llm-safe-pl

⚠️ DEPRECATED — use pii-toolkit instead

llm-safe-pl is end-of-life as of v0.2.1 (2026-04-29). No further features, fixes, or security updates will be released. The library has been superseded by the pii-toolkit family on PyPI:

Was	Now
llm_safe_pl.Shield, Anonymizer, Deanonymizer	pii-veil
llm_safe_pl.detectors.*	pii-core
llm_safe_pl.validators.* (checksums)	pii-core
Microsoft Presidio integration	pii-presidio

See MIGRATION.md for a full symbol-by-symbol map and before/after code samples.

v0.2.1 is a deprecation tombstone: the public API still works exactly as it did in 0.2.0, but import llm_safe_pl now emits a DeprecationWarning. Existing pinned installs keep functioning; you just won't get any new releases. Migrate at your own pace.

Why pii-toolkit replaces this

llm-safe-pl started as a Polish-only reversible-anonymization library. Building it surfaced a cleaner three-package architecture:

• pii-core — language-agnostic detectors and checksum validators with zero runtime dependencies. Polish detectors (PESEL, NIP, REGON, Polish IBAN, ID card, passport) ship out of the box alongside multi-language ones.
• pii-veil — the reversible Shield-style facade for LLM workflows, layered on top of pii-core. Same round-trip contract (anonymize → call LLM → deanonymize) you had in llm-safe-pl.
• pii-presidio — optional Microsoft Presidio plugin for projects that already standardize on Presidio's recognizer ecosystem.

The split lets each piece evolve independently: detectors without dragging in the LLM-workflow surface, the LLM workflow without locking in any one detection backend, and Presidio integration without infecting the dependency tree of users who don't use it.

Migrating

The fastest path:

pip uninstall llm-safe-pl
pip install pii-veil

Then update imports:

# Before (llm-safe-pl 0.2.x)
from llm_safe_pl import Shield

# After (pii-veil)
from pii_veil import Shield

Polish detectors are part of pii-core's default registry and are auto-loaded by pii-veil — no extra setup needed for the common case. For the full map including Match, Mapping, AnonymizeResult, PIIType, individual detector classes, and checksum validators, see MIGRATION.md.

The CLI is replaced by pii-veil's CLI; same subcommands (detect, anonymize, deanonymize).

Reporting issues

Issues against llm-safe-pl are no longer accepted. If you hit a bug while migrating, please open it against the relevant successor package:

• pii-toolkit/issues

Legacy documentation (v0.2.0)

The original v0.2.0 documentation is preserved below for users still on pinned installs. Nothing in this section will receive further updates.

What this library does

Reversible PII anonymization for Polish documents, designed for LLM workflows.

When you send a Polish document to an LLM (OpenAI, Anthropic, a local model), you're exposing PESEL numbers, NIPs, ID card numbers, addresses, and names to a third party. llm-safe-pl is built around the full round-trip:

1. Anonymize — detect Polish PII, replace with stable tokens, return a reversible mapping.
2. Call the LLM — the request contains no raw PII.
3. Deanonymize — restore original values in the response using the saved mapping.

Checksum validation (PESEL, NIP, REGON, Luhn, IBAN mod-97) is first-class, so valid-looking-but-wrong numbers are rejected before they become false positives.

Quick example — Python API

from llm_safe_pl import Shield

shield = Shield()

result = shield.anonymize(
    "Jan Kowalski ma PESEL 44051401359, NIP 526-000-12-46, email jan@example.pl."
)
# result.text    -> "Jan Kowalski ma PESEL [PESEL_001], NIP [NIP_001], email [EMAIL_001]."
# result.mapping -> reversible Mapping object (JSON-serializable)
# result.matches -> tuple[Match, ...] for audit

# Safe to send to an LLM now.
# response = call_any_llm(result.text)

restored = shield.deanonymize(result.text)
# "Jan Kowalski ma PESEL 44051401359, NIP 526-000-12-46, email jan@example.pl."

Quick example — CLI

# Detect PII without modifying the file
llm-safe detect document.txt

# Anonymize: writes rewritten text and a reversible mapping
llm-safe anonymize document.txt -o anon.txt -m mapping.json --force

# Restore original values
llm-safe deanonymize anon.txt -m mapping.json

What's supported (v0.2.0)

License

MIT. See LICENSE.