Tell the chat what to test. Ranveo reads your app's source code, drives the live screen, writes the test, runs it and fixes it until it passes. When something breaks later, it tells you why.
In the demo above, nobody said which buttons to tap or by how much the setpoint moves. Ranveo found that in the thermostat’s source and on the live screen, wrote nine steps and ran them. Flows, limits, schedules and error paths work the same way.
Every test is a plain file in your repo: JSON, Python or TypeScript, reviewed like code.
Or ask for
Point it at the repository and the running app. It learns the screens, the object names, the state behind them, and the limits and error paths in the code.
Ask in a sentence and the AI writes the steps. Or record them in the live view, or program them in Python or TypeScript.
It checks every expected result on the screenshots, and the test has to fail with a key action left out. Only then does it count.
On your PC, in CI and on the team server, on real hardware, emulators and browsers, each test with its own app and its own clock.
It reads the commits since the last pass, the app’s log and the code. A bug becomes a ticket with its cause; a change made on purpose, an updated test to review.
The AI, the recorder and your own code write the same tests: one runner, one report, one history.
It reads the source and the live screen and writes the steps, and a second AI audits them. You review the file like any other change.
Tap through the app in the live view. Each tap becomes a step, and you add a check wherever you want one.
Loops, test data and your own helpers, with the same steps as functions. Run them from the same CI as your other tests.
test_setpoint_steps.py
from ranveo import test
@test("Each tap adds 0.5 °C, up to 30.0 °C")
def setpoint_steps(app):
app.set("heat_setpoint", 210)
for _ in range(18):
app.tap(name="btn_plus")
app.expect("heat_setpoint", 300)
# one tap more: it stays at the limit
app.tap(name="btn_plus")
app.expect("heat_setpoint", 300)
app.expect_text(name="lbl_setpoint", equals="30.0")setpoint-steps.test.ts
import { test } from '@ranveo/test';
test('Each tap adds 0.5 °C, up to 30.0 °C', async (app) => {
await app.set('heat_setpoint', 210);
for (let i = 0; i < 18; i++) {
await app.tap({ name: 'btn_plus' });
}
await app.expect('heat_setpoint', 300);
// one tap more: it stays at the limit
await app.tap({ name: 'btn_plus' });
await app.expect('heat_setpoint', 300);
await app.expectText({ name: 'lbl_setpoint' }, { equals: '30.0' });
});setpoint-steps.test.json
{
"name": "Each tap adds 0.5 °C, up to 30.0 °C",
"steps": [
{ "set": { "subject": "heat_setpoint", "value": 210 } },
{ "repeat": { "times": 18 } },
{ "tap": { "name": "btn_plus" } },
{ "end": true },
{ "expect": { "subject": "heat_setpoint", "value": 300 } },
{ "tap": { "name": "btn_plus" } },
{ "expect": { "subject": "heat_setpoint", "value": 300 } },
{ "expect": { "text": { "name": "lbl_setpoint" }, "equals": "30.0" } }
]
}One test, three ways to write it: each tap adds 0.5 °C, and the setpoint stops at 30.0 °C.
The test states up front what must be seen. A second AI, independent of the one that wrote it, looks for each of those results on the run’s screenshots. Then Ranveo runs the test again with a key action left out, and it has to fail. Only a verified test joins the suite.
Switched on per project on the team server.
A test can check
setpoint-everywhere.test.jsonset 22.0 → splbl_setpoint = ${sp}"${sp} °C" within 5 s
Ranveo tests web apps, mobile apps on Android and iOS, LVGL devices on real hardware and Unity apps, with the same tests, the same runner and one team server. And a single test can drive all of them at once.
Platforms
The same steps, locators and checks on every platform, wherever the platform can do them.
Headless Chrome, Chromium or Edge, and Firefox or Safari over WebDriver. Nothing to add to the page, and an error in its console can fail the test.
No SDKElectronUploads, reloadsAndroid apps on real devices and emulators today, iOS on devices and simulators next. No Appium server to install, and the companion app joins the same test as the device.
