> ## Documentation Index
> Fetch the complete documentation index at: https://docs.julian.11x.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Guardrails and security

> Keep the agent off topics you don't want discussed, and control which sites can load your widget.

## Guardrails

Every chat agent has built-in safety guardrails. Add your own in **Widget > Data & Compliance > Guardrails**:

| Setting | What it does |
| - | - |
| **Restricted Topics** | Topics the agent won't discuss, each with an optional reason for your team's reference. The agent politely steers away from them |
| **Blocked Phrases** | Words or phrases the agent must not use |
| **Redirect Message** | What the agent says when it steers away from a restricted topic |
| **Custom Instructions** | Under **Advanced Settings**: extra moderation rules in your own words |

Typical restricted topics are competitors' pricing, legal or medical advice, and anything the agent shouldn't promise,
such as discounts or delivery dates.

## Where the widget can load

List your domains under **Allowed Origins** in **Widget > Data & Compliance > Page Rules**, and the widget loads only on
those sites. You can also limit a domain to some of its pages. See [Install the widget](https://docs.julian.11x.ai/chat/install.md#add-the-embed-code).

## App ID

The App ID in the embed code identifies your widget. If it's exposed somewhere it shouldn't be, regenerate it from the
**Install** tab: the old ID stops working within five minutes, so update the embed code on your site at the same time.


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