dataiku/kiji-proxy ? reverse-engineered prompt

Build me a privacy proxy for AI requests that sits between my apps and services like OpenAI, and automatically detects sensitive info in prompts before anything leaves my machine. I want it to mask things like emails, phone numbers, credit cards, SSNs, and addresses with fake but believable values, then restore the real data in the response so my app still works normally.

Please make it easy to run on macOS as a desktop app and on Linux as a standalone server, with a simple local web UI where I can see what got masked and delete saved mappings if I want. It should work transparently for browser traffic too, ideally with automatic browser setup on Mac, plus a Chrome extension for AI chat sites like ChatGPT, Claude, and Gemini. Use local model based detection if possible, and keep it fast with no external privacy checks. If you need to look up current docs online, go ahead.