redact-ai-stream

Bi-directional PII Redaction for AI Streams

redact-ai-stream is a lightweight, specialized Node.js library designed to secure your AI applications. It acts as a middleware layer, automatically redacting Personally Identifiable Information (PII) from data streams before they exit your secure boundary (e.g., to OpenAI, Anthropic), and transparently restoring that data in the incoming response stream.

Why use this?

When building RAG requests or chat interfaces, you often need to send user context to an LLM. However, sending raw email addresses, phone numbers, or credit card details violates privacy compliance (GDPR, CCPA) and security best practices. redact-ai-stream solves this by tokenizing sensitive data on the fly.

Features

• Stream-based Redaction: Integrates natively with Node.js Transform streams.
• Bi-directional: Redacts outgoing data, restores incoming data.
• Session-based Security: Tokens are unique per session (<EMAIL_UUID>).
• Zero-Persistence: Original PII is held in memory only for the duration of the stream; never stored on disk.
• TypeScript Support: Written in TypeScript with full type definitions included.

Installation

npm install redact-ai-stream

Usage

TypeScript / ES Modules

import RedactionSession from 'redact-ai-stream';
import { Readable } from 'stream';

// 1. Create a session
const session = new RedactionSession();

// 2. Mock input stream (e.g., from an API request)
const userInput = Readable.from(["My email is [email protected]."]);

// 3. Pipe through redaction
const redactedStream = userInput.pipe(session.redact());

redactedStream.on('data', (chunk) => {
    console.log('To LLM:', chunk.toString());
    // Output: "To LLM: My email is <EMAIL_1234-5678...>"
});

// ... Send to AI ...

// 4. Restore AI response
const aiResponse = Readable.from(["Sending confirmation to <EMAIL_1234-5678...>."]);
const finalStream = aiResponse.pipe(session.restore());

finalStream.on('data', (chunk) => {
    console.log('To User:', chunk.toString());
    // Output: "To User: Sending confirmation to [email protected]."
});

CommonJS

const RedactionSession = require('redact-ai-stream');
// Usage is identical to above

Supported Redactions

| Type | Pattern Example | Token Format | | :--- | :--- | :--- | | Email | [email protected] | <EMAIL_UUID> | | Credit Card | 4532 1234 5678 9012 | <CC_UUID> | | Phone | +1-555-0123 | <PHONE_UUID> |

License

MIT © Godfrey Lebo