Awesome Infra for AI › LLM Gateways & Proxies

packyme/privacy-filter

⭐ 339 Go repository created 2026-05-29

Privacy Filter is a high-performance, pure Go solution designed to safeguard sensitive user data (PII and secrets) by redacting it from text before it interacts with Large Language Models (LLMs). It operates as a privacy gateway, supporting deployment as a core Go package for direct integration into gateways, an HTTP service with a REST API, or a gRPC service using standard interfaces. The system employs a dual-layer detection mechanism: structured PII (emails, phone numbers, national IDs, bank cards, IP addresses) is identified using regex patterns, while secrets and credentials (API keys, tokens, private keys, high-entropy strings) are detected using a robust gitleaks ruleset, contextual regex, and Shannon-entropy fallback. Detected sensitive information is replaced with irreversible, typed placeholders. A key design principle is speed and efficiency; it's implemented without external models, GPUs, or CGO, ensuring millisecond-level latency across text lengths, even in production environments. It prioritizes the identification of high-risk identity data and avoids slower, model-based methods like NER for person/place/organization recognition. This tool is crucial for companies handling sensitive data and aiming for compliance by preventing unintended data exposure to LLMs.

https://github.com/packyme/privacy-filter

llm-privacydata-redactionpii-detectionsecrets-detectionllm-gatewaygoprivacy-filterai-securitydata-governanceinference-middleware