Guardrails

InferiaLLM's Guardrail system ensures that both inputs (prompts) and outputs (responses) comply with your organization's safety and security policies.

Why Use Guardrails?

• Prevent Toxic Content: Filter hate speech, violence, and inappropriate content
• Protect Sensitive Data: Detect and redact PII automatically
• Security: Block prompt injection and jailbreak attempts
• Compliance: Ensure interactions meet regulatory standards

Supported Providers

1. LLM Guard (Local)

• Best for: Privacy-first, air-gapped deployments
• Description: Runs safety models locally. No data leaves your infrastructure.
• Features: Toxicity, secrets, code safety scanning
• Configuration: Default provider, no API key needed

2. Llama Guard (Groq)

• Best for: High-speed, chat-optimized applications
• Description: Meta's Llama Guard via Groq for ultra-low latency
• Configuration: API key managed in Dashboard > Settings > Providers

3. Lakera Guard (API)

• Best for: Security-critical applications
• Description: Specializes in prompt injection and jailbreak detection
• Configuration: API key managed in Dashboard > Settings > Providers

PII Detection & Redaction

The PII service scans every request for sensitive information:

• Detection: Emails, phone numbers, credit cards, SSNs, IP addresses
• Redaction: Replaces with placeholders (e.g., <EMAIL_ADDRESS>)
• Privacy: LLM providers never see your users' private data

Configuration

Environment Variables

Per-Scanner Flags

GUARDRAIL_ENABLE_TOXICITY=true
GUARDRAIL_ENABLE_PROMPT_INJECTION=true
GUARDRAIL_ENABLE_SECRETS=true
GUARDRAIL_ENABLE_PII=true

Dashboard Configuration

Guardrails can be configured per-deployment in the Dashboard:

1. Navigate to Deployments > [Your Deployment] > Guardrails
2. Enable/disable specific scanners
3. Configure provider-specific settings

API Provider Keys

Provider API keys (Groq, Lakera) are managed through the Dashboard:

1. Go to Settings > Providers
2. Select the provider category
3. Enter your API key (encrypted at rest)