Describe the core flow in one sentence. A tester uses your release APK on real Samsung phones, runs the same path again with the network off, fonts at 130% and the phone in Japanese, and returns a report your coding agent can act on. No scripts, no SDK, no device slots.
claude mcp add autony https://api.autony.dev/mcp \
--header "Authorization: Bearer autony_sk_…"
No coding agent? Upload the APK in your browser — same report. · 2 free runs on your own app · prices below
“Create a note and save it.” The tester reads the screen and finds the way, step by step, until it's done. One named flow, one phone per run, assigned automatically.
The same path, again, with the network off, fonts at 130% and the phone in Japanese. On our Galaxy phones (Android 13–15), not an emulator.
Markdown with evidence links for Claude Code or Cursor; HTML with screenshots for you. Your app's text can't steer the tester.
A verdict, defects ranked by how much they hurt the user, each with a screenshot, the log line and a suggested fix — and the exact path the tester took, so “no defects” still shows what was covered.
| Criterion | Device clouds | AI test authoring | autony |
|---|---|---|---|
| You write | Appium / Espresso scripts | Test steps in plain English | One sentence |
| Who runs it | Your scripts, their phones | Their AI, often their phones | Our tester, our phones, offline / fonts / locale re-runs |
| Output | Pass/fail, video | Pass/fail, screenshots | Defects with severity, suggested fixes, Markdown for your agent |
| Pricing | Per device slot or minute | Per seat or credits | Per run, from [PRICE/RUN] |
One run = one test of one flow on one phone, about 5 minutes once it starts. Runs that reach the phone are charged, even with no defects. If the install fails or our lab can't complete the run, the run is returned. Runs never expire.