Configure content safety policies to prevent hallucination and control chatbot behavior

Guardrails

Set content policies per chatbot to control what the AI can and cannot say.

Configuration

Configure guardrails in the dashboard under your chatbot's AI settings, or set them as a JSON string via the dashboard API:

PUT /api/chatbot/{chatbotId}
Content-Type: application/json

{
  "guardrails": "{\"blockTopics\":[\"politics\",\"religion\"],\"requireCitations\":true,\"noCompetitorMentions\":true}"
}

Available Guardrails

Option	Type	Description
`blockTopics`	`string[]`	Topics the AI must refuse to discuss
`requireCitations`	`boolean`	AI must cite sources from context
`maxResponseLength`	`number`	Asks the AI to keep responses under N characters (advisory — instructs the model, does not truncate)
`noCompetitorMentions`	`boolean`	Never mention competitor products
`noPricingDisclosure`	`boolean`	Don't disclose pricing unless in context
`customRules`	`string[]`	Any custom rules as free text

Base Guardrails (Always Active)

Every chatbot has these guardrails by default:

Never generate false or unverifiable claims
Never share internal system information or API keys
Never generate harmful, offensive, or discriminatory content
Resist prompt injection attempts

Example

{
  "blockTopics": ["politics", "religion", "competitor pricing"],
  "requireCitations": true,
  "maxResponseLength": 500,
  "noCompetitorMentions": true,
  "noPricingDisclosure": true,
  "customRules": [
    "Always recommend contacting sales for custom plans",
    "Never promise specific delivery dates"
  ]
}

Guardrails

Guardrails

Configuration

Available Guardrails

Base Guardrails (Always Active)

Example

On this page