IronWarden

IronWarden is a high-performance, sovereign AI reverse proxy and privacy firewall, meticulously designed using bare-metal Rust, Axum, and Tokio. This project stands out for its ultra-low latency, offering exceptional capabilities for streaming PII redaction and real-time services without requiring any changes to existing applications.

Core Features

• Real-Time Streaming PII Redaction: IronWarden allows seamless integration with SDKs from OpenAI, Anthropic, and Ollama/vLLM. It provides real-time protection by automatically redacting personally identifiable information (PII) during data transactions.
• Sliding-Window SSE Token Rehydration: The system employs a unique asynchronous state machine that stitches together split tokens on-the-fly, ensuring minimal latency and high-throughput performance.
• Prompt Injection Defense: Built-in security measures protect against prompt injection vulnerabilities, reinforcing the integrity of data transactions.
• Cryptographic HMAC Audit Chaining: This feature ensures that all actions are logged securely, providing a tamper-evident audit trail essential for compliance with data privacy regulations.

Technical Superiority

IronWarden offers remarkable performance metrics as compared with similar technologies:

Architecture Overview

IronWarden’s architecture is built for scalability and efficiency. It directly connects client requests via HTTP/2 and supports streaming SSE to a core gateway that efficiently manages connections and processing:

       [ Client / Microservices / OpenAI & Anthropic SDKs ]
                   │
                   ▼ (HTTP/2, Streaming SSE, JSON-RPC)
       ┌─────────────────────────────────────────────────────────────┐
       │                   IronWarden Core Gateway                   │
       └─────────────────────────────────────────────────────────────┘

Compatibility

IronWarden seamlessly integrates with various SDKs, requiring minimal changes to existing implementations:

• Use the base URL to point to IronWarden’s gateway, easily adaptable for models such as OpenAI and Anthropic Claude.
• Automatic model routing is supported, facilitating communication with both local and cloud-based infrastructure.

Performance Benchmarks

Recently benchmarked results show:

• Streaming SSE Rehydration: Achieving a latency of just 0.04 ms per chunk, ensuring no perceived delay in data streaming.
• Added Gateway Overhead: Total overhead remains under 1.8 ms, contributing to a streamlined performance that is conducive to end-user experience.

Observability

With built-in observability, IronWarden provides:

• Prometheus Metrics for monitoring and analyzing request counts, PII redaction, and security events.
• A pre-configured Grafana Dashboard for visualizing performance and traffic metrics in real-time.

Conclusion

IronWarden is an advanced solution for organizations needing a robust, secure, and efficient AI proxy with privacy-centric features. It encapsulates the necessary tools to meet compliance mandates while providing flexibility and speed in real-world applications. Optimize data handling with IronWarden to ensure superior performance and security.