grid-safety

GRID Safety layer and Cognitive Privacy Shield: real-time safety enforcement, PII detection, masking, and compliance presets (GDPR, HIPAA, PCI-DSS) for AI inference.

Install

pip install grid-safety

Minimal usage

Cognitive Privacy Shield (library)

from safety.privacy import get_privacy_engine, PrivacyPreset, AsyncPIIDetector

# Get engine with a compliance preset
engine = get_privacy_engine(preset=PrivacyPreset.BALANCED)

# Detect PII
detections = await engine.detect("Contact me at user@example.com")

# Process (detect + mask/block/ask)
result = await engine.process("My SSN is 123-45-6789")
print(result.processed_text)  # Masked or blocked per preset
print(result.detections)

Run the Safety API server

safety-api

Runs the FastAPI app (auth, rate limiting, pre-check, Privacy Shield middleware, inference queue). Configure via environment variables:

Endpoints include:

• POST /infer — Submit inference (enqueued to Redis)
• POST /privacy/detect — Detect PII in text
• POST /privacy/mask — Mask or block PII
• POST /privacy/batch — Batch PII detection/masking (up to 100 texts)
• GET /health, GET /metrics

Set SAFETY_DEGRADED_MODE=true to run without Redis (e.g. for tests).

Architecture

The Cognitive Privacy Shield is GRID's core safety and privacy API. For pipeline details, integration points, and key files, see:

• Safety Layer & Cognitive Privacy Shield — Architecture (in the GRID repo)

License

MIT. See LICENSE.