YAML Config

Define your guardrails in YAML — no code changes needed.

Basic Config

Create guardrail.yaml:

version: "1"
pipelines:
  input:
    mode: fail-fast
    guards:
      - type: prompt-injection
        action: block
      - type: pii
        action: mask
        config:
          entities: [email, phone]
  output:
    mode: run-all
    guards:
      - type: toxicity
        action: warn
      - type: hallucination
        action: block

Loading Config

import { OpenGuardrail } from 'open-guardrail';

const engine = await OpenGuardrail.fromConfig('./guardrail.yaml');
const result = await engine.run(text);

CLI Validation

npx open-guardrail-cli init       # generate guardrail.yaml
npx open-guardrail-cli validate   # validate config

Using Presets

version: "1"
preset: korean    # loads ISMS-P + PIPA + resident ID guards

Available presets: default, strict, korean, security, content, ai-basic-act-kr, eu-ai-act

Guard Config Options

Each guard supports a config block for guard-specific options:

guards:
  - type: keyword
    action: block
    config:
      denied: [hack, exploit, jailbreak]
      caseSensitive: false
  - type: word-count
    action: warn
    config:
      max: 500
      unit: words