AI test suites / Record and run Android workflows
Record a workflow on a connected Android phone or emulator, review the captured steps, and use a selected AI profile to rewrite them as editable actions and expected results. Then run a case or suite with an agent on an authorized device, review its live evidence, and add a separate judge if you want another check.
01 / Record and review
Connect an authorized adb device or emulator and perform the workflow. Indagium records the inputs as a draft; you can inspect, edit, reorder, or remove steps before applying them to a case.
Use the device, emulator window, or Indagium mirror. The recorder observes input and prepares ordered rows for review; it does not inject actions.
Edit the captured rows yourself or ask a selected AI profile to turn them into human-readable actions and expected results. Check each suggestion against the evidence.
Nothing is added to the case until you apply the reviewed steps. The actions and expected results remain editable in the case editor.
The recording shows a Codex account for the rewrite. Use any AI profile configured in Indagium; the provider and model are selected in the app.
02 / Run and inspect
Choose a case or suite, an agent profile, and a connected authorized device. The run view follows progress and evidence; a separate judge is optional and can check every step or only failed steps.
Select the AI agent profile and adb device or emulator. Configure recording and evidence retention, then choose whether to use a separate judge.
See the current step, latest screenshot, recent logcat, and agent activity as the lane runs. Runs can also be paused or stopped.
Inspect each step's outcome and the evidence retained for that run. Failed results can be turned into issues kept locally or filed through a configured tracker.
This recording uses a Codex agent and a separate judge. Configure agent and judge profiles independently; a judge is optional.
Configure the workflow
Start with a case, then choose how its steps are authored, run, checked, and retained. The map below links to the full AI test suite guide.
Group cases under a suite with a target package, agent instructions, goals, and preconditions. Reuse shared step sequences in cases or setup and teardown hooks; named suite values are included as plain context for the agent. Suite guide.
Add steps by hand, draft them with AI, record a device workflow, paste log lines as checks, improve saved steps, or rewrite a retained recording. Review and edit previews before applying them. Authoring guide.
Steps can be performed by an AI agent, a tester, a script, or eligible recorded-input replay. Optional steps run only when their condition is present. Set step timeouts and retries, then stop, continue, create a draft issue, or pause after a failure. Step types.
Check for matching or absent logs, script results, and device state. Screen judge and Ask the judge checks use AI and need a configured judge. Add golden screenshots or reference logs as examples for a step. Checks and examples.
Choose agent profiles, model and reasoning effort, or an external MCP lane. Up to four distinct devices can run in parallel; lanes sharing a device run in sequence. Repeat cases 1, 3, or 5 times, set case and step retries plus tool-call budgets, and choose no judge, failed-step or every-step judging with inline, after-case, or background timing. Run guide.
Inspect live screenshots, logcat, and agent activity. Choose capture tabs and saved evidence. Reports include a case-and-step matrix, attempts, agent/judge disagreements, and flaky outcomes; compare runs, rerun failed cases or through a step, or export JSON, Markdown, and evidence ZIPs. Reports.
Expose typed-parameter scripts from the host shell or adb shell with ask-first, automatic, or setup/teardown-only permissions. Keep issues locally, add them to Notes, copy Markdown, or file through a configured tracker MCP server. Scripts · Issues.
Configure separate suite, run, and issue folders. The Free edition allows one suite with five cases; additional entries are locked, not deleted. MCP clients can author, run, compare, export reports, and handle issue evidence, with approval required for gated actions. Limits · MCP guide.
Questions
Yes. A recording starts as a reviewable draft. You can ask AI to turn it into editable actions and expected results that use the current screen and recording as guidance, or apply the recorded input as replay steps. A case can contain either kind.
No. Choose No judge, failed steps only, or every step when configuring a run. The judge has its own AI profile when enabled.
Suites, saved runs, and issues are kept in their configured folders on your computer. You can change those folders in Settings → General.
Suites, saved runs, and issues stay in their configured folders. When you ask an AI profile to rewrite a recording or run a test, Indagium sends the relevant prompts and evidence to the provider you selected. A recording rewrite can include captured inputs, screen readings, text entered, screenshots, and requested video frames; recognized password content is masked or withheld. AI use and judging are optional.
Continue exploring