Ten agents that work from your real data.
Each agent does one job. Two of them use a live browser. The rest read results you've already recorded, and a model helps where it's useful. When there isn't enough data to say anything, the agent tells you that and doesn't guess.
How agents run
You run an agent from the Agent page or the API. You can also chain several into a workflow: steps that don't depend on each other run side by side, and a step that depends on another waits for it.
Every run is saved with its input, its output, any error, and the credits it used, so you can look back at what an agent did. The Explorer can take a few minutes, so it runs in the background and shows as running until it finishes.
Agents run when you start them. Nothing runs on its own in the background.
The agents
What each one does, and what it needs to work.
Explorer
explorerWhat it does
You give it your app's URL. It opens the app in a real browser, clicks through pages, fills in forms with test data and notes what happens, including JavaScript errors and broken pages. It then writes test cases from the steps it took.
Good to know
It stays on your site, never reaches private networks, and is never offered controls like delete, pay, checkout, log out or invite. It stops at 30 steps or 5 minutes at the latest. It can sign in only with a test account your administrator set up. The test cases are saved as drafts for a person to approve.
Drift Detector
drift-detectorWhat it does
It opens a page in a real browser and saves a snapshot: the title, the visible text, and every button, link and field. On the next run it takes a new snapshot and lists what was added, removed or reworded, then classifies each change as intentional, suspicious or critical.
Good to know
The first run only saves the baseline. If nothing changed, it says so and doesn't ask a model anything. It needs a browser on the server.
Flaky Manager
flaky-managerWhat it does
It counts, from your recorded results, how often each test flips between pass and fail, and whether the same commit both passed and failed. That is the clearest sign of a flaky test. A model then suggests a likely cause from the real error messages. With auto-quarantine on, flaky tests are tagged so their failures stop blocking verdicts, and they are released after five clean runs in a row.
Good to know
A test needs at least five results before it is judged, and the agent tells you when it doesn't have them. Failures it ignores are always listed in the verdict.
Bug Reporter
bug-reporterWhat it does
It takes a failing test, or every failure found for a pull request, and files a bug with the error, how often the test failed recently, and the first failing commit when your runs record one. If you connect Jira, it creates the Jira issue and keeps its key.
Good to know
It won't file a second bug for a test that already has one open, and it skips failures that look flaky. A model writes the report text, and the failure counts in it come from your data.
Strategist
strategistWhat it does
It reads your test and run data and points out where coverage and testing effort are lopsided: for example, a module with many tests and no failures, next to a critical area with a couple of tests and a poor pass rate.
Good to know
It works from the data already in the project, so it is only as useful as the results you have recorded.
Execution Optimizer
execution-optimizerWhat it does
Given a description of a code change and the files it touches, it picks the tests most likely to be affected, ordered by risk, and says which it would skip and why.
Good to know
It works from your test list and the change you describe. Tests linked to code with covers tags give it the most to go on.
Perf Analyst
perf-analystWhat it does
It reads your recorded performance metrics (response time, throughput, error rate, LCP, TTFB) and flags where a metric has drifted from its own history, so you don't have to set a threshold for every one.
Good to know
You need metrics recorded in RuyQA first. It runs when you run it.
Data Synthesizer
data-synthesizerWhat it does
You describe a schema and how many rows you want. It returns realistic, varied test data, so your tests don't run on rows like 12345 and lorem ipsum.
Good to know
It generates data from the schema you give it. It never reads production data.
Visual Analyst
visual-analystWhat it does
It takes a description of a baseline screenshot and of the new one, for example from your visual testing tool, and judges which differences are intended design changes and which look like regressions.
Good to know
It reads descriptions and cannot look at the images itself. It refuses to run without both descriptions.
CrossBrowser
cross-browserWhat it does
It compares the results you've already recorded for the same test in Chromium, Firefox and WebKit, and separates failures that only happen in one browser from ones that happen everywhere.
Good to know
A test needs results in at least two browsers to be compared. It does not run the tests itself.
AI you can use without an agent.
Generate a test case
Describe what you want to test in plain language. RuyQA drafts a test case with a title, steps, expected result and priority. You edit it, approve it and run it.
Refine a test case
Pick an existing test case and ask for improvements. RuyQA rewrites the steps to be clearer or more complete, and you decide whether to keep the new version.
Analyse a failure
Paste a failure log. RuyQA suggests a likely root cause and a fix for that specific failure.
Ask your data a question
Ask something like "Which modules have a pass rate below 80% this week?" and RuyQA answers from your project's data. The queries it writes are limited to reading.
AI credits depend on your plan.
Agent runs, test generation and analysis draw from your monthly credits, and each run records what it used. Every plan can run the agents within its credits.
Try an agent on your own project.
A free account is enough to run the agents within the free credits.