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

# Guardrails

> Add safety rules that constrain what your agents accept and produce.

Guardrails are safety policies applied to a workflow — checking what goes in and what comes out. Use them to protect sensitive information and to keep workflows behaving the way you intend.

<Frame caption="The Guardrails page, with a grid of guardrail cards showing their types.">
  <img src="https://mintcdn.com/aiplanet/Jy4Vl1U9sSF_9LZH/images/guardrails.png?fit=max&auto=format&n=Jy4Vl1U9sSF_9LZH&q=85&s=756ca8c177f0470dc621bf513ab754e6" alt="Guardrails management page" width="1600" height="900" data-path="images/guardrails.png" />
</Frame>

## Guardrail types

The platform supports two guardrail types:

<CardGroup cols={2}>
  <Card title="PII detection" icon="user-shield">
    Detects personal information — such as email addresses, phone numbers, and other sensitive identifiers — and masks it so it isn't exposed.
  </Card>

  <Card title="Prompt injection protection" icon="shield-halved">
    Detects and blocks attempts to manipulate a workflow into ignoring its instructions.
  </Card>
</CardGroup>

## Creating a guardrail

Click **Create New** and configure:

* **Name** — a clear label, such as `Mask Personal Info`.
* **Type** — PII detection or prompt injection protection.
* **Settings** — type-specific options:
  * **PII detection** — choose which entity types to detect (emails, phone numbers, credit cards, IDs, and more) and the scope it applies to (model calls, tools, or both).
  * **Prompt injection protection** — set the confidence threshold for what counts as an injection attempt.

## Testing a guardrail

Before relying on a guardrail, test it. Open the guardrail's test panel and enter sample text — for instance, text containing an email address — and confirm the guardrail catches and handles it as expected.

<Tip>
  Test with realistic examples from your actual use case. A guardrail that's too strict can block legitimate input; one that's too loose can let things through.
</Tip>

## Applying guardrails

Guardrails are applied at the **workflow** level. In the workflow's settings panel, choose which guardrails the workflow uses — you can apply one PII detection guardrail and one prompt injection guardrail. See [The workflow builder](/building-workflows/builder-overview#settings).

## Managing guardrails

From each guardrail card you can **edit** its settings, **test** it, or **delete** it. If you delete a guardrail that's in use, the workflows using it will no longer have that protection.

## Next steps

<Card title="Running workflows" icon="play" href="/running-monitoring/running-workflows">
  See how runs work once your workflow is built and protected.
</Card>