Early accessAndroidNo AppiumDesktop builds over loopback, firmware over a serial line, USB or the network. The MIT-licensed ranveo_lvgl library comes in from CMake or as an ESP-IDF component.
uGUI test builds with the Ranveo SDK for Unity, a UPM package that compiles only under the RANVEO symbol, so release builds stay clean.
Every test gets its own copy of the app and its own clock, so tests run side by side. When one fails, the AI compares the run with the last pass, the commits since and the app's log, says what changed and files the bug.
AUR-1A setpoint change holds until the next schedule event1.84 sAUR-3A schedule event added by a command is the next one0.96 sAUR-2Each tap raises the setpoint by 0.5 °C up to 30.0 °C2.31 sscreen = home · ≤ 10000 mshold.test.json:#1100preset = 0hold.test.json:#315hold = 0hold.test.json:#45heat_setpoint = 210hold.test.json:#55hvac_action = 2hold.test.json:#65name=btn_temp_uphold.test.json:#720name=btn_temp_uphold.test.json:#88heat_setpoint = 220hold.test.json:#106hold = 1hold.test.json:#115hold_until = 510hold.test.json:#125name=lbl_setpoint text = "22.0"hold.test.json:#1322$ ranveowall test tests --jobs 4 --junit junit.xmlAbove: the Aurora T1 demo tests, run on one Windows PC.
On premises
The server, its machines, the AI assistant, the bug tracker and the reports stay with you, without internet if you like. At run time it is one native binary: no Python, no Node.js, no database.
The target's screen, its input and a control panel built from its state, shared by everyone on the project.
Each test in its own process with its own clock. When every instance is busy, the run waits its turn instead of failing.
A failure becomes a ticket, failures with the same cause merge, and tickets move on to GitLab, Jira, GitHub or a webhook.
The first run records the picture; a person approves it. Differences are mapped onto the object tree.
Every run with its logs, failure pictures and traces, and which tests have started to wobble.
HTML, PDF and JUnit, plus a trace viewer built into the binary for the test that failed.
Nightly runs, staging and production, and secrets that never land in a test file.
Viewers, editors and admins. Who changed what is written down.
The AI compares a failed run with the last pass, the commits since and the app’s log, and writes down what changed. Failures with the same cause become one ticket, with pictures and a trace, and it goes on to Jira, GitLab or GitHub.
AUR-B8 The Fan tile shows Auto while the fan is offAUR-12 · fan.test.json · — · ×1 yesterdayAUR-B4 The kitchen tile shows °F after a restartAUR-9 · units.test.json · — · ×2 last weekAUR-B3 The schedule list scrolls past its last rowAUR-7 · schedule.test.json · — · ×1 last weekAUR-B2 An error dialog is cut off in GermanAUR-11 · dialogs.test.json · — · ×1 2 weeks agoAUR-B1 Away mode heats to 18.0 °C instead of 16.0 °CAUR-3 · away.test.json · — · ×4 3 weeks agoAUR-B0 The Humidity tile rounds 45.5% down to 45%AUR-14 · humidity.test.json · — · ×1 last monthAUR-A9 The clock skips 02:00 on the DST changeAUR-15 · clock.test.json · — · ×2 last monthWhy it failed
AI summary
Step 6 expects heat_setpoint = 215 after one tap, but got 220. Commit 7c2e91a changed SETPOINT_STEP from 5 to 10.
State
OpenAssigneeAssign to me
NobodyTest
AUR-2 Each tap raises the setpoint by 0.5 °C up to 30.0 °Csetpoint-steps.test.json
Seen
in 3 runs
first 2 minutes agoTicket
Labels
regressionthermostat
Click a bug to watch how Ranveo found its cause.
Pricing
Unlimited users, CI included, no run limits. A test takes one slot however many apps, devices or browsers it drives, and a run that finds no free slot waits in line.
Web and desktop LVGL for one person or a side project.
For teams that ship web and Electron apps only.
For companies under $150k yearly revenue.
Devices, web and Unity, in one test if you like.
For several sites and air-gapped labs.
Prices in USD. Every server starts with a 30-day trial of every platform and module. The demo video module is +$490 a year (included in Enterprise, watermarked in Free).
Start the server, connect your app and ask for the first test today. The trial has every platform for 30 days.
