Language moderation

The Language moderation guardrail uses a dedicated LLM provider with a hardcoded prompt to detect content that falls into standard moderation categories. You select which categories to enforce from a predefined list.

This is different from the Moderation Model guardrail which uses a purpose-built moderation API. The Language moderation guardrail uses a regular LLM provider, making it more flexible (customizable categories) but slower and more expensive.

It can be applied before sending the prompt to the LLM and after to validate the LLM response.

How it works

The guardrail sends messages to a validation LLM with a system prompt that lists the selected moderation categories. The LLM evaluates the content and responds with true (content is clean) or false (content violates one or more categories).

Configuration

"guardrails": [
  {
    "enabled": true,
    "before": true,
    "after": true,
    "id": "moderation",
    "config": {
      "provider": "provider_xxxxxxxxx",
      "moderation_items": [
        "hate",
        "hate/threatening",
        "harassment",
        "harassment/threatening",
        "self-harm",
        "self-harm/intent",
        "self-harm/instructions",
        "sexual",
        "sexual/minors",
        "violence",
        "violence/graphic"
      ]
    }
  }
]

Field explanations

• enabled: true — The guardrail is active
• before: true — The guardrail applies to user input before sending to the LLM
• after: true — The guardrail applies to the LLM response

Config section

Parameter	Type	Required	Default	Description
provider	string	Yes	—	Reference ID of the LLM provider used for moderation evaluation. Must be different from the main provider.
moderation_items	array of strings	No	[]	List of moderation categories to enforce.
err_msg	string	No	"This message has been blocked by the 'language-moderation' guardrail !"	Custom error message returned when a message is blocked.

Available moderation categories

Category	Description
hate	Content that expresses hate toward a group
hate/threatening	Hateful content that includes threats of violence
harassment	Content that harasses an individual or group
harassment/threatening	Harassment that includes threats
self-harm	Content that promotes self-harm
self-harm/intent	Content expressing intent to self-harm
self-harm/instructions	Instructions for self-harm
sexual	Sexually explicit content
sexual/minors	Sexual content involving minors
violence	Content depicting violence
violence/graphic	Graphic depictions of violence
profanity	Content containing profane language