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

# Configure guardrails

> Switch guardrails on, set what they check, save a draft, and apply it as the live version or complete an assessment.

Open an AI system, go to its **Guardrails** tab and open the editor, or choose **Configure guardrails** from an assessment. The page is titled **Guardrails for** followed by the AI system's name.

## The editor

The list on the left shows the eight guardrails and whether each is on. It marks a guardrail that **Needs your setup** (on, with nothing chosen yet) or that has **Unsaved** changes. Choose one to open it beside the list. On a phone the list comes first.

Each guardrail has two tabs:

* **Configuration**: the switch, **Applies to**, **When it fires**, and its own settings. While it is off, its settings are greyed out.
* **Risks and controls**: the risks of the latest assessment that this guardrail helps cover, with their controls.

The page address names the guardrail you are looking at, so a reload or a shared link opens the same one.

<Steps>
  <Step title="Switch a guardrail on">
    Choose the guardrail and turn it on.
  </Step>

  <Step title="Set it up">
    Choose what it applies to and what it does when it fires, then fill in what it looks for. See [Guardrail types](/guardrails/types).
  </Step>

  <Step title="Save the draft">
    Choose **Save draft**. Every changed guardrail is saved together. A badge says **Unsaved changes** until you do, then **Saved, not applied**.
  </Step>

  <Step title="Test it">
    Choose **Test guardrails** to check messages against the saved draft. See [Test guardrails](/guardrails/test).
  </Step>

  <Step title="Apply it">
    Choose **Apply** and confirm: "Your saved draft becomes the new live version." A message says **Version N is live.**
  </Step>
</Steps>

**Apply** is available when:

* the draft is saved,
* at least one guardrail is on, and
* every guardrail that is on has what it needs.

A guardrail that is on and empty is refused with a message beside it, such as "Choose at least one of topics, or switch this guardrail off." You can save up to 120 times and apply up to 30 times an hour.

## While an assessment is open

When the AI system has an assessment **In review**, the editor works for that assessment:

* **Apply** and **Versions** are hidden. **Complete Assessment** takes their place.
* **Back to review assessment** returns to the results, as often as you like.
* **Complete Assessment** asks you to confirm: "Your saved guardrail settings go live and the assessment closes. If one cannot be applied, the assessment stays open and names it." A message says **The assessment is complete.**

## If applying fails

If Altrum cannot publish the version, the previous live version stays live and your draft is kept. You see "The guardrail service is busy or cannot be reached. Your settings are saved: try again in a few minutes."

If the version is published but the AI Gateway cannot be told at that moment, the version is live in Altrum and the answer says the AI Gateway has not been told yet. Altrum keeps trying until it has, and you do not need to apply again.


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.