Add OpenObscure — on-device privacy firewall for AI agents

[srianant]

What is OpenObscure?

OpenObscure is an on-device privacy firewall for AI agents. It intercepts LLM traffic and:

• Encrypts PII with FF1 format-preserving encryption (15+ PII types via regex + CRF + TinyBERT NER ensemble) — the LLM sees realistic-looking fake data, not [REDACTED] tokens
• Redacts images — solid-fills faces (SCRFD/BlazeFace), OCR text regions (PaddleOCR v4), and blocks NSFW content
• Voice PII detection — keyword spotting for PII trigger phrases in audio
• Cognitive firewall — scans LLM responses for persuasion/manipulation techniques

Why it belongs here

• Privacy-first: all processing happens on-device, real PII never leaves the machine
• Open source: MIT / Apache-2.0 dual license
• No tracking: project website has no analytics or user tracking
• Runs locally: gateway proxy mode (macOS/Linux/Windows) or embedded native library (iOS/Android via UniFFI)
• 1,800+ tests, 99.7% PII recall

Placement

Added to the ChatGPT section alongside PasteGuard and Shimmy (similar privacy-proxy category). Alphabetical order maintained.