prompt-privacy

Comparing classical ML, neural networks, DistilBERT, and Phi-4 for multilingual PII detection in LLM prompts.

Browser extension that catches PII in your prompt and lets you redact it before it leaves your machine - local detection, reversible tokenization, zero network calls.

CLI-first OpenAI-compatible LLM router/proxy for coding agents that picks the cheapest fast-enough model and proves speed, savings, and prompt-free privacy.

Redact PII in the browser before a prompt ever leaves the machine, then rehydrate the reply locally: 11 entity types, Luhn/SSN/IBAN checksum validation, and a type-checked Egress Guard on the send path. TypeScript. 173 tests, 100% coverage.

Local-first privacy gateway for LLM prompts: detects and pseudonymizes sensitive data before you send it, then restores it locally. Python library and CLI.