Data Maskit maskit

Data Maskit is a local privacy gateway for large language model tools. It masks sensitive data in outgoing requests as structured placeholders before they leave your machine and restores them in real time as model responses stream back. It works with any tool that accepts a custom Base URL, such as Cursor, Claude Code, and Codex. All processing happens locally with zero telemetry. 💡 Why Maskit? When using Cursor, Claude Code, Codex, Pi, OpenCode, ChatGPT, or any AI coding assistant, sensitive information can easily be transmitted to external LLM providers: - 🔑 Credentials & Secrets: sk-proj-..., ghp_..., Cloud AccessKeys, JWT Tokens, private keys; - 🌐 Internal Infrastructure: DB connection strings (mysql://root:Pass123@192.168.1.50:3306/db), private IP addresses (10.x, 172.16.x, 192.168.x); - 👤 Business Privacy (PII): Phone numbers, ID numbers, real names, credit cards, proprietary internal project names. Maskit's Mission: Act as a transparent local privacy gateway between your developer tools and external AI providers — mask sensitive tokens before requests leave your machine, and restore them in real-time typewriter stream as answers arrive! ✨ Highlights & Feature Overview 🛡️ 1. Deep Masking with Multi-turn Consistency - 19 Built-in Scanner Rules: API Keys/Tokens, PEM keys, DB connection strings, phone numbers, ID cards, emails, credit cards, private IPs; - Custom Wordlists & Regex: Categorized custom dictionary for names, codenames, and proprietary business terms; full regex support; - Sliding-window Placeholder Reuse: Placeholders remain consistent across long conversations. "Alice" is assigned the exact same token in turn 1 and turn 20, preserving model reasoning consistency. ⚡ 2. Millisecond SSE Stream Takeover (Native Typewriter Flow) - Intercepts text/event-stream chunk by chunk; - Automatically reassembles split tokens across chunk boundaries, maintaining native typewriter responsiveness without lag. 