# Turn on Self-driving

URL: https://testmode.ai/docs/self-driving/

> Self-driving is Testmode's agent for one project: it explores your app on its own, keeps a map of it, writes and fixes tests and investigates failed runs. Turn it on, choose where it works and how much, and tell it what matters.

Self-driving is the part of Testmode that tests your app without being asked. Once it is on, it explores the app a little every day, keeps a [map](https://testmode.ai/docs/map/) of its screens and journeys, writes and fixes test cases, groups them in plans and schedules them, and [investigates failed runs](https://testmode.ai/docs/investigate-failed-runs/). What needs a decision from you arrives in the [Inbox](https://testmode.ai/docs/inbox/); everything it did is listed under [Activity](https://testmode.ai/docs/self-driving-activity/).

**Note**

Self-driving comes with every paid plan, the Pro trial included. On the Free plan its pages are there, but it can't be turned on.

## Turn it on

You need the Editor role or above, and the project needs at least one [environment](https://testmode.ai/docs/environments/).

1. Click the gear beside **Self-driving** in the sidebar. **Self-driving settings** opens.

2. Under **Where it works**, choose the **Environment** it explores. It suggests a non-production one when the project has one: learn the app on staging and run the tests on production.

3. Optionally choose a **Login credential**. Self-driving signs in with it before it explores, so it can reach the screens behind your login. Without one it stays signed out.

4. Under **How much**, choose **Light**, **Standard** or **Thorough** (see below), and click **Save**.

5. Turn on the switch at the top of the page.

Its first exploration starts at once. The sidebar shows what Self-driving is doing beside its name: **Exploring**, **Analysing**, **On**, **Off** or **Paused**.

[Screenshot: Self-driving settings with Self-driving on: where it works (the Staging (known bugs) environment, no login, paths it stays out of), how much (Light, Standard selected, Thorough, with their pages and credits a month, and Map everything now), what it may do (Investigate failed runs on, Make them selected), what matters most with the team's goal, and Connect Slack. Caption: Self-driving's settings: where it works, how much, what it may do, the goal and Slack.]

**Caution: Exploring production**

Self-driving reads pages; it fills in no form but the login and clicks nothing but a cookie banner. Every page it opens is still a real request, so on a production environment list any paths whose links change data under **Paths it stays out of**, for example `/admin, /export, /orders/*/invoice`.

## Tell it what matters most

Under **What matters most**, write the project's goal in your own words, and click **Save**. Every exploration, review and investigation reads it. For example:

```text
Customers must always be able to find a product, add it to the cart and check out.
Search and the product pages matter most. We release on Tuesdays.
```

Saving a new goal starts a review, unless one ran recently (see below). **Review now** starts one straight away, whenever you click it.

## How much it does

Each level sets how many pages Self-driving explores a day and how often it reviews the project. The credits are for its explorations and reviews; the tests it plans and runs use credits like any other test.

| Level | Pages a day while it learns | Pages a day after | Credits a month, about |
| --- | --- | --- | --- |
| **Light** | Up to 10 | About 5 | 2,700 |
| **Standard** | Up to 25 | About 10 | 5,000 |
| **Thorough** | Up to 60 | About 20 | 9,000 |

- **Learning your app.** At first it explores up to the day's pages, starting from the screens it has seen a link to but not read yet. When none are left, or after a week, it moves on.
- **Keeping up with your app.** Then it explores once a day: screens that changed, new links, the first screens of important journeys and the screens it has not seen for longest.
- **Map everything now** lifts the day's limit: one exploration after another until it has read every screen it found a link to, for a day at most. **Stop mapping** returns it to its daily pace.

Schedules Self-driving plans may use up to 70% of your organization's monthly credits, shared by all your projects, so your own runs and CI always have room. See [How credits are counted](https://testmode.ai/docs/how-credits-are-counted/).

## What it may do

Under **What it may do**:

- **Investigate failed runs.** On by default. When a scheduled, CI or plan run fails, Self-driving re-runs the failures and files a verdict in the Inbox. See [Investigate failed runs](https://testmode.ai/docs/investigate-failed-runs/).
- **Changes to test cases and plans.** **Make them** (the default): it writes, fixes and switches off test cases, groups them in plans and schedules them, and lists every change under Activity with **Undo**. **Suggest them**: it says what it would change, and you make the changes.

It always explores and keeps the map up to date while it is on.

## When it reviews the project

A review reads the goal, the map, the test suite and what happened lately, and then acts on it. One starts:

- when you save a new goal, or click **Review now**;
- when an exploration finishes;
- when your pipeline reports a deploy (see [REST API](https://testmode.ai/docs/rest-api/#report-a-deploy));
- when Testmode notices the app changed: every few hours it compares the key screens with how they looked, and reviews when they differ;
- once a week, on Monday at 06:00 UTC, for projects with a goal.

Reviews are spaced out by level: at most one every 60 hours on Light, 36 on Standard and 24 on Thorough, so each level has a known cost. **Review now** is never held back. See [How Self-driving works](https://testmode.ai/docs/how-self-driving-works/) for what a review does.

## Turn it off or pause it

- **Turn it off** with the switch on the settings page. It explores, reviews and investigates nothing until you turn it on again; an exploration already going finishes. What it learned stays on the map, and your own test cases, plans and schedules keep running.
- **Pause testing** in the project's **Settings** stops everything that runs on its own at once: schedules, CI runs and Self-driving. See [Pause testing and protect data](https://testmode.ai/docs/pause-testing-and-protect-data/).

## Related

- [How Self-driving works](https://testmode.ai/docs/how-self-driving-works/): What it does on its own, and where you come in.
- [The Inbox](https://testmode.ai/docs/inbox/): Questions, verdicts and suggestions that need you.
- [Self-driving's activity](https://testmode.ai/docs/self-driving-activity/): Everything it did, with Undo.
- [The map](https://testmode.ai/docs/map/): What Testmode knows about your app.