state of the world around them.  In a navigation scene, that includes geo location and who is in the car.   All state stays on device, where consumer privacy is guaranteed.

Language from knowledge

We use AI models to derive language expectations live, as they listen to your customers.  

Much more intelligent than traditional systems given model sentences, because as with humans, language expectations are derived live, from the state of the world around our agents, the context provided by the knowledge graphs, and the stimuli coming in from the mobile agent running on device.

host & Seeding 

Binary installations make injecting our intelligence into your host apps straightforward and simple.  

Storage + a connection to the internet, and an intelligence seed, which when open, creates the right connections to generate the right local intelligence from.    

Our knowledge creator tool creates these seeds while working with our team of knowledge workers.

Distributed intelligence node

Each agent maintains a lightweight connection with our distributed intelligence