> For the complete documentation index, see [llms.txt](https://docs.xenioo.com/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.xenioo.com/ai/large-language-models/guardrails.md).

# Guardrails

The **Guardrails** page lists every check made by the AI Guardrail actions of your bot. Each time an AI Guardrail action runs, its result is stored here, whether the text passed, was blocked or the check failed to run.

Use the blocked results to verify that your guidelines are working as expected and to refine your bot instructions. Use the passed results to spot anything that slipped through and should have been stopped.

#### The Results List

Each row of the list contains the following information:

| Column       | Description                                                                   |
| ------------ | ----------------------------------------------------------------------------- |
| Date         | The date and time of the check                                                |
| Outcome      | The result of the check: *Passed*, *Blocked* or *Error*                       |
| Checked text | The text that was verified, usually the AI reply                              |
| Reason       | Why the text was blocked or, for errors, why the check could not be completed |
| Interaction  | The behaviour and interaction where the AI Guardrail action was executed      |
| Model        | The AI provider and model used to run the check                               |
| Time         | The number of seconds the check took                                          |

Checks executed from the preview chat are marked with a *Preview* label next to the outcome.

Expand a row to see the full details of the check:

* The user message that originated the AI reply
* The complete checked text
* The list of broken guidelines, each one with the part of the text that broke it
* The error message, if the check failed
* The guidelines that were active when the check ran
* The conversation id and the number of tokens consumed by the check

#### Filtering And Managing Results

Use the outcome selector at the top of the page to show only *Blocked*, *Passed* or *Error* results. Click **Refresh View** to load the latest results.

Results can be deleted one by one from the row menu, or in bulk by selecting multiple rows and clicking **Delete Selected**.

{% hint style="info" %}
Results are kept for 30 days and then automatically deleted.
{% endhint %}

#### Remarks

The *Error* outcome means the check itself could not be completed, for example because of a timeout or a provider error. It does not mean the text was wrong. Depending on the *Treat a hard fail as a block* setting of the action, an error lets the conversation move on or is handled as a block.

A high number of errors usually means the selected model is too slow for the bot API timeout or the API key is not valid anymore. A high number of blocks for the same guideline usually means the bot instructions should be updated so that the AI stops producing that kind of reply in the first place.

{% hint style="warning" %}
Checked texts and user messages are stored as they were sent. Keep this in mind if your conversations can contain personal data.
{% endhint %}