@radzor/guardrails

Validate LLM inputs and outputs against configurable safety rules including PII detection, content filtering, prompt injection detection, and length limits. Provides a rule engine for building custom validation pipelines.

AI & MLv0.1.0typescriptpythonServerguardrailssafetypiicontent-filtervalidationaillmsecurityby Radzor

Install

View source on GitHub →

$npx radzor@latest add guardrails

⚠ Constraints: No API keys required — all validation is local. PII detection uses regex patterns and is not 100% accurate. For production PII handling, consider combining with a dedicated PII service.

Inputs

Name	Type	Default	Description
enableBuiltinRules	boolean	true	Whether to load built-in rules (PII, prompt injection, profanity) on initialization.
maxInputLength	number	50000	Maximum allowed character length for input text.
maxOutputLength	number	100000	Maximum allowed character length for output text.

enableBuiltinRulesboolean

Whether to load built-in rules (PII, prompt injection, profanity) on initialization.

default: true

maxInputLengthnumber

Maximum allowed character length for input text.

default: 50000

maxOutputLength

{ "llm": { "constraints": "No API keys required — all validation is local. PII detection uses regex patterns and is not 100% accurate. For production PII handling, consider combining with a dedicated PII service.", "usageExamples": "llm/examples.md", "integrationPrompt": "llm/integration.md" }, "name": "@radzor/guardrails", "tags": [ "guardrails", "safety", "pii", "content-filter", "validation", "ai", "llm", "security" ], "events": [ { "name": "onViolation", "payload": { "rule": "string", "text": "string", "severity": "string", "direction": "string" }, "description": "Fired each time a rule violation is detected." } ], "inputs": [ { "name": "enableBuiltinRules", "type": "boolean", "default": true, "required": false, "description": "Whether to load built-in rules (PII, prompt injection, profanity) on initialization." }, { "name": "maxInputLength", "type": "number", "default": 50000, "required": false, "description": "Maximum allowed character length for input text." }, { "name": "maxOutputLength", "type": "number", "default": 100000, "required": false, "description": "Maximum allowed character length for output text." } ], "radzor": "1.0.0", "actions": [ { "name": "validateInput", "params": [ { "name": "text", "type": "string", "description": "The input text (user prompt) to validate." } ], "returns": "ValidationResult", "description": "Validate input text against all active rules." }, { "name": "validateOutput", "params": [ { "name": "text", "type": "string", "description": "The output text (LLM response) to validate." } ], "returns": "ValidationResult", "description": "Validate output text against all active rules." }, { "name": "addRule", "params": [ { "name": "rule", "type": "GuardrailRule", "description": "The rule definition to add." } ], "returns": "void", "description": "Add a custom validation rule." }, { "name": "listRules", "params": [], "returns": "RuleInfo[]", "description": "List all active validation rules." } ], "outputs": [ { "name": "validationResult", "type": "ValidationResult", "fields": [ { "name": "passed", "type": "boolean", "description": "Whether the text passed all validation rules." }, { "name": "violations", "type": "Array<{ rule: string; severity: string; message: string; matches?: string[] }>", "description": "List of rule violations found in the text." } ], "description": "Result of validating text against the active rules." } ], "runtime": "server", "version": "0.1.0", "category": "ai", "languages": [ "typescript", "python" ], "description": "Validate LLM inputs and outputs against configurable safety rules including PII detection, content filtering, prompt injection detection, and length limits. Provides a rule engine for building custom validation pipelines.", "dependencies": { "radzor": [], "packages": {} }, "composability": { "connectsTo": [ { "event": "onViolation", "description": "Alert on content policy violations", "compatibleWith": [ "@radzor/log-aggregator.action.warn.message", "@radzor/error-tracker.action.captureMessage.message", "@radzor/slack-bot.action.sendMessage.text" ] }, { "output": "validationResult", "mapField": "passed", "compatibleWith": [ "@radzor/event-tracker.action.track.eventName" ] } ] } }

@radzor/guardrails

Install

Inputs

Outputs

Actions

Events

Composability

radzor.manifest.json