Your Android Automotive app, on every head unit, reviewed before it ships

Upload an APK. Auto App Test runs it on real Android Automotive devices in the cloud - every AAOS version, resolution and day/night theme - and an AI agent drives your app to review each screen. A scored verdict and a shareable report in minutes.

Request access See the pipeline

Real AAOS 14, 15 and 17 system images · verdict in minutes · advisory by default

AAOS 14 · 1920x1080 · day/night verdict: PASS

Every head unit your users actually drive

One upload fans out across AAOS versions, resolutions, densities and day/night themes.

AAOS 14, 15 and 17 · with and without OEM compatibility mode · day and night · your choice of resolution and DPI

Why it pays off

A bug shipped to a head unit is expensive. A bug caught before release is a line in a report.

Head units disagree about everything: resolution, density, aspect ratio, day and night themes, OEM compatibility modes. Testing that spread by hand is slow, repetitive and easy to skip under deadline pressure - and a field report from a vehicle costs more than a release cycle of automated runs. Auto App Test runs the entire matrix on every build, so regressions are caught while they are still cheap.

From APK to verdict in five steps

The same fixed pipeline runs on every device in the matrix. No test scripts to write, no flaky suites to maintain.

install

Your APK goes onto a real AAOS system image over adb - genuine rendering, media sessions and behaviour, not a rough emulation.

launch

The app is started and smoke-checked on every configuration: does it come up, stay up and reach its first frame?

screenshot

Screens are captured across the whole matrix and compared against your baseline, so visual and DPI regressions surface immediately.

a11y lint

Deterministic checks flag clipped text, overlapping elements and touch targets too small for a moving car.

ai review

An AI agent drives your app across every navigation destination and settings screen, judging each one the way a user would. It returns structured findings, a 0-100 score and a self-contained report - in the portal or over the API.

Judge the judge

This is an unedited report from a real validation run of an open-source AAOS app - install to verdict, exactly as the API returns it. We ran an active AI review against a public weather app and it came back with a score of 17. We would rather show you a hard grader than a vanity number: the low score is the point. Those nine findings are real touch targets too small for a moving car and controls a screen reader cannot name, and the report links every one to the screen it was found on.

View the full report
Auto Weather v1.1.0
WARN
17 / 100
8 major
1 minor
9 findings across 8 screens
AAOS 14 (Cuttlefish) active review 2026-07-10

Watch the agent test

An active review of Auto Weather on AAOS 14, recorded as it ran - the Planner picks destinations, the Actor drives the device, the Verifier checks each screen.

Run a1a8c95d · Auto Weather v1.1.0 · AAOS 14 Cuttlefish · active review

Planner Reads the accessibility tree and decides which screens to explore next, prioritising untested destinations.
Actor Issues adb tap, swipe, type and key events. Retries on missed targets using the updated tree.
Verifier Independent pass after each action - checks contrast, touch target size, label presence and distraction rules.

Take the wheel when you need to

Automation catches the regressions, but sometimes you want your hands on the screen. Stream any device in the pool straight to your browser and click through your app live - no SDK, no cables, no head unit on your desk. It works from the passenger seat too: the native iOS app streams the same live session to your phone, and now starts reviews and shows their verdict, score, findings and screenshots on the go.

Agent-first. API always.

Everything the portal does is available over the Developer API - and your AI agent gets its own seat: a first-party MCP server with real adb access to pool devices.

From CI

release.yml · ci
$ curl -X POST https://api.auto-apptest.com/v1/validate \
    -H "X-API-Key: $AAT_KEY" \
    -F apk=@app-release.apk \
    -F target_id=aaos14-cuttlefish -F target_id=aaos15-cuttlefish \
    -F review_mode=active \
    -F callback_url=https://ci.example.com/hooks/aat

{ "validation_id": "val_0142", "status": "queued", "cells": 2 }

From your agent

mcp session
$ claude mcp add --transport http autoapptest https://mcp.auto-apptest.com -H "Authorization: Bearer $AAT_API_KEY"

agent> submit_validation aaos=14 resolution=1920x1080
lease granted · aaos14-1 · adb ready

agent> adb shell input tap 640 360
agent> screenshot
home.png · 1920x1080 · captured

agent> get_verdict val_0142
WARN · score 78 · 6 findings · report ready

Lease a real device in seconds, drive it over adb, pull screens and scored verdicts - the same pipeline your CI uses. Submit an APK, poll for a verdict and report, or take a signed validation.completed webhook when it finishes.

API reference MCP setup

Frequently asked questions

Common questions about supported devices, timings and access.

Which AAOS versions are supported?

Auto App Test runs on AAOS 14, 15 and 17. AAOS 14 and 15 are each available in two variants: standard and OEM compatibility mode (compat). AAOS 17 ships as a single standard build. Compat mode mirrors the behaviour of head units that ship with an OEM adaptation layer on top of stock AAOS, which can differ significantly from the standard build in layout, touch target sizing and theme handling.

How long does a validation take?

A passive validation - install, smoke check, screenshot sweep and a11y lint - finishes in under two minutes for a typical APK across a single device target. An active AI review that drives the app through its navigation tree adds roughly two to five minutes per target depending on app complexity. Running across the full device matrix in parallel does not add proportional time; all targets run concurrently.

What does the AI review actually check?

The AI agent drives your app screen by screen - tapping through navigation destinations, settings panels and media controls - and evaluates each state against AAOS UX guidelines: readability at arm's length, touch target sizing, day/night theme consistency, distraction-optimisation compliance and overall layout correctness. Results are advisory by default: vision and accessibility tree findings are reported as warnings unless you opt into gated verdicts (ai_gated_fail=true) that can block a release. Hard failures from deterministic checks (install failure, crash on launch, a11y lint critical) always gate regardless of that setting.

How do I get access?

Auto App Test is in early access. Send an email to apps@hirschmann.io with a brief description of your app and which AAOS versions you target. You will receive an API key and portal credentials within one business day.

Can my AI agent use Auto App Test directly?

Yes. There are two integration paths for agents. The Developer API lets any HTTP-capable agent submit APKs, poll verdicts and receive signed webhooks using an API key. The MCP server goes further: your agent can lease a real AAOS device from the pool, drive it over adb (tap, swipe, type, screenshot) and fetch structured verdicts - all within a single MCP session. See the API reference and MCP setup guide for details.

Ready to test before you ship?

Get early access to Auto App Test and put your next Android Automotive release through its paces.

Request access