Patent Pending · U.S. Provisional filed June 13, 2026

Ama

The on-screen mechanic that knows your car.

An offline, in-vehicle AI that replaces the check-engine light with a mechanic that explains what's wrong, why, and how to fix it — and learns your specific vehicle over time.

Talk to us See how it works
100% offline · no cloud Warns before the light Never changes your car without consent
292M
vehicles on U.S. roads
$500–$10K
cost of one unexpected repair
1 in 3
drivers can't afford that repair
The Problem

The check-engine light hasn't changed in 40 years.

One amber bulb for thousands of possible failures — and it only turns on after the damage is done.

Reactive

It only warns you after the failure has already happened — never before.

Generic

The same vague code for every car on the road. Nothing about yours.

Silent

An amber light and a number. No what. No why. No how to fix it.

The Solution

Ama is three mechanics in one.

The same diagnosis, spoken to the person in the seat — driver, do-it-yourselfer, or professional.

Driver

Plain language, proactive. No codes, no jargon — just what you need to know, before it becomes a problem.

Ama says
"Your brakes need attention in about three weeks — nothing urgent today."Tap for the why, or just let me remind you.
How It Works

Five agents, one offline brain.

Each vehicle system gets its own specialist agent. Together they read the car, reason about it, and explain it — entirely on a computer in the vehicle.

CAN Bus
& OBD
›
5 Domain
Agents
›
Ama
Knowledge Engine
›
Reasoning
& Render
›
In-Car
Display
Everything runs on a single trunk-mounted compute unit.No cloud. No subscription. No network required for any core function — your data never leaves the car.
Why It's Different

Not a code reader — a mechanic that learns your car.

Predicts before the code

Catches drift in correlated signals weeks early — a cylinder misfire creeping up 600 miles before any fault code would trigger.

Learns this car

A per-vehicle baseline, not a model-year average. Proprietary data no competitor can license — it compounds with every mile.

Fully offline

All AI runs in the vehicle. No cloud, no subscription, no privacy exposure — data never leaves the car.

Proof

We caught cylinder 7 — 600 miles before the light.

The reference build is a 2015 Dodge Charger Pursuit (5.7 HEMI) — chosen deliberately because its well-documented failure modes make the "caught it early" demonstration concrete and verifiable.

Cylinder 7 — misfire trend vs. learned baseline misfire trend fault-code threshold
fault-code threshold learned baseline Ama flags it here light sets
← roughly 600 miles of runway between Ama's early warning and when a conventional light would set →
1

Drift detected

Per-cylinder misfire trend rises against the learned baseline.

2

Cause identified

Correlated to cylinder-deactivation lifter wear — a known $4K failure.

3

Driver warned

Plain-language alert with a timeframe — weeks of runway.

4

Light would set

A conventional system illuminates only now — far too late.

Technology & Defensibility

A filed patent and data that compounds.

6 inventions

filed in the provisional application:

  • Predictive fault detection before any code sets
  • Context-conditioned per-vehicle baseline learning
  • Self-rebaselining for a modified or swapped subsystem
  • Offline, in-vehicle multi-model reasoning pipeline
  • Cross-agent freeze-frame event capture
  • In-vehicle agent-presence visualization

First-party data

Every mile a vehicle drives makes its own model sharper. The longer Ama runs on a car, the better it knows that car — and that history is data no competitor can buy or license.

The provisional dates the invention to June 13, 2026; the data moat begins the day the first unit powers on.

Founder

One founder. A full system, already built.

Wyatt Austin Naugle, founder of Shiro AI

Wyatt Austin Naugle

Founder & Engineer · Ava, Missouri
  • Built the entire Shiro multi-agent architecture solo — simulation engine, Unreal Engine integration, and the automotive system.
  • Authored the IP and filed the patent independently — a 50-page technical specification and a U.S. provisional patent, without a legal team.
  • Deep, hands-on systems engineering — CUDA / GPU pipelines, on-device LLMs, multi-agent reinforcement learning, and real-time rendering.
Where We Are

From architecture to filed IP — built solo.

Architecture

validated

Provisional patent

filed Jun 2026

Reference build

underway

In-vehicle prototype

next

Fleet pilot

with funding

We're building the mechanic every car should have come with.

Investors, fleet partners, and the curious — we'd love to talk.

wyattnaugle@shiroai.co