Guardrails AI

Guardrails AI validates and corrects LLM outputs against defined rules, catching issues like hallucinations, PII leakage, off-topic responses, and format violations before they reach users. It wraps your LLM calls with a validation layer that automatically re-prompts when outputs fail checks.

Key use cases include:

• Hallucination detection — verifying that LLM outputs are grounded in provided context
• PII redaction — preventing sensitive data from appearing in responses
• Toxicity filtering — blocking harmful, offensive, or inappropriate content
• Format enforcement — ensuring outputs match required JSON schemas, regex patterns, or length constraints
• Topic adherence — keeping responses on-topic and within scope
• Custom business rules — validating outputs against domain-specific constraints

Guardrails AI is used by teams deploying LLMs in production where output quality and safety are critical — customer-facing applications, regulated industries, and enterprise environments where unexpected LLM behavior has real consequences.

The Guardrails Hub provides a library of community-contributed validators that can be composed together.

1. Install Guardrails:

   pip install guardrails-ai

2. Install validators from the Hub:

   guardrails hub install hub://guardrails/toxic_language
   guardrails hub install hub://guardrails/detect_pii

3. Create a guarded LLM call:

   from guardrails import Guard
   from guardrails.hub import ToxicLanguage, DetectPII
   
   guard = Guard().use_many(
       ToxicLanguage(on_fail='fix'),
       DetectPII(on_fail='fix')
   )
   result = guard(
       model='gpt-4o',
       messages=[{'role': 'user', 'content': 'Tell me about...'}]
   )

Pricing: Guardrails AI is free and open source. Some Hub validators use LLM calls (cost depends on provider). Guardrails Server (managed) offers team features with custom pricing.