Custom code guardrails

12 ready-made guardrails you can copy, customize and deploy on the LiteLLM AI Gateway to check requests and responses. See the custom code guardrails docs.

Security guardrails

• PII Detector: Detects and blocks personally identifiable information like SSNs, credit cards, and phone numbers
• PII Redactor: Redacts PII from inputs by replacing with [REDACTED] markers
• Prompt Injection Detector: Detects common prompt injection patterns and jailbreak attempts
• URL Validator: Validates and filters URLs in requests, blocking suspicious domains
• API Key Detector: Detects and blocks accidental API key exposure in prompts

Validation guardrails

• JSON Schema Validator: Validates that LLM responses conform to a specified JSON schema
• Word Count Limit: Enforces maximum word count on inputs or outputs
• Language Detector: Detects and optionally restricts content to specific languages

Content guardrails

• Code Detector: Detects code in responses and optionally blocks or flags certain languages

Formatting guardrails

• Response Formatter: Formats and cleans LLM responses (removes markdown, extra whitespace, etc.)

Moderation guardrails

• Toxic Content Filter: Filters toxic, offensive, or inappropriate content using keyword matching

Rate-limiting guardrails

• Rate Limit by User: Implements per-user rate limiting using request metadata