Vix AI

MeetVix AI

First responders don't need another chatbot. Vix is an assistant that understands incident protocols, apparatus, and the tempo of a working shift. Vix 1.0 Preview is live inside Aevix Operations today, with 1.0 Pro and an API shipping later this year.

By 2032, Aevix will be running on our own foundation models, built for the field and fully on-device. That means Vix works when there's no internet and no signal, because a fire scene, a rural call, or a basement rescue can't wait on the cloud.

See the roadmap
Vix 1.0 Preview
What's the status on Engine 2?
Engine 2 is in service. FF Davis ran the morning check at 06:14. All SCBA bottles green, one flashlight flagged for a battery swap. Last out-of-service was three days ago for the quarterly pump test. Want me to open a work order for the flashlight?
Ask Vix anything...

Vix AI can make mistakes. Please double-check responses.

Where we are, where we're going

The arc from shipping product to a foundation model built from scratch.

I
TodayShipping

Vix 1.0 Preview is live inside Aevix Operations.

Vix 1.0 Preview answers operational questions in plain language. Protocol lookup, incident briefings on demand, and apparatus and resource reference specific to your department. Any agency running Aevix Operations has it right now.

II
Near-termLater this year

Vix 1.0 Pro and the API expand from answering to noticing.

Vix AI Pro and the API hold context across an active call instead of starting fresh each prompt.

It flags protocol conflicts during incident documentation before anyone asks.

Vix AI Pro and the API also draw on the rest of the Aevix platform (Operations, Stride, Orision), so answers reflect the same operational picture your command staff already sees.

Each department's Vix AI Pro instance and API access tune to their own SOPs, so the intelligence feels local instead of generic.

III
2032Six-year build

A foundation model built from the ground up for emergency services.

Most AI in public safety is a general-purpose model fine-tuned to sound like it understands the work. Ours is different in kind.

A model from scratch. New pretraining, a tokenizer built on field language, and architectural choices tuned to what actually shows up on a call: apparatus status, protocol language, CAD records, incident taxonomy, voice cut with sirens and engine noise.

It runs on the apparatus. On-device inference, no network, no cloud. A fire scene, a rural call, or a basement rescue can't wait on a signal that isn't there.

Call it a six-year build. We'll tell you where we are on it.

Active Research

What We're Working on Right Now

Runs on the Apparatus

We're building Vix to run on the truck itself. No cell signal, no cloud, no data leaving the device. When the connection drops, Vix keeps working.

Built for Cab Noise

Most voice recognition is trained on clean studio audio. We train ours on what an apparatus cab actually sounds like: sirens, air horns, radio cross-talk, diesel, wind through an open window.

Knows When to Say 'I Don't Know'

A confident wrong answer on a call is worse than no answer at all. We teach Vix to say it doesn't know when it shouldn't guess, and we measure how often it gets that right, not just how often it responds.

Tested on Real Calls

We test Vix against recorded incidents with real radio traffic and real crew phrasing, not clean benchmarks. The calls it gets wrong become training data for the next version.

Partnerships

Building the intelligence layer for emergency services.

If you invest in deep technical infrastructure or run a research program in public safety AI, we want to talk. Aevix is selective about who we work with and open about where we are today.

Start a conversation