pii-protect 0.1.0

Pluggable, on-premise-first PII masking, unmasking, and redaction library.

pii-shield

A pluggable, on-premise-first PII masking, unmasking, and redaction library for Python.

pii-shield detects personally identifiable and sensitive business information in free text (emails, phone numbers, GST/PAN/IBAN numbers, person and organisation names, bank details, invoice/PO numbers, and more), and gives you three ways to handle it:

• mask it into a reversible placeholder token, encrypted at rest
• unmask a previously masked token back to its original value
• redact it permanently, with no way to recover the original

There is no server, no API, and no hard dependency on any particular database — it's a library you import and call directly. Where your encrypted PII values live is a pluggable choice: in-memory, a local file, Redis, or PostgreSQL, or a backend you write yourself.

---

Install

pip install pii-shield                      # core: regex detection + in-memory/filesystem storage
pip install "pii-shield[postgres]"           # + PostgreSQL storage backend
pip install "pii-shield[redis]"              # + Redis storage backend
pip install "pii-shield[spacy]"              # + spaCy NER layer (PERSON/ORG/GPE)
pip install "pii-shield[privacy-filter]"     # + transformer token-classification layer
pip install "pii-shield[all]"                # everything

Only cryptography is a hard dependency. asyncpg, redis, spacy, and transformers/torch are all opt-in extras. If you use a backend or detection layer without installing its extra, you get a clear OptionalDependencyMissingError telling you exactly what to install — never a bare ImportError or a silent failure.

---

Quick start

import asyncio
from pii_shield import PIIMaskingEngine
from pii_shield.storage import InMemoryStorage

async def main():
    async with PIIMaskingEngine(storage=InMemoryStorage()) as engine:
        result = await engine.mask("Contact [e-mail] about GST 27AAPFU0939F1ZV")
        print(result.masked_text)
        # "Contact {{EMAIL:abcc2}} about GST {{GST:9a03b}}"

        original = await engine.unmask(result.masked_text)
        print(original)
        # "Contact [e-mail] about GST 27AAPFU0939F1ZV"

        print(engine.redact("Contact [e-mail] about GST 27AAPFU0939F1ZV"))
        # "Contact [REDACTED:EMAIL] about GST [REDACTED:GST]"  — irreversible, nothing stored

asyncio.run(main())

PIIMaskingEngine is an async context manager — initialise() connects the storage backend, close() releases it. Use async with unless you need to control that lifecycle yourself.

---

The three operations

Method	Reversible?	Touches storage?	Use for
mask(text, scope=None)	Yes, via unmask()	Yes	Sending documents to an LLM/cloud service while keeping raw PII on-premise
unmask(masked_text, scope=None)	—	Yes (read)	Restoring original values before writing back to source systems
redact(text)	No	No	Logs, analytics exports, anything that must never contain recoverable PII

result = await engine.mask(text, scope="invoice-2026-00417")
# result.masked_text     -> text with {{TYPE:xxxxx}} placeholders
# result.token_count     -> number of PII spans masked
# result.entity_counts   -> {"EMAIL": 1, "GST": 1, ...}
# result.entities        -> per-span detail (type, offsets, token, confidence, source)

text_back = await engine.unmask(result.masked_text, scope="invoice-2026-00417")

scrubbed = engine.redact(text)   # synchronous — no storage or encryption involved

mask_dict() / unmask_dict() do the same over JSON-serialisable dicts, masking all string leaf values in one pass so a repeated value across fields still maps to the same token.

scope is a free-form string (e.g. a document or invoice ID). The same PII value repeated within one scope deduplicates to a single stored token instead of being encrypted and stored twice — pass the same scope to mask() and the matching unmask() call.

