Local prompt redaction gateway

Stop secrets and PII before they leave your machine for a public LLM

Problem

Developers paste code, logs, and customer data into ChatGPT, Claude, and Gemini all day, easily dragging API keys, passwords, card numbers, and emails along. Browser extensions catch some of this, but they miss CLI agents, desktop clients, and custom scripts, which is exactly where developers leak the most.

Solution

A local daemon or CLI wrapper that intercepts requests headed to LLM endpoints, using rules plus a lightweight local model to detect keys, PII, and internal domains. Matches get replaced with placeholders and logged, restored on the way back, or blocked outright with a prompt.

Users

Developers using CLI agents like Claude Code, Codex, and OpenClaw, plus small teams with compliance needs

MVP

HTTP proxy mode covering common LLM endpoints; built-in rules for keys, emails, card numbers, and internal domains; block-and-warn on match with a replace-and-send option; local log of blocked events

Revenue

Open-source core plus a team subscription ($5/seat/mo) for central policy, audit logs, and SSO

Why now

Agentic tools push far more context into LLMs than chat ever did, and existing protection only covers the browser; local-only detection avoids vendor dependencies and suits one builder