> For the complete documentation index, see [llms.txt](https://docs.qyrus.com/device-farm/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.qyrus.com/device-farm/ai-session/aisession-faqs.md).

# AI Session FAQs

## 1. What is a Device Farm AI Session?

You give an objective in plain language, pick a real device or a URL for web, and an AI agent drives the app step by step: tapping, typing, scrolling, and verifying what it sees on screen. Every action, screenshot, and verdict is recorded into a report you can replay and share.

## 2. How long can a session run? Does it fall apart on big flows?

Sessions can run for 3 to 4 hours, depending on device and browser timeouts. The agent manages its own history in the background as the run grows.

## 3. What are milestones and when does the agent create them?

When an objective asks for several separate outcomes, the agent splits it into checkpoints and works through them in order. The trigger is breadth, not length. This keeps the agent from declaring success after only the first part works and lets later milestones depend on earlier ones.

## 4. What is a nudge?

A nudge is a short note on one step about what may genuinely differ on the next run, so the agent adapts instead of stopping. It does not change the recorded action or step order. You can send one to a live session or save one on a step of a saved test.

## 5. When should I add a nudge to a step?

Add a nudge when content changes between runs, a target moves or is renamed, a step may not apply every time, or the screen cannot show success clearly. The agent can also write nudges when it notices a legitimate variation during a session.

## 6. Can I edit a saved test before re-running it?

Yes. Steps can be removed, reordered, and given nudges. Changes apply to future runs only; past reports stay unchanged.

## 7. What happens to each recorded step during a re-run?

The re-run uses the recorded locator first. If it cannot find the element, the agent reads the live screen and re-aims the recorded action. A bounded rescue attempt can then recover the step. If recovery is not possible, the step is skipped with a reason.

## 8. What do the report statuses mean?

* **Passed:** The step replayed successfully.
* **Failed:** The expected result did not occur.
* **Skipped:** The step could not apply, with the reason saved.
* **Rescued:** The recorded path changed, but the agent reached the same outcome another way.

## 9. Is rescue safe?

Yes. The rescue agent receives a fixed number of turns and steps. It cannot become a fresh exploratory session. If it cannot recover within its budget, the step is reported honestly.

## 10. What is Record for Automation and why is it marked Experimental?

Record for Automation captures an element locator for each action so the session can be saved and re-run as an automation test without the agent re-deciding every step. It is marked Experimental because some apps may expose no locators or only partial locator coverage.

## 11. What does the effort setting control?

The effort setting controls how long the agent may wait for a screen’s page source while recording. Higher settings allow more time for large or deeply nested screens.

## 12. What happens to screens that exceed the effort budget?

They are remembered for the app and are not retried unnecessarily. Steps on those screens can still replay by screen position instead of by locator.

## 13. Does recording slow the session down?

Yes. Recording adds a locator read to each step, so the first recorded session may run more slowly. Later re-runs can recover that time through locator replay.

## 14. What does a run cost?

Run costs depend on model usage, session length, and the selected device or platform. Intelligence costs are separate from device-time and platform costs.

## 15. iOS or Android: which is better for sessions and recording?

Both platforms can be used. The time required to retrieve page-source information can differ by platform and app, so a higher effort setting may help when recording slower screens.

## 16. How do I share or escalate a run?

Runs provide actions such as Copy run ID, Copy deep link, and Create JIRA ticket when an integration is configured. The deep link opens the report directly for anyone with access.

## 17. Can I drive AI Session from the command line?

Yes, with the Qyrion CLI. Use it to run objectives on mobile devices or web sessions, view and re-run sessions, access artifacts, and work with reports.

## 18. My Android app renders a black screen. What can I do?

The app may be using secure-screen flags that blank screenshots and the device stream. Contact your internal Qyrus support or platform team for approved troubleshooting guidance. Do not distribute internal tools or links to clients or end users.