---

Architecture

                    ┌─────────────────────────────────────────┐
                    │             PIIMaskingEngine             │
                    │        (pii_shield.engine)               │
                    │                                           │
                    │   mask()    unmask()    redact()          │
                    └───────┬───────────┬───────────┬───────────┘
                            │           │           │
              ┌─────────────┘           │           └─── (redact never
              │                         │                  leaves this box —
              ▼                         │                  no encrypt, no store)
   ┌─────────────────────┐              │
   │      NEREngine        │            │
   │   (pii_shield.ner)     │            │
   │                        │            │
   │  RegexNERLayer   (always on)        │
   │  SpacyNERLayer   (optional)         │
   │  PrivacyFilterLayer (optional)      │
   │        │                            │
   │  TokenizerSafeSpanMerger            │
   │  SpanConflictResolver               │
   └──────────┬─────────────┘            │
              │ DetectedSpan[]           │
              ▼                          │
   ┌─────────────────────┐               │
   │ DeterministicToken-   │             │
   │ Generator              │             │
   │ (pii_shield.tokens)    │             │
   │                        │             │
   │  {{TYPE:xxxxx}}        │             │
   │  find_tokens_in_text() │◄────────────┘
   └──────────┬─────────────┘
              │ token, value_hash
              ▼
   ┌─────────────────────┐        ┌───────────────────────────────┐
   │   AESGCMCipher        │       │        StorageBackend           │
   │  (pii_shield.crypto)   │──────▶       (pii_shield.storage)       │
   │                        │       │                                 │
   │  encrypt() / decrypt() │       │  InMemoryStorage                │
   │  AES-256-GCM            │      │  FileSystemStorage               │
   └────────────────────────┘       │  RedisStorage       (extra)      │
                                     │  PostgresStorage    (extra)      │
                                     └───────────────────────────────┘

Components

PIIMaskingEngine (pii_shield.engine) is the single public entry point. It owns one NEREngine, one DeterministicTokenGenerator, one AESGCMCipher, and one StorageBackend, and wires them together for mask() / unmask() / redact(). This is the only class most callers need to import.

NEREngine (pii_shield.ner) does detection only — it never touches encryption or storage. It runs one or more layers over the input text and merges their output into a single non-overlapping span list:

• RegexNERLayer — always on, no extra dependencies. High-precision patterns for structured PII: GST, PAN, TAN, ABN, VAT, IBAN, SWIFT, account/sort-code/routing numbers, credit cards, email, phone (India + international), invoice and PO references.
• SpacyNERLayer — optional. Adds PERSON / ORGANISATION / ADDRESS detection via a local spaCy model. Requires pii-shield[spacy].
• PrivacyFilterLayer — optional. Adds detection via any HuggingFace token-classification model you point it at, run entirely on-premise through transformers.pipeline. Requires pii-shield[privacy-filter].

When layers disagree or overlap, SpanConflictResolver picks a winner (regex-validated spans win first, then financial-entity-over-phone, then higher confidence, then longer span), and TokenizerSafeSpanMerger stitches back together sub-word fragments that some transformer models emit at token boundaries.

DeterministicTokenGenerator (pii_shield.tokens) turns a detected span into a {{ENTITY_TYPE:xxxxx}} placeholder — a 5-hex-character suffix derived from SHA-256(value | entity_type | salt). Same value + same entity type always produces the same token within one salted instance, which is what makes within-document deduplication and find_tokens_in_text() (used by unmask() to locate placeholders) work. It also computes an unsalted value_hash used purely for storage-side deduplication, so multiple engine instances backed by the same storage can recognise a value they've each seen before, even though their salted tokens differ.

AESGCMCipher (pii_shield.crypto) is the only component that ever sees plaintext PII outside of the NEREngine. Each value is encrypted with AES-256-GCM using a fresh 96-bit IV, with the entity type bound in as additional authenticated data (AAD) — so a stored ciphertext can't be replayed under a different entity type. Storage backends only ever receive ciphertext, IV, and tag; they never see plaintext.

StorageBackend (pii_shield.storage) is an abstract interface with five methods a backend must implement: put, get, get_many, find_by_value_hash, touch (plus an optional log_access audit hook). PIIMaskingEngine depends only on this interface, which is what makes storage swappable without touching detection, tokenisation, or encryption code. Four implementations ship out of the box:

Backend	Persistence	Extra required	Notes
InMemoryStorage	None (process lifetime)	—	tests, short scripts
FileSystemStorage	Single JSON file, atomic writes	—	single-process, no external infra
RedisStorage	Redis hashes per token	pii-shield[redis]	shared across processes/hosts
PostgresStorage	Relational table, auto-migrated schema	pii-shield[postgres]	shared, queryable, audit-loggable

Writing a fifth backend (S3, DynamoDB, Vault, etc.) means subclassing StorageBackend and implementing those five methods — PIIMaskingEngine needs no changes.

Data flow

mask(): NEREngine.detect() finds spans → for each span, compute value_hash and check the backend for an existing token in this scope (dedup) → if new, AESGCMCipher.encrypt() the value → StorageBackend.put() the ciphertext/IV/tag → splice the {{TYPE:xxxxx}} token into the text in place of the original span.

unmask(): DeterministicTokenGenerator.find_tokens_in_text() locates every placeholder → StorageBackend.get_many() fetches all matching records in one round trip → AESGCMCipher.decrypt() each → splice the decrypted values back into the text. Tokens with no matching record are left in place with a [UNRESOLVED] suffix rather than raising, so a partially-available vault degrades instead of failing the whole call.

redact(): NEREngine.detect() finds spans → each span is replaced in-place with [REDACTED:ENTITY_TYPE]. Nothing downstream of detection is invoked — no cipher, no storage — which is what makes it genuinely irreversible rather than just "not currently reversed."

Design choices worth knowing about

• Encryption keys are supplied by the caller, not derived from or stored alongside vault data. This keeps a compromised storage backend from being sufficient on its own to decrypt anything, and keeps key rotation an application-level concern independent of which storage backend you choose.
• Detection is fully separated from storage. You can swap InMemoryStorage for PostgresStorage without changing anything about how PII is found, and you can add spaCy/transformer layers without touching storage at all.
• All storage backend methods are async, including InMemoryStorage and FileSystemStorage — so the same calling code works unmodified whether the backend is a Python dict or a networked database.

---

Configuring detection

from pii_shield import NEREngine, PIIMaskingEngine
from pii_shield.storage import InMemoryStorage

ner = NEREngine(
    enable_spacy=True,                 # PERSON / ORGANISATION / ADDRESS
    spacy_model="en_core_web_sm",
    enable_privacy_filter=True,        # any HF token-classification model
    privacy_filter_model="your/token-classification-model",
    privacy_filter_threshold=0.5,
    privacy_filter_device="cpu",
)

engine = PIIMaskingEngine(storage=InMemoryStorage(), ner_engine=ner)

NEREngine() with no arguments runs regex detection only, with zero extra dependencies.

---

Encryption key

from pii_shield.crypto import AESGCMCipher

engine = PIIMaskingEngine(storage=..., encryption_key="<64-char hex string>")
# or
engine = PIIMaskingEngine(storage=..., encryption_key=AESGCMCipher.generate_key())

If you omit encryption_key, an ephemeral one is generated and a warning is logged — anything masked in that session becomes permanently unrecoverable once the process exits. Always pass a stable key outside of quick experiments; losing the key makes every previously masked value permanently unrecoverable, by design.

---

Storage backend examples

from pii_shield.storage import InMemoryStorage, FileSystemStorage, RedisStorage, PostgresStorage

InMemoryStorage()
FileSystemStorage("./vault.json")
RedisStorage("redis://localhost:6379/0")
PostgresStorage("postgresql://user:pass@host:5432/mydb")   # creates its own schema on connect()

Custom backend:

from pii_shield.storage import StorageBackend
from pii_shield.types import TokenRecord

class MyBackend(StorageBackend):
    async def put(self, record: TokenRecord) -> None: ...
    async def get(self, token_value: str) -> TokenRecord | None: ...
    async def get_many(self, token_values: list[str]) -> dict[str, TokenRecord]: ...
    async def find_by_value_hash(self, value_hash: str, scope: str | None) -> str | None: ...
    async def touch(self, token_value: str) -> None: ...

---

Error handling

from pii_shield import (
    PIIShieldError,               # base class for everything below
    EngineNotInitialisedError,     # mask()/unmask() called before initialise()
    DecryptionError,               # AES-GCM tag verification failed
    StorageBackendError,           # backend-specific I/O/connection failure
    OptionalDependencyMissingError, # used a backend/layer without its extra installed
)

Running the tests

pip install -e ".[dev]"
pytest