> ## Documentation Index
> Fetch the complete documentation index at: https://aibco.mintlify.site/llms.txt
> Use this file to discover all available pages before exploring further.

# Set Guardrails for Your AI Chat Behavior

> Guardrails let you add custom instructions that shape how your AI responds, layered on top of the platform's built-in safety rules.

Guardrails are custom instructions you write to control the tone, style, and boundaries of your AI chat. They help the AI stay on-brand and avoid answers that do not match how you want to represent your business.

## How guardrails work

Guardrails layer on top of your knowledge base. When a visitor asks a question, the AI considers the knowledge you have added, then applies your guardrails to shape the final response. For example, a guardrail can instruct the AI to always recommend a consultation before quoting a price.

## Add a guardrail

<Steps>
  <Step title="Open the Guardrails tab">
    In your dashboard, go to **Chat → Guardrails**.
  </Step>

  <Step title="Write your instruction">
    Type any extra instructions (e.g. *Always recommend scheduling a consultation before quoting a price*).

    <img src="https://mintcdn.com/aibco/Itg-OrkMwe5i239l/images/Screenshot-2026-08-29-155617.png?fit=max&auto=format&n=Itg-OrkMwe5i239l&q=85&s=b64d7e470c371ed5558c08933b6b6ba6" alt="Screenshot 2026 08 29 155617" width="747" height="638" data-path="images/Screenshot-2026-08-29-155617.png" />
  </Step>

  <Step title="Save">
    Click **Save behavior**. Changes take effect immediately for the next visitor chat session.
  </Step>
</Steps>

## Example guardrail

> Always recommend scheduling a consultation before quoting a price.

With this guardrail in place, if a visitor asks "How much does this cost?", the AI will explain that pricing depends on the specific project and suggest booking a consultation rather than giving a fixed number.

<Note>
  Guardrails work alongside the platform's core safety rules, not instead of them. You cannot remove built-in safety behaviors here.
</Note>

<Warning>
  The `[render]` and `[contact]` placeholders work inside chat messages you send to visitors (like a greeting or suggested question). Placing them in Guardrails instructions has no effect. They are stripped before the AI sees them.
</Warning>

## Next steps

* [Add knowledge sources](/chat/knowledge)
* [Configure the greeting message](/chat/greeting)