ph-eye

ph-eye is a lightweight HTTP service that hosts AI/NLP models for finding PII and PHI in text. It exposes a simple REST API and is designed for use with Phileas and Philter, though it can also be used standalone.

• Docker images: https://hub.docker.com/repository/docker/philterd/ph-eye
• Documentation: https://philterd.github.io/ph-eye

Available models

Each Docker image bundles a single model, selected at build time. Images are air-gapped — the model is downloaded into the image during docker build and no network access is required at runtime.

PHEYE_MODEL	Language	Entities	Underlying model
pii_en_small (default)	English	Person names (label name)	philterd/ph-eye-pii-en-small
pii_en_xsmall	English	Person names (label name)	philterd/ph-eye-pii-en-xsmall
pii_en_medium	English	Person names (label name)	philterd/ph-eye-pii-en-medium
pii_en_large	English	Person names (label name)	philterd/ph-eye-pii-en-large
pii_base (deprecated)	English	Person names (label Person)	philterd/ph-eye-pii-base
hospitals	English	Hospital, room number	knowledgator/gliner-pii-base-v1.0
medical_conditions	English	Disease/disorder	blaze999/Medical-NER
french_persons	French	Person	EmergentMethods/gliner_medium_news-v2.1
french_medical	French	Disease (Maladie)	almanach/camembert-bio-gliner-v0.1

pii_en_small is the default when PHEYE_MODEL is not set. The English name models come in four sizes (pii_en_xsmall, pii_en_small, pii_en_medium, pii_en_large) that trade speed for accuracy; each has its own recommended default threshold (0.50, 0.90, 0.70, and 0.95 respectively). pii_base is deprecated and remains available only when explicitly selected.

Building

Pass PHEYE_MODEL as a build argument to select the model. The model is downloaded from Hugging Face and cached inside the image at build time.

docker build --build-arg PHEYE_MODEL=pii_en_small        -t pheye:1.3.0-pii-en-small .
docker build --build-arg PHEYE_MODEL=pii_en_xsmall       -t pheye:1.3.0-pii-en-xsmall .
docker build --build-arg PHEYE_MODEL=pii_en_medium       -t pheye:1.3.0-pii-en-medium .
docker build --build-arg PHEYE_MODEL=pii_en_large        -t pheye:1.3.0-pii-en-large .
docker build --build-arg PHEYE_MODEL=pii_base            -t pheye:1.3.0-pii-base .
docker build --build-arg PHEYE_MODEL=hospitals           -t pheye:1.3.0-hospitals .
docker build --build-arg PHEYE_MODEL=medical_conditions  -t pheye:1.3.0-medical-conditions .
docker build --build-arg PHEYE_MODEL=french_persons      -t pheye:1.3.0-french-persons .
docker build --build-arg PHEYE_MODEL=french_medical      -t pheye:1.3.0-french-medical .

Running

docker run -p 5000:5000 pheye:1.3.0-pii-en-small

To run all models together using Docker Compose:

docker compose up                        # CPU
docker compose -f docker-compose.gpu.yaml up   # GPU

Check that the service is up:

curl http://localhost:5000/status

API

POST /find

Find entities in text. labels and threshold are optional — defaults are model-specific.

Request

{
  "text": "Please forward the invoice to Toni Levine and copy Maria Gonzalez.",
  "labels": [
    "name"
  ],
  "threshold": 0.9
}

The default pii_en_small model uses the label name and a threshold of 0.9. Omit labels and threshold to use the model defaults.

Response

[
  {
    "label": "name",
    "text": "Toni Levine",
    "score": 0.9712,
    "start": 30,
    "end": 41
  },
  {
    "label": "name",
    "text": "Maria Gonzalez",
    "score": 0.9685,
    "start": 51,
    "end": 65
  }
]

GET /status

Returns healthy when the service is running and the model is loaded.

License

ph-eye is licensed under the Apache License, version 2.0. The model bundled in each image may be licensed separately - refer to the model's Hugging Face page for details.

Copyright 2024-2026 Philterd, LLC.