Transparent OpenAI Client PII Redaction Wrapper for Node.js & TypeScript

The Solution: In-Memory 1-Line Client Wrapping

The PrivacyScrubber Developer Tools ecosystem provides a native client wrapper for the official OpenAI SDK (openai). By wrapping your OpenAI instance with wrapOpenAI, you establish an air-gapped security boundary directly in your application's volatile RAM.

import OpenAI from 'openai';
import { wrapOpenAI } from '@privacyscrubber/sdk'; // 1. Initialize OpenAI client with 1-line Zero-Trust wrapper
const openai = wrapOpenAI(new OpenAI({ apiKey: process.env.OPENAI_API_KEY }), { profile: 'Legal', // 25+ industry presets (Medical, Finance, HR, Security) detectSecrets: true, // Automatically strip AWS keys, JWTs, DB passwords licenseKey: process.env.PRIVACYSCRUBBER_KEY
}); // 2. Call chat.completions.create with standard OpenAI signatures
const response = await openai.chat.completions.create({ model: 'gpt-4o', messages: [ { role: 'user', content: 'Please summarize settlement agreement for John Doe (john@acme.com) with SSN 421-55-0147.' } ]
}); // 3. Prompt was sanitized in RAM before transmission; response is restored automatically!
console.log(response.choices[0].message.content);

How the Egress & Ingress Cycle Works

Unlike server proxies that require external network hops, wrapOpenAI operates within your existing Node.js event loop:

• Outbound Interception: When your code executes chat.completions.create(), the wrapper scans all messages in the payload against the active profile (e.g., Legal, Healthcare, Financial) and replaces sensitive substrings with deterministic tokens ([NAME_1], [EMAIL_1]).
• Volatile Session Isolation: The mappings between original values and tokens are stored in an ephemeral sessionMap allocated strictly within local heap memory. No disk writes or external database queries occur.
• Safe Model Ingestion: OpenAI's servers receive only sanitized tokens. The LLM processes the contextual structure of the prompt without learning private personal data or proprietary API credentials.
• Transparent Re-Hydration: When the completion is returned by OpenAI, the wrapper matches the tokens in the assistant's message against the volatile session map and reconstructs the original plaintext before returning the object to your calling function.

Enterprise Compliance & Pipeline Architecture

Integrating wrapOpenAI satisfies the strict data minimization requirements of SOC 2 Type II controls and ISO 27001 Annex A.8.11. For multi-stage architectures, this wrapper can be combined with the NPM PII Redaction SDK and Zero-Trust AI Data Pipelines to protect both direct LLM calls and background batch jobs.