Skip to content

Data Flow Guide

This guide documents how data flows within Enprivacy 3.0 and what external network connections the product makes. It is intended to support network egress validation and data residency verification.

Enprivacy 3.0 is designed with a data-residency-first approach:

  • All customer data is stored in your own infrastructure.
  • Enprivacy has no access to your data.
  • Outbound internet connectivity is minimised to a single runtime endpoint (auth.enprivacy.com) for user management. The only model downloads are a self-hosted LLM’s, from the Hugging Face Hub on first use, and pre-loading eliminates them — see External Connections.

Services communicate over the following default ports within the deployment’s private network. The Web and Job services hold all state in the database or object storage.

Source Destination Data Port Protocol
End user / admin Web Interface and API requests 8080 HTTP
Web Database Application state 5432 TCP
Job Database Application state 5432 TCP
Web LLM Detection / classification inference 8000 HTTP
Job LLM Detection / classification inference 8000 HTTP
Web Inference Adapter Document text for embedding, reranking and entity detection 8443 HTTPS
Job Inference Adapter Document text for embedding, reranking and entity detection 8443 HTTPS
Web OCR Text extraction 5001 HTTP
Job OCR Text extraction 5001 HTTP
Web / Job Object storage Document uploads and redacted output — Storage protocol

Interservice traffic uses plain HTTP/TCP by default, except to the Inference Adapter, which requires TLS and an API key, and is intended to run within a trusted private network. See the Network Security Guide.

The product makes the following outbound connections to external services:

Destination Port Protocol Purpose Data transmitted
auth.enprivacy.com 443 HTTPS User authentication and entitlement User identity (email, user ID, roles)
Hugging Face Hub (huggingface.co and *.hf.co CDNs) 443 HTTPS Model downloads on first use, for a self-hosted LLM only No customer data — model artefacts only; eliminated by pre-loading models

No customer data is transmitted to Enprivacy or any other external party.

The product does not phone home to check for updates or send telemetry. The Web, Job and Inference Adapter images carry their models and data. With a self-hosted LLM’s models pre-loaded, or a hosted LLM inside your network, all features work fully air-gapped except user authentication, which requires auth.enprivacy.com.

See Data Security & Privacy for details on where data is stored and encryption recommendations.

You can validate the product’s network egress by:

  1. Deploying with a network egress proxy or firewall logging enabled.
  2. Allowing only auth.enprivacy.com:443 in your egress allow-list.
  3. Blocking all other outbound traffic and verifying that no application features break (except authentication).