PII Masking Proxy for LLMs

Strip user PII before prompts leave your server

Problem

More teams are calling LLM APIs, but sending real user data to OpenAI or Claude raises privacy and compliance worries. Someone built an open-source proxy that masks PII before a request leaves and rehydrates values in the response, and is now validating whether a B2B market exists. Meanwhile, many SaaS teams rebuild Presidio, NER and token maps from scratch—repetitive and error-prone.

Solution

A lightweight proxy that can be self-hosted or managed: point your LLM base URL at it, and it detects PII, replaces it locally, forwards to the upstream model, then restores values in the response. Signed audit logs are included; works with OpenAI, Claude and similar APIs. Focused on leak prevention rather than being a general gateway.

Users

SaaS and internal-tool teams sending user data to third-party LLMs, especially SMBs and regulated industries

MVP

Proxy supporting OpenAI-compatible endpoints; built-in PII rules (email, phone, name, card number); local mapping store with restore; one-command or Docker deploy; basic audit log.

Revenue

Open-source core plus managed subscription; usage-based or per-seat pricing; self-hosted enterprise license.

Why now

LLM calls are now in production, but privacy tooling is either expensive or heavyweight. An indie can own a self-hostable, minutes-to-integrate slice and earn small-team trust first.