Privacy Filter: Detect PII in Text from Laravel

Privacy Filter is a Laravel package that detects private entities — names, emails, and other personally identifiable information — in arbitrary text. It wraps the privacy-filter.cpp command-line binary, a GGML inference engine for OpenAI's privacy-filter token-classification models, and exposes the results through a PHP API.

#Classifying Text into Entities

The main entry point is the entities() method, which runs the model over a string and returns Entity instances. Each entity carries its type, the matched text, the start and end byte offsets, and a confidence score:

use DirectoryTree\PrivacyFilter\Facades\PrivacyFilter;
 
$entities = PrivacyFilter::entities('Contact John Doe at jdoe@example.com.');
 
foreach ($entities as $entity) {
    echo $entity->type; // private_email
    echo $entity->text; // jdoe@example.com
}

Because the offsets are exact UTF-8 byte positions, you can map each detected entity back to its location in the original text rather than relying on string matching alone.

#Confidence Thresholds

Classifications default to a 0.5 confidence threshold. You can raise or lower it per call to trade recall for precision — a higher threshold returns fewer, more certain entities:

$entities = PrivacyFilter::entities(
    text: 'Contact John Doe at jdoe@example.com.',
    threshold: 0.75,
);

#Redacting Detected Text

The package detects entities; what you do with them is up to you. A redaction pass is a reduce over the returned entities:

$redacted = collect($entities)->reduce(function (string $text, $entity) {
    return str_replace($entity->text, '[redacted]', $text);
}, $text);

#Faking Classifications in Tests

Running the binary and loading the model in a test suite is slow, so Privacy Filter ships a fake() method that returns predetermined entities without invoking the binary:

use DirectoryTree\PrivacyFilter\Entity;
use DirectoryTree\PrivacyFilter\Facades\PrivacyFilter;
 
PrivacyFilter::fake([
    new Entity(type: 'private_email', start: 20, end: 36, score: 0.98),
]);

Your code under test then receives those entities as if the model had produced them.

#Installation and Setup

Install the package with Composer, then run the installer to fetch the compiled binary for your operating system and download the required GGUF model:

composer require directorytree/privacy-filter
php artisan privacy-filter:install

Prebuilt binaries for Linux, macOS (including ARM64), and Windows are distributed through PrivacyFilterBinaries. Pass --force to overwrite existing files. To customize the binary path, model path, process timeout, model URL, or release source, publish the configuration file:

php artisan vendor:publish --tag=privacy-filter-config

Every classification loads the model into memory, so the package recommends running classifications on dedicated queue workers in production to keep memory usage predictable.

You can find the source and documentation on GitHub.