# Configuring guardrails

> Bring your own subscription to a supported guardrail provider and make its validators available as agent guardrails in the AI Trust Layer.

The **Guardrail configurations** tab lets you bring your own subscription to a supported safety vendor and offer its validators as guardrails for the agents in your tenant. You keep the vendor account and the subscription, and UiPath calls your provider whenever a guardrail based on your configuration runs.

During preview, you can bring a subscription only to the vendors listed in [Supported guardrail providers and validators](#supported-guardrail-providers-and-validators). Any other vendor is out of scope, because each supported one relies on a UiPath-provided connector template.

:::note
Access to guardrail configurations depends on the cloud platform you use. For details, refer to [Test Cloud feature availability](test-cloud-feature-availability.md#admin).
:::

Your configurations do not replace the UiPath-managed guardrails. Both sets are offered together, and each guardrail shows the provider that evaluates it. Several configurations can reuse the same Integration Service connection, so one vendor account can serve every validator it supports.

:::note
The **Fall back to UiPath-managed guardrail** option determines what happens when your provider returns an error or becomes unreachable. With the option enabled, the evaluation runs against the UiPath-managed guardrail instead. With it disabled, the evaluation fails and the agent step is blocked.
:::

## Prerequisites

* You are an organization administrator.
* An Integration Service connection to your safety vendor exists in a folder you can access. If you do not have one yet, you can create both the connector and the connection during the procedure.

## Supported guardrail providers and validators

The connectors in the following table are the only guardrail providers supported during preview. The validators you can enforce depend on the connector you select, and each connector corresponds to one safety vendor service.

 
  
    Connector 
    Validators 
  
 
 
  
    Azure AI Language Guardrails 
    PII Detection 
  
  
    Azure Content Safety Guardrails 
    Harmful Content DetectionUser Prompt Attack DetectionProtected Material Detection for TextProtected Material Detection for Code 
  
  
    Noma Guardrails 
    Prompt Injection Detection 
  
 

## Creating a guardrail configuration

1. Navigate to **Admin > AI Trust Layer > Guardrail configurations**.
2. Select the **Tenant** the configuration applies to.
3. Select **Add configuration** to add a new configuration.
4. Choose the **Connections Folder** that stores your guardrail connection.

   The folder you choose determines where the guardrail connection is saved, the same as for a BYO LLM configuration. It does not restrict who can use the guardrail. Every user in the tenant can see and apply the configuration.
5. From the **Connector** list, choose a guardrail connector you already created, or create one from a UiPath-provided template without leaving the configuration flow. The templates cover the supported vendors only:
   1. Select **Create custom connector**.
   2. Select a template from the available options: **Noma Guardrails**, **Azure Content Safety Guardrails**, or **Azure AI Language Guardrails**.
   3. Select **Create connector**.
   4. In Connector Builder, configure the authentication method for your vendor account, then publish the connector. For details, refer to [Publishing a connector](https://docs.uipath.com/integration-service/automation-cloud/latest/user-guide/connector-builder-publishing).
   5. Return to the AI Trust Layer tab, then select your new connector from the **Connector** list.
6. From the **Connection** list, choose the Integration Service connection for your vendor account. If no connection exists yet, select **Add connection** and create one against your connector.
   :::note
   Your provider's credentials stay in Integration Service. A guardrail configuration stores a reference to the connection, never the secret itself.
   :::
7. From the **Validator** list, choose the validation capability to enforce, such as **PII Detection** or **Harmful Content Detection**. The list shows only the validators that the selected connector supports.
8. Enter a **Configuration name**. This is the name the guardrail appears under when it is applied.
9. Set **Fall back to UiPath-managed guardrail** according to how you want provider failures handled.
10. Select **Test**. UiPath sends a test prompt to your provider through the connection to confirm that the validator responds.
11. Select **Save**.

The configuration appears in the list with its connector, validator, and status.

## Updating or deleting a configuration

Select the tenant whose configurations you want to manage. The list shows every configuration for that tenant, and you can update or delete any of them.

## Applying your guardrails

A saved guardrail configuration is applied the same two ways as a UiPath-managed guardrail:

* **Per agent** — Agent developers add it from **Add guardrails** in Agent Builder or Maestro Flow. In the **Choose guardrail** panel, your configurations are grouped under the folder they were created in, above the UiPath guardrails, each marked **BYO** and listing its connector as the provider. Developers then set the categories to detect, the thresholds, the scopes, and the action to take, as with any other guardrail. For details, refer to [Out-of-the-box guardrails](https://docs.uipath.com/agents/automation-cloud/latest/user-guide/out-of-the-box-guardrails).
* **Organization-wide** — Administrators enable it in the **Bring your own guardrails** section of the **Centralized guardrails** tab in the AI Trust Layer policy. They select the tenant, then enforce each guardrail for all LLM calls or for individual products. For details, refer to [Bring your own guardrails](https://docs.uipath.com/automation-ops/automation-cloud/latest/user-guide/settings-for-ai-trust-layer-policies#bring-your-own-guardrails).

Agent traces record which configuration ran, the validator type, the connector and connection used, and whether the evaluation fell back to the UiPath-managed guardrail.