Guardrails Node

The guardrails node scans a text value for policy violations using regex-based detectors. Violations can block execution, redact the offending content, or flag it while allowing the workflow to continue.

Configuration Fields

Field	Type	Default	Description
`guardrailType`	`pii` \| `moderation` \| `jailbreak` \| `all`	—	Shorthand to enable a check category
`piiEnabled`	boolean	`false`	Enable PII detection (overrides `guardrailType`)
`moderationEnabled`	boolean	`false`	Enable harmful content detection
`jailbreakEnabled`	boolean	`false`	Enable prompt injection / jailbreak detection
`actionOnViolation`	`block` \| `redact` \| `warn`	`block`	What to do when a violation is found
`inputField`	string	`lastOutput`	Variable name to inspect

guardrailType: 'all' enables all three check categories simultaneously.

Detectors

PII

Regex patterns for:

Email addresses
US phone numbers
Social Security Numbers (\b\d{3}-\d{2}-\d{4}\b)
Credit card numbers (16-digit)
IPv4 addresses

Jailbreak

Phrase-based detection for common prompt injection patterns including: "ignore previous instructions", "act as if", "jailbreak", "dan mode", and similar.

Moderation

Keyword patterns for violent or harmful content (hate speech, threats, etc.).

Actions

Action	Behavior
`block`	Throws an error — the node fails and the execution is marked `failed`
`redact`	Replaces detected patterns with `[TYPE_REDACTED]` and passes the cleaned text downstream
`warn`	Passes through unchanged but adds a `violations` array to the output

Output

When passed: true:

{ "passed": true, "violations": [] }

When violations are found with action: 'warn':

{
  "passed": false,
  "violations": [
    { "type": "pii", "detail": "Detected email: 1 instance(s)" }
  ]
}

When violations are found with action: 'redact':

{
  "passed": false,
  "violations": [...],
  "redactedText": "Contact [EMAIL_REDACTED] for support.",
  "originalText": "Contact alice@example.com for support."
}

These are regex-based detectors — they produce false positives (IPv4 patterns may match version numbers) and false negatives (novel jailbreak phrasing not in the list). Do not rely on this node as a sole security control for untrusted user input.

Example Config

{
  "piiEnabled": true,
  "jailbreakEnabled": true,
  "actionOnViolation": "redact",
  "inputField": "lastOutput"
}

Guardrails Node

On this page