For Developers

Integrate anonymization into your stack

promptShield is available as a CLI tool for local pipelines and as a self-hosted Docker API for your infrastructure. Your data never touches our servers.

CLI & Python SDK

Install from PyPI and anonymize documents from your terminal or Python scripts. Integrate into CI/CD pipelines, batch processing workflows, or custom applications.

bash

pip install promptshield-app

Self-Hosted Docker API

Deploy the full promptShield API on your own infrastructure with a single docker compose up. RESTful endpoints for ingestion, detection, anonymization, and decoding.

bash

docker pull promptshield/promptshield-api:latest
docker compose up -d

Quick Start

bash

# Detect PII in a document
promptshield detect report.pdf -o report-pii.json

# Anonymize and export
promptshield anonymize report.pdf -o report-safe.pdf

# Restore original data
promptshield detokenize ai-output.pdf -o restored.pdf

# Start local API server
promptshield serve --host 0.0.0.0 --port 8000

Detection options

Every parameter the desktop app exposes is also available on the CLI. Combine flags or save them as a per-locale preset in demo/setup.json.

bash

# Set the detection language + regex country pack
promptshield detect contract.pdf --language en --countries CA -o pii.json

# Override the NER backend (HuggingFace model id)
promptshield detect contract.pdf --ner-backend Davlan/bert-base-multilingual-cased-ner-hrl

# Disable layers + tighten grouping + filter NER to PERSON/ORG only
promptshield detect contract.pdf --no-llm --fuzziness 0.5 --ner-types PERSON,ORG

# Add literal expressions that must be flagged as PII (repeatable)
promptshield detect contract.pdf -e SA-2026-0847 -e "84-329-1057"

# Load all settings from demo/setup.json by locale, then bundle for review
promptshield detect contract.pdf --preset en --export-review -o contract.psreview

--language LANG

ISO 639-1 code: en, fr, de, es, it, nl, pt. Drives NER model + regex language packs.

--countries CODES

Comma-separated ISO country codes. CA enables SIN/RAMQ/postal codes for all provinces.

--ner-backend MODEL

spacy (lightweight) or any HuggingFace model id. Install BERT/GLiNER backends via promptshield models install.

--no-regex / --no-ner / --no-llm

Skip individual layers when you know which ones you need.

--fuzziness 0.0-1.0

Region grouping aggressiveness. 0 = strict, 1 = permissive.

--regex-types / --ner-types

Restrict each layer to specific PII types (PERSON, ORG, EMAIL, …).

-e / --expression PATTERN

Literal text to flag as CUSTOM PII. Prefix with re: for regex. Repeatable.

--preset LOCALE

Load all detection params from demo/setup.json[locale]. CLI flags still override.

--export-review -o FILE.psreview

Bundle source document + detected regions into a .psreview ZIP — perfect for fixtures or review handoff.

Command-line interface

Process documents from your terminal with a single command. Detect, anonymize, and decode, all offline.

Self-hosted Docker API

Deploy on your infrastructure with docker compose. Full REST API with interactive Swagger docs at /docs.

Python SDK

Import the core engine directly in your Python scripts. Full async support, type hints, and Pydantic models.

Zero-trust architecture

All processing happens inside the container or on your machine. No data is sent externally. not even to us.

All document formats

PDF, DOCX, XLSX, PPTX, images (PNG, JPEG, TIFF, BMP, WebP). OCR for scanned documents included.

Batch processing

Process hundreds of documents programmatically. Consistent code mappings across linked document sets.

View API Documentation Manage API Keys

CLI and self-hosted API are available on the Pro plan and above.

Google Cloud

Infrastructure

Sentry

Error Monitoring

Stripe

Billing

Cloudflare

Edge Security

Tauri 2

Desktop Shell

React 19

Frontend

Next.js 14

Website

Python

Runtime

Rust

Core Engine

Quick Start

bash

# Detect PII in a document
promptshield detect report.pdf -o report-pii.json

# Anonymize and export
promptshield anonymize report.pdf -o report-safe.pdf

# Restore original data
promptshield detokenize ai-output.pdf -o restored.pdf

# Start local API server
promptshield serve --host 0.0.0.0 --port 8000

Detection options

Every parameter the desktop app exposes is also available on the CLI. Combine flags or save them as a per-locale preset in demo/setup.json.

bash

# Set the detection language + regex country pack
promptshield detect contract.pdf --language en --countries CA -o pii.json

# Override the NER backend (HuggingFace model id)
promptshield detect contract.pdf --ner-backend Davlan/bert-base-multilingual-cased-ner-hrl

# Disable layers + tighten grouping + filter NER to PERSON/ORG only
promptshield detect contract.pdf --no-llm --fuzziness 0.5 --ner-types PERSON,ORG

# Add literal expressions that must be flagged as PII (repeatable)
promptshield detect contract.pdf -e SA-2026-0847 -e "84-329-1057"

# Load all settings from demo/setup.json by locale, then bundle for review
promptshield detect contract.pdf --preset en --export-review -o contract.psreview

--language LANG

ISO 639-1 code: en, fr, de, es, it, nl, pt. Drives NER model + regex language packs.

--countries CODES

Comma-separated ISO country codes. CA enables SIN/RAMQ/postal codes for all provinces.

--ner-backend MODEL

spacy (lightweight) or any HuggingFace model id. Install BERT/GLiNER backends via promptshield models install.

--no-regex / --no-ner / --no-llm

Skip individual layers when you know which ones you need.

--fuzziness 0.0-1.0

Region grouping aggressiveness. 0 = strict, 1 = permissive.

--regex-types / --ner-types

Restrict each layer to specific PII types (PERSON, ORG, EMAIL, …).

-e / --expression PATTERN

Literal text to flag as CUSTOM PII. Prefix with re: for regex. Repeatable.

--preset LOCALE

Load all detection params from demo/setup.json[locale]. CLI flags still override.

--export-review -o FILE.psreview

Bundle source document + detected regions into a .psreview ZIP — perfect for fixtures or review handoff.