Precise PII Detection at Scale

Runs air-gapped on your infrastructure with better accuracy than the hyperscaler alternatives.

Core Operations

PII Eraser offers four modes of operation to detect and anonymize PII, PCI and other sensitive data.

Detect (NER)

Identify selected entities, returning character offsets and confidence scores. Supports overlapping entities and nested PII.

Redact

Detect and replace entities with entity-specific placeholders like <NAME>. Ideal for preparing datasets for ML training and sanitizing LLM prompts.

Mask

Obfuscate characters while preserving format length (e.g., John -> ####). Useful for UX logs and where partial visibility is required.

Hash

Replace entities with consistent SHA-256 or SHA-512 hashes. Enables analytics on unique users without revealing identity.

Python Input

import requests

# Detect sensitive data instances
response = requests.post(
    "<base url>/text/detect",
    json={
        "text": ["Hallo Matthias"],
    }
)

import requests

# Replace sensitive data with entity types
response = requests.post(
    "<base url>/text/transform",
    json={
        "text": ["Contact [email protected]"],
        "operator": "redact", 
    }
)

import requests

# Mask characters with asterisks
response = requests.post(
    "<base url>/text/transform",
    json={
        "text": ["My ID is 452-992-112"],
        "operator": "mask", 
    }
)

import requests

# Cryptographic hashing for analytics
response = requests.post(
    "<base url>/text/transform",
    json={
        "text": ["User: John Smith"],
        "operator": "hash", 
    }
)

Output JSON

response.json()

{
  "entities": [
    [
      {
        "entity_type": "NAME",
        "start": 6,
        "end": 14,
        "score": 0.995
      }
    ]
  ],
  "stats": { "total_tokens": 4, "tps": 5420 }
}

{
  "text": [
    "Contact [EMAIL_ADDRESS]"
  ],
  "entities": [
    [
      {
        "entity_type": "EMAIL_ADDRESS",
        "output_start": 8,
        "output_end": 23
      }
    ]
  ],
  "stats": { "total_tokens": 8, "tps": 5420 }
}

{
  "text": [
    "My ID is ***-***-112"
  ],
  "entities": [
    [
      {
        "entity_type": "US_SSN",
        "output_start": 9,
        "output_end": 20
      }
    ]
  ],
  "stats": { "total_tokens": 10, "tps": 5100 }
}

{
  "text": [
    "User: <PERSON_8f4a2c>"
  ],
  "entities": [
    [
      {
        "entity_type": "PERSON",
        "output_start": 6,
        "output_end": 21
      }
    ]
  ],
  "stats": { "total_tokens": 5, "tps": 4950 }
}

For the full range of options, including how to configure entity types via YAML, please visit the documentation

Designed for the Agentic AI Era

Built for LLM guardrail use, PII Eraser natively supports OpenAI-format chats to ensure seamless, context-aware privacy.

Process All Messages

Slow

High

Scans entire history every turn. High accuracy, but scales poorly.

Hi, I need to update the beneficiary details for the 'Project Alpha' contract payouts.

I can help with that. Which specific banking detail do you need to amend?

The bank account number has changed for our UK entity.

Understood. Please provide the new 8-digit account number.

<BANK_ACCT>

Process Last Message

Fast

Low

Scans only the latest prompt. Fast, but misses PII requiring context.

Hi, I need to update the beneficiary details for the 'Project Alpha' contract payouts.

I can help with that. Which specific banking detail do you need to amend?

The bank account number has changed for our UK entity.

Understood. Please provide the new 8-digit account number.

82910453

Smart Context

Fast

High

Automatically includes relevant context. The optimal balance.

Hi, I need to update the beneficiary details for the 'Project Alpha' contract payouts.

I can help with that. Which specific banking detail do you need to amend?

The bank account number has changed for our UK entity.

Understood. Please provide the new 8-digit account number.

<BANK_ACCT>

OpenAI Chat Support

Pools context between messages for higher accuracy without processing history needlessly.

Flexible Configuration

Configure scanning for just completions or the last user prompt with intelligent context handling.

RAG Robustness

Optimized to handle RAG artifacts like chunking errors, scraping noise, and mixed-language content.

Great Accuracy, Globally

Accurate identification of 150+ entity types across Western Europe, North America and Australia.

150+ Localized Entity Types

Most systems are built for the US and retrofitted for the world with US-centric entity names like "SSN". PII Eraser is built from the ground up for global coverage with a focus on Western Europe, as well as Canada, US and Australia.

Regular Model Updates

The world changes fast. Older models fail on terms like "COVID" and aren't familiar with MCP tool calls. We continuously update our models to recognize contemporary entities and the shifting GenAI landscape.

No Regex Maintenance

PII Eraser relies on large encoder transformer models, freeing your team from maintaining fragile regex-based solutions. We also offer model updates free of charge in case we do miss something.

150+ Entities

System Architecture

Customer Infrastructure (VPC / On-Prem)

Self-hosted, air-gapped deployment within your secure environment.

Your Application /
Data Pipeline

REST API POST

Text or Chat with PII

PII Eraser Container

NLU & Entity Detection

Transformer Models

CPU Optimized

ARM (Graviton) & x86

Air-gapped Operation

No external network calls

JSON RESPONSE

Sanitized Data

Your Application /
Storage

ML Model / RAG

Optional: config.yaml Custom policies & entities

High Throughput

Capable of processing >5000 tokens/sec on a single instance via ARM SVX or x86 AMX instructions, such as AWS Graviton 4.

Flexible Deployment

Multiple deployment options including AWS Fargate, ECS and Azure Container Instances. CPU-only operation makes security scans, startup and autoscaling easy.

Seamless Migration

PII Eraser features Microsoft Presidio Analyzer compatibility endpoints, allowing for easy drop-in replacement.

Why PII Eraser?

Compare us against cloud APIs, open source libraries, and LLM-based approaches.

Deployment Method	Self-hosted Container	Cloud API	Local Library	LLM (Generative)
Data Sovereignty	100% Local / Air-gapped	Cloud Only	Local	Cloud (Mostly)
Cost Model	Hourly (Unlimited Tokens)	Per Character (Expensive)	Free (Maintenance Heavy)	Per Token (Very Expensive)
EU Localization	Native (DACH, FR, IT, etc.)	Limited (EN, ES, FR)	Requires Custom Regex	Varies (Training Bias)
Latency (p99)	< 20ms	~100ms	~200ms (Python overhead)	> 1000ms
Context Window	1M+ Tokens	Short (Chunks required)	Short	Variable
Hallucinations	Zero (Deterministic)	Zero	Zero	Possible

Get started in minutes with our free trial

Per-hour pricing on AWS and Azure Marketplaces. No sales calls and no lock-in.

Get Started