iron_cage_safety

PII detection and output validation for multi-agent systems.

Installation

[dependencies]
iron_cage_safety = { version = "0.1", features = ["full"] }

Features

• enabled (default): Core PII detection without audit logging
• full: All functionality including audit trail
• audit: SQLite-based audit logging

Quick Start

use iron_cage_safety::PiiDetector;

// Initialize PII detector with default patterns
let detector = PiiDetector::new()?;

// Check for PII in agent output
let text = "Contact me at john@example.com";
if detector.check(text) {
    // Redact PII before displaying
    let safe = detector.redact(text);
    println!("Safe output: {}", safe);
    // Output: "Safe output: Contact me at [EMAIL_REDACTED]"
}

Documentation

• API Reference
• Safety Patterns Guide

Responsibilities: Detects and redacts PII (emails, phones, SSNs, credit cards) from AI agent outputs with configurable safety policies and optional SQLite-based audit logging. Provides real-time validation preventing data breaches before agent responses reach users. Requires Rust 1.75+, all platforms supported, optional SQLite audit trail with audit feature.

In Scope:

• PII pattern detection (emails, phones, SSNs, credit cards)
• Output redaction and sanitization
• Safety audit trail (with audit feature)
• Configurable safety policies
• Real-time validation of agent outputs

Out of Scope:

• Cost tracking (see iron_cage_cost)
• Circuit breaker logic (see iron_cage_reliability)
• Agent lifecycle management (see iron_cage_cli)
• Configuration management (see iron_cage_types)

Source Files

Notes:

• Entries marked 'TBD' require manual documentation
• Entries marked '⚠️ ANTI-PATTERN' should be renamed to specific responsibilities

License

MIT