jevtest

Plain-English end-to-end tests for Android and iOS apps.
Write what a user does and what they should see. Jev works out the taps. CI replays every run exactly.

Documentation · Getting started · Steps · CI

app: build/app-debug.apk
device: { android: emulator-5554 }

tests:
  - name: Sign in
    fresh: true
    steps:
      - do: Sign in with email "${DEMO_EMAIL}" and password "${DEMO_PASSWORD}"
        expect: The home screen is showing
        see: Welcome, ${DEMO_EMAIL}

$ jevtest run login.yaml --lock frozen --out results
jevtest 0.9.0 · android · emulator-5554 · dev.jevtest.jevtest_demo · jev-1.13.0 · lockfile: frozen

▶ Sign in
  ✓ do: Sign in with email "${DEMO_EMAIL}" and password "${DEMO_PASSWORD}" (2.6s) — 3 action(s)
      → type "${DEMO_EMAIL}" into text_field 'Email'  (confidence 0.83)
      → type "${DEMO_PASSWORD}" into password_field 'Password'  (confidence 0.76)
      → tap button 'Sign in'  (confidence 0.93)
      → done  (confidence 0.96)
      ✓ expect: The home screen is showing — Jev 0.95
      ✓ see: Welcome, ${DEMO_EMAIL}
  PASS Sign in (7.1s)

1/1 passed in 7s
Jev: 5 decisions, 5 from lockfile, 0 asked live in 0.0s (0% of run time), $0.0000

Why jevtest

• Tests read like the spec. One action, then what should be true. do: takes a plain-English goal; tap:, type:, swipe:, scroll_to: and 20 more give exact control.
• Deterministic. Every Jev decision is recorded in a lockfile. The same screen always gets the same answer; --lock frozen replays a run exactly, with no network and no API key.
• Nothing assumed. No default device, no guessing what a typo meant, and settings with good defaults and strict limits. The whole test file is checked before a device is touched, and every problem is reported at once with what to fix.
• Real apps, real phones. Android emulators and phones, iOS simulators and iPhones. Native, Flutter, React Native and in-app WebViews, driven the same way. Animations stay on, and anything a step changes on the device is put back.
• Wait until, or fail. Each step waits until what it needs is on screen, up to its timeout, then fails saying what it waited for. No sleeps.
• Built for scale and CI. ${SECRETS} from .env or CI, shared test libraries, whole folders in one command, several devices at once, JUnit XML, JSON reports and failure screenshots.

Install

pip install jevtest

You also need the platform tools for your apps: the Android SDK and a JDK, and/or Xcode. A TypeSafe API key gives access to Jev. See Getting started.

Run

jevtest run tests.yaml --lock record --out results     # asks Jev about new screens, records the answers
jevtest run tests/ --lock frozen --out results         # a whole folder, replayed exactly from the lockfiles

How it works

A small agent on the device reads the accessibility tree in milliseconds. jevtest describes the screen as text and asks Jev, TypeSafe's decision model, to choose the next action and element from the options on screen. Jev never writes free text, so everything typed comes from your test file. The steps it works out for each do: are saved, and every later run repeats them exactly, without asking Jev. More.

Status

New, and tested end to end on a Flutter demo app with native and web screens: the Android emulator (API 37), a Pixel 4a (Android 13), iOS simulators (iOS 26) and an iPhone 17 (iOS 27).

Built to a high bar: a clean architecture whose layer rules are checked on every commit; strict mypy and strict pyright with no Any; every ruff rule; no function above a complexity of 10; 100% line and branch coverage in random order, with Hypothesis properties on every input jevtest reads; dependencies audited and every CI action pinned to a commit. Issues and pull requests are welcome; see Contributing.

License

MIT