Privacy Python Server

A standalone PII redaction API wrapping OpenAI's privacy-filter model. POST text, get back redacted text with typed numbered placeholders and detected span metadata.

Quick Start

uv sync
cp .env.example .env
DEV_MODE=true uv run python server.py

The first request triggers a model download (~1.5GB) from HuggingFace, then caches locally.

Example

curl -X POST http://localhost:8000/redact \
  -H "Content-Type: application/json" \
  -d '{"text": "My name is Alice Smith, email alice@smith.com, phone 555-0123."}'

{
  "redacted_text": "My name is <PRIVATE_PERSON_1>, email <PRIVATE_EMAIL_1>, phone <PRIVATE_PHONE_1>.",
  "spans": [
    { "label": "PRIVATE_PERSON", "id": 1, "text": " Alice Smith", "start": 10, "end": 22, "score": 0.999999 },
    { "label": "PRIVATE_EMAIL", "id": 1, "text": " alice@smith.com", "start": 31, "end": 48, "score": 0.999999 },
    { "label": "PRIVATE_PHONE", "id": 1, "text": " 555-0123", "start": 56, "end": 65, "score": 0.999998 }
  ],
  "summary": { "total_spans": 3, "by_label": { "PRIVATE_PERSON": 1, "PRIVATE_EMAIL": 1, "PRIVATE_PHONE": 1 } }
}

API

Method	Path	Auth	Description
GET	/health	No	Health check
POST	/redact	Optional	Redact PII from text

POST /redact accepts { "text": "..." } (1–128,000 chars, no extra fields). Returns RedactResponse with the redacted text, detected spans, and a summary.

Configuration

All config via env vars (see .env.example):

Variable	Default	Description
DEVICE	cpu	Inference device: cpu or cuda
HOST	0.0.0.0	Server bind address
PORT	8000	Server port
DEV_MODE	false	Debug logging
PRETTY_JSON	false	2-space indented JSON responses
AUTH_KEYS	(disabled)	Comma-separated Bearer tokens
RATE_LIMIT	60/minute	Per-IP rate limit
CORS_ORIGINS	(disabled)	Comma-separated allowed origins

Detected PII Types

account_number, private_address, private_email, private_person, private_phone, private_url, private_date, secret

Docker

docker build -t privacy-filter .
docker run -p 8000:8000 --env-file .env privacy-filter

The Dockerfile uses CPU-only PyTorch (~200MB vs ~2.5GB with CUDA) and pre-downloads the model during build.