A self-healing LLM firewall with cryptographic data protection
Real-time data redaction/tokenization with format-preserving encryption (FF3-1) and reversible detokenization with policy gating.
Semantic violation detection, constitutional AI reflection, prompt rewriting, and tool/function call guarding.
Works with OpenAI, Anthropic, Mistral, Hugging Face, and Ollama with streaming support and mid-stream inspection.
OPA-style policy evaluation with versioning, multi-tenant management, and audit trails for compliance reporting.
BYOK/HSM integration, envelope encryption with AES-256-GCM, HKDF-SHA-512 key derivation, and tamper-evident audit logs.
Available as Docker images, Helm charts, language SDKs, or build from source with comprehensive documentation.
Protect your LLM applications with enterprise-grade security
View on GitHub Read Documentation