Introduction

Highvolt is an AI-powered Personally Identifiable Information (PII) detection system built by Key9, Inc. It scans documents — PDFs, images, Office files, archives, text files, and more — and uses a Large Language Model (LLM) to determine whether each file contains sensitive data.

What it does

Highvolt accepts files from multiple sources, routes them through an analysis pipeline, and records the results in OpenSearch. Each file is evaluated by an LLM that returns a structured JSON verdict:

{
  "has_sensitive_data": true,
  "confidence": "high",
  "reasoning": "Document contains social security numbers and full names.",
  "description": "HR record with PII",
  "status": "success"
}

Results are deduplicated by SHA256 hash — a file already analyzed is never re-submitted.

Components

Component	Description
highvolt-server	Central REST API server. Receives file submissions, queues work, invokes the LLM, and stores results in OpenSearch.
voltage	Endpoint agent. Scans local filesystems (Linux, macOS, Windows) for files matching configured MIME types and submits them to highvolt-server.
suricata	Network integration. Listens on a Redis pub/sub queue populated by Suricata IDS and submits network-captured files for analysis.
aws-s3	Batch cloud scanner. Traverses all existing buckets across an AWS Organization and submits matching objects to highvolt-server. Run once to establish a baseline.
aws-s3-lambda	Real-time cloud scanner. An AWS Lambda triggered by S3 object-creation events across the entire organization. Analyzes new uploads within seconds of arrival.

License

Highvolt is licensed under the GNU Affero General Public License v3.0 (AGPL-3.0).