Status: [UNTESTED] Endpoints found: http://127.0.0.1:8787/v1/chat/completions Source: https://github.com/Exestence/redaction-proxy # redaction-proxy **English** · [繁體中文](README.zh-TW.md) · [简体中文](README.zh-CN.md) · [日本語](README.ja.md) A small proxy on your own machine that replaces emails, API keys, ID numbers, and other personal data with placeholders before a prompt reaches a cloud LLM, and puts the real values back in the reply.  It works with any OpenAI- or Anthropic-compatible client, streaming included, and needs no local model. It **reduces** leaks of personal data and secrets; it does **not** make a conversation private. If content must never leave your machine, answer it with a local model instead; this proxy is for when you choose a cloud model and still want fewer leaks. > **Status:** 0.3.1, not yet on npm. Verified with real services: DeepSeek Chat Completions (2026-10-08), Anthropic Messages through Claude Code and dsh (2026-10-14), and the official Anthropic SDK against Anthropic Messages (2026-10-21); see below. The OpenAI API and Codex CLI are not verified yet. > This is a learning and exploration project. Issues and pull requests are welcome; I'll look at them when I have time. ## Intro video [](docs/media/intro.mp4) The recording replays real output of `pnpm demo` and of one more request through the proxy (captions in English). [MP4 version](docs/media/intro.mp4). ## Quick start (30 seconds) Requires Node 22.19+ or 24+ and pnpm. From a clone of this repository: ```sh pnpm install pnpm demo # no API key: fake upstream + proxy + one streamed request pnpm build node lib/cli.js # the proxy, on http://127.0.0.1:8787 ``` `pnpm demo` print