> Documentation index: [/llms.txt](/llms.txt). Every page as one file: [/llms-full.txt](/llms-full.txt).

# Build an environment

Build an environment [#build-an-environment]

An environment is a test you keep: the checks every answer must pass and the cases the agent plays. Every new version of an agent is measured on the same test.

<img alt="The Environments page with the Refund questions environment for the Refund Policy Assistant: 2 people, 2 checks" src="__img0" />

Create an environment [#create-an-environment]

1. Open **Lab** > **Environments** and select **New environment**.

2. Under **What is it for**, pick **My environment**, or **A catalog entry** to publish it for others, and **Who is it for**: **One of your agents**, or **Any agent**. This cannot be changed later.

3. Under **Checks**, pick what every answer must satisfy, or select **Suggest checks**. **Rules** pass or fail exactly; **Judged by AI** covers what rules cannot catch.

   <img alt="The New environment dialog on the Checks step" src="__img1" />

4. Under **Default inputs**, choose **Simulated people** or **Fixed inputs**, then add the cases. Select **Attach file** on a simulated person, or the attach icon on a fixed input, to give a case the documents it brings.

5. Under **Review & save**, enter a **Name** and select **Create environment**.

| Default inputs       | What happens in a run                                                                                                          |
| -------------------- | ------------------------------------------------------------------------------------------------------------------------------ |
| **Simulated people** | Realistic people chat with the agent the way your users would, each with a goal of their own. Needs an agent picked in step 2. |
| **Fixed inputs**     | The same messages go to the agent every round, and each answer is checked.                                                     |

Changing an environment's checks or cases makes later runs not strictly comparable with earlier ones.

Run an environment [#run-an-environment]

Select **Run** (▷) on the environment's row. **Evaluate an agent** opens with the environment picked; see [Test an agent](/docs/platform/lab/test-an-agent).

**Run** is unavailable when the environment's agent was deleted. Duplicate the environment for another agent instead.

Manage environments [#manage-environments]

Open &#x2A;*⋮** on a row:

| Action                 | What it does                                                                                                                                                             |
| ---------------------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------ |
| **Edit**               | Opens the environment in the same dialog. Selecting the row does the same.                                                                                               |
| **Duplicate**          | Creates a separate copy, for the agent you pick. A copy for a different agent keeps the checks and fixed inputs, not the simulated people.                               |
| **Publish to catalog** | Adds a copy to the **Catalog** tab, so others can start from it. Yours stays as it is.                                                                                   |
| **Export**             | Downloads it as a file, with or without its cases. Documents are never exported. **Import** on the Environments page loads one, from this or another Alomana deployment. |
| **Delete**             | Removes it from your pickers. Past runs keep their results. This cannot be undone.                                                                                       |

Use the catalog [#use-the-catalog]

The **Catalog** tab holds ready-made environments: some published by your workspace, some provided by Alomana.

1. Select an environment's card to see its checks and inputs.
2. Select **Add**, pick the agent, and give it a name. Your own editable copy goes to **My environments**; for an agent other than the one it was made for, it keeps the checks and fixed inputs, not the simulated people.

| Catalog       | Who can publish                                                               | Who can use it                                  |
| ------------- | ----------------------------------------------------------------------------- | ----------------------------------------------- |
| **Workspace** | Anyone who can create environments, in a workspace with more than one member. | Everyone in the workspace.                      |
| **Platform**  | Platform admins only.                                                         | Everyone on the deployment, in every workspace. |

Only the person who published an entry can edit or **Unpublish** it. Copies people already made stay theirs.