Prompt validator

The prompt validator plugin validates the content of consumer messages against regex-based allow/deny patterns before they reach the LLM provider. This is a lightweight alternative to LLM-based guardrails for pattern-based content filtering.

How it works

1. The plugin stores its allow/deny patterns in the request attributes
2. The LLM proxy reads these patterns and validates every message content against them
3. If validation fails, the request is rejected

Validation logic

• Allow patterns: if any are defined, at least one must match the message content
• Deny patterns: if any match the message content, the message is denied
• A message passes validation only when it is not denied AND allowed

Plugin configuration

Add the Cloud APIM - LLM Proxy - prompt validator plugin to your route, before the LLM proxy plugin:

{
  "enabled": true,
  "plugin": "cp:otoroshi_plugins.com.cloud.apim.otoroshi.extensions.aigateway.plugins.AiPromptValidator",
  "config": {
    "allow": [],
    "deny": []
  }
}

Parameter	Type	Default	Description
allow	array	[]	Regex patterns — if non-empty, at least one must match for the message to pass
deny	array	[]	Regex patterns — if any matches, the message is denied

Stacking validators

Multiple prompt validator plugins can be stacked on the same route. All validators must pass for the request to be allowed — their allow/deny patterns accumulate.

Examples

Block requests containing code

{
  "deny": ["```.*```", "<script.*>", "eval\\(", "exec\\("]
}

Only allow questions about a specific topic

{
  "allow": ["(?i).*(weather|forecast|temperature|climate).*"]
}

Combine allow and deny

{
  "allow": ["(?i).*(product|pricing|support).*"],
  "deny": ["(?i).*(competitor|internal|confidential).*"]
}

This allows only messages about products, pricing, or support, while blocking any mention of competitors or confidential information.