DontFeedTheAI

Why we included this project

This proxy sits between your shell and an LLM such as Claude or OpenAI, swapping out IPs, hostnames, credentials, and other sensitive values before they leave your machine and restoring them on the way back. Detection runs in two layers: regex catches structured items like IPs and API keys, while a local Ollama model picks up hostnames and credentials written in prose. Because both run locally, nothing sensitive crosses the boundary. It works with Claude Code, the OpenAI SDK, OpenRouter, and other OpenAI-compatible clients, and includes a per-engagement vault plus an audit dashboard for reviewing every mapping. Pentesters get the obvious benefit, but developers debugging with production data in regulated environments and consultants handling client contracts will find the same value.