CosyRedactGateway: local-first LLM privacy gateway adds high-entropy credential detection beyond known key formats

Products·concept

preflight — local proxy that scans LLM requests for secrets before they reach the model

preflight is a local proxy that intercepts LLM API requests and attachments, scanning them for secrets before they reach the model. It sits between any OpenAI-compatible client and the model, decoding and inspecting payloads, then redacting or blocking sensitive content. It solves accidental secret leakage for developers using coding harnesses and LLM tools.

preflight is a local proxy that intercepts LLM API requests and attachments, scanning them for secrets before they reach the model. It sits between any OpenAI-compatible client and the model, decoding and inspecting payloads, then redacting or blocking sensitive content. It solves accidental secret leakage for developers using coding harnesses and LLM tools.

Keyfence is a local proxy that sits between applications and LLM APIs to stop secrets and API keys from being sent in prompts. It addresses the risk of developers accidentally leaking credentials when using LLM services. It is aimed at developers and teams using LLM APIs.

Releases·Tool Release

Data Maskit: 本地隐私脱敏网关可自动对 AI 编码工具的请求进行脱敏并恢复响应

Data Maskit (数据面具) is a local privacy-masking gateway for large language models: it automatically redacts sensitive data in requests and transparently restores it in streaming responses. It targets AI coding assistants like Cursor, Claude Code, Codex, Pi, and any tool with a configurable Base URL. All processing runs 100% locally with zero telemetry, making it relevant to developers sending proprietary code to external model providers.

Data Maskit (数据面具) is a local privacy-masking gateway for large language models: it automatically redacts sensitive data in requests and transparently restores it in streaming responses. It targets AI coding assistants like Cursor, Claude Code, Codex, Pi, and any tool with a configurable Base URL. All processing runs 100% locally with zero telemetry, making it relevant to developers sending proprietary code to external model providers.

CleanPrompt is a single-page web utility that automatically masks secrets like API keys, JWTs, emails, and IPs in text before developers paste it into ChatGPT or Claude. It runs entirely client-side in the browser with no backend, so sensitive data never leaves the user's machine. It solves the risk of accidentally leaking credentials or private data when sharing code and logs with LLMs.

A new open-source Go tool, guardrails-llm-filter, acts as a transparent reverse proxy between clients and LLM providers, stripping sensitive data (PII, secrets) from requests and restoring them in responses. The model sees only placeholders, while the client gets original data. The tool is licensed under Apache 2.0 and released as v0.1.0.