Configuring Guardrails and Deploying Your Agent

Navigate to the agent you want to configure and open its detail page.

Navigate to Agents in the navigation menu. Click on the agent you created. This opens the agent detail page where you can configure guardrails, test behavior, and get the embed code.

Guardrails define what your agent should and should not discuss. Set up topic and word restrictions to keep conversations focused.

Add topic restrictions: Define topics the agent should not discuss. For example, block discussions about competitor products, internal pricing, or unrelated subjects.

Add word restrictions: Block specific words or phrases from appearing in responses. Useful for preventing profanity, confidential terms, or brand-sensitive language.

Set match types: Choose how restrictions are matched. Options include exact match, contains, or regex patterns for advanced matching.

Set actions: Choose what happens when a restriction is triggered.

Before deploying, verify that your guardrails behave as expected by testing with real questions.

Ask the agent questions that should trigger your restrictions. Try variations and edge cases. Verify the agent responds appropriately: blocking, warning, or redirecting depending on your configuration. Adjust sensitivity if restrictions are too strict (blocking legitimate questions) or too loose (allowing off-topic responses).

Once your guardrails are working, grab the embed code to add the agent to your website.

Click Widget or Embed in the agent settings panel. Copy the embed code. You can choose between an iframe snippet or a JavaScript snippet depending on your website's requirements. Customize the widget appearance: colors, position on screen, size, and welcome message.

Place the embed code on your website to make the agent available to visitors.

Paste the embed code into your website's HTML, typically before the closing </body> tag. The widget appears as a chat bubble in the corner of your page. Visitors can click it to start a conversation with your agent. Test on both desktop and mobile devices to verify the widget loads correctly and conversations work as expected.

Once your agent is live, monitor how it handles real conversations and catch issues early.

Navigate to Agent Operations to see conversations in real time. Check for escalations, which are questions the agent could not answer confidently. These highlight gaps in your knowledge base. Review response quality in the first few days. Look for patterns in what users ask and how the agent responds.