Redactory

Redactory is a privacy‑first utility for detecting and removing personally identifiable information (PII) from text files. It transforms documentation into AI-ready web content without exposing PII or export-controlled data. All processing happens on your machine unless you set the AZURE_BLOB_SAS_URL environment variable, in which case scrubbed files are uploaded to Azure Blob Storage.

Features

• Detects common PII types (EMAIL, PHONE, SSN and ICD10 codes) using regular expressions
• Policy‑driven actions to MASK, REDACT or ALLOW detected entities
• Streaming transform for processing large files
• Command line interface with scrub, preview, ingest and policy validate
• Optional upload of scrubbed files to Azure Blob Storage
• Converts documentation into AI-ready web content without exposing PII or export-controlled data

Installation

npm install redactory

Quick Start

Create a policy file describing which entity types to detect and how they should be handled. An example policy is included in this repository:

version: 1
entityTypes:
  - EMAIL
  - PHONE
  - SSN
  - ICD10
actions:
  EMAIL: MASK
  PHONE: MASK
  SSN: REDACT
  ICD10: ALLOW
thresholds:
  default: 0.7
  SSN: 0.9
mask:
  char: "*"
  keepLast: 4
fallback: BLOCK

CLI usage

Build the project and run the CLI with npx:

npm run build
npx redactory scrub synthetic-data/sample.txt

Available commands:

• scrub <file> – redact a file in place
• preview <file> – show a diff of changes without modifying the file
• ingest <dir> – scrub all .txt, .html and .json files in a directory
• policy validate <file> – verify a policy file is valid

If the AZURE_BLOB_SAS_URL environment variable is set, scrubbed files will automatically be uploaded to Azure Blob Storage and the resulting blob URL will be printed.

Programmatic API

You can also use Redactory from your own Node.js code:

import { Scrubber, loadPolicy } from 'redactory';

const policy = loadPolicy('policy.yaml');
const scrubber = new Scrubber(policy);

const { result } = scrubber.scrub('Contact me at [email protected]');
console.log(result);

Using an ONNX NER model

Redactory can optionally load an ONNX model to detect entities using machine learning. A CPU build of onnxruntime-node is installed automatically during npm install. To use the GPU build set the environment variable ONNXRUNTIME_GPU=1 before installing. After obtaining a vocabulary mapping of tokens to IDs, provide the model path and vocabulary when constructing the Scrubber:

import fs from 'fs';
import { Scrubber, loadPolicy } from 'redactory';

const policy = loadPolicy('policy.yaml');
const vocab = JSON.parse(fs.readFileSync('vocab.json', 'utf8'));

const scrubber = new Scrubber(policy, {
  ner: { modelPath: 'ner-model.onnx', vocab }
});

const { result, entities } = scrubber.scrub('Alice met Bob.');
console.log(result, entities);

Streaming API

Redactory can scrub data from Node.js streams using the scrubStream helper:

import { Scrubber, loadPolicy, scrubStream } from 'redactory';
import { Readable } from 'stream';

const policy = loadPolicy('policy.yaml');
const scrubber = new Scrubber(policy);
const input = Readable.from(['Contact me at [email protected]']);

// pipe the redacted output elsewhere
const redacted = scrubStream(input, scrubber);
redacted.on('data', chunk => process.stdout.write(chunk));

This makes it easy to pipe the redacted output into other streams such as file writes or network uploads.

Synthetic Test Data

A sample file containing fabricated sensitive data lives in synthetic-data/sample.txt. Try scrubbing it with the CLI to see the output.

Development

Compile the TypeScript sources and run the test suite:

npm run build
npm test

License

This project is licensed under the Apache 2.0 License. See the LICENSE file for details.

Contributing

We welcome community contributions! Feel free to open an issue or submit a pull request on GitHub if you discover problems or have improvements to share.