> ## Documentation Index
> Fetch the complete documentation index at: https://docs.revyl.com/llms.txt
> Use this file to discover all available pages before exploring further.

# Tests via Recorded Session

> Record device interactions and generate test steps

Instead of writing steps by hand, interact with the live app and let Revyl generate test steps from your actions.

<img src="https://mintcdn.com/cognisim/fQuvYyuV8mxPZ97a/images/diagrams/session-compile.png?fit=max&auto=format&n=fQuvYyuV8mxPZ97a&q=85&s=2fbc2b1e4f93b9f56a95e2d3a1571f44" alt="Raw device actions flow through an AI pipeline to produce structured test steps" width="1800" height="600" data-path="images/diagrams/session-compile.png" />

There are two ways to do this: [recording live](#recording-live) inside the test editor, or [compiling](#session-compile) steps from a past session's history. Both produce the same kind of output, but fit different workflows:

|            | Recording Live              | Session Compile              |
| ---------- | --------------------------- | ---------------------------- |
| **Where**  | Inside the test editor      | From the sessions history    |
| **When**   | While building a test       | After an exploratory session |
| **Result** | Steps added to current test | New test or module created   |
| **Scope**  | Short, focused interactions | Full sessions or time ranges |

## Recording Live

### Create a new Test

From the **Tests** page, click **Create test** in the top right.

Fill in the creation modal:

* **Test Name**: a descriptive name for your test
* **Platform**: iOS or Android
* **App**: select an existing [app](/builds/index), or create one inline
* **Tags**: optional labels for filtering (e.g., `smoke`, `regression`)
* **Advanced Settings**: AI config and YAML import (optional)

After clicking **Create test**, you land in the test editor.

### Start the Recording

1. Start the device by clicking on the device screen or clicking the `Start` button
2. Click the **Record** button
3. Interact with the device: tap buttons, type text, swipe, navigate
4. Click **Stop** when you're done

Revyl captures your actions during the recording window and sends them through the AI compile pipeline. After a few seconds, you get a set of generated steps.

### Reviewing Generated Steps

After recording stops, a review panel shows each generated step with:

* **Step type** (instruction, validation, manual) that you can toggle
* **Description** in plain English, editable inline
* **Confidence score** from the AI

You can also:

* **Retry with feedback** if the results aren't right ("focus on the login flow", "ignore the settings detour")
* **Copy as YAML** for use in CI/CD
* **Save as module** to reuse the steps across tests

When you're happy with the result, click **Add to test** to insert the steps into your test at the current position. You can also **Save as module** to reuse the recorded flow.

## Session Compile

Session compile takes your raw device interactions and uses AI to generate structured, natural-language test steps. Explore your app freely, then let Revyl turn that exploration into a test you can run again.

### How It Works

1. **Start a session** and interact with your app normally
2. **Stop the session** (or let it idle-timeout)
3. From the sessions history, click **Make a test** on any session that has recorded actions
4. Revyl's AI analyzes your actions, groups them into logical steps, and generates natural-language instructions
5. **Review the generated steps** in the editor as described in [Reviewing Generated Steps](#reviewing-generated-steps).

### The Compile Pipeline

The compile runs through several stages:

`Queued` > `Preparing` > `Summarizing` > `Merging` > `Formatting` > `Complete`

Each stage shows progress. If you close the modal while it's running, a badge appears so you can come back to it.

### Partial Compile

You don't have to compile the entire session. Select a time range to compile just a portion, which is useful for long exploratory sessions where only part is relevant.

## Tips

* Testing via recorded sessions works best when:
  * You're exploring a new feature and want to turn that exploration into a regression test
  * A QA team member found a bug manually and wants to create a reproducible test
  * You want to bootstrap tests quickly without writing steps from scratch
* Keep recordings short and focused on a single flow. It's easier to record login, then record checkout, than to record everything at once.
* The AI works best when your actions are intentional. Avoid aimless scrolling or backtracking during a recording.
* You can record multiple times and build up a test incrementally.