open-guardrail-guards

30 built-in guards for open-guardrail — the open-source guardrail engine for LLM applications.

Install

npm install open-guardrail-guards open-guardrail-core

Or use the all-in-one package: npm install open-guardrail

Guards (30)

Security

• promptInjection — Detect jailbreak and prompt injection attempts
• regex — Custom pattern matching (ReDoS safe)
• keyword — Deny/allow keyword lists

Privacy

• pii — Detect and mask PII (email, phone, card, SSN)

Content

• toxicity — Profanity, hate speech, threats, harassment
• topicDeny / topicAllow — Topic control
• bias — Gender, racial, religious, age bias detection
• language — Restrict to allowed languages

Format

• wordCount — Min/max word/character limits
• schemaGuard — JSON schema output validation

AI Delegation

• llmJudge — Delegate judgment to external LLM
• hallucination — Fact-check via LLM
• relevance / groundedness — RAG verification via LLM

Operational

• costGuard — Token usage and cost limits
• rateLimit — Per-key request rate limiting
• dataLeakage — System prompt leak detection
• sentiment — Emotional tone control

Agent Safety

• toolCallValidator — Tool call argument validation (email, uuid, SQL injection)
• codeSafety — Dangerous code pattern detection (eval, exec, rm -rf)

Advanced

• copyright — Copyright notices, trademarks, verbatim reproduction
• watermarkDetect — AI-generated text detection
• multiTurnContext — Multi-turn manipulation detection

Korea / ISMS-P / PIPA

• piiKr — Korean PII (주민등록번호, 여권, 면허, 사업자등록번호)
• profanityKr — Korean profanity (초성, 변형 포함)
• residentId — Resident ID checksum + masking
• creditInfo — Financial info protection
• ismsP / pipa — Korean compliance presets

Usage

import { promptInjection, pii, keyword } from 'open-guardrail-guards';
import { pipe } from 'open-guardrail-core';

const result = await pipe(
  promptInjection({ action: 'block' }),
  pii({ entities: ['email'], action: 'mask' }),
  keyword({ denied: ['hack'], action: 'block' }),
).run('user input');

License

MIT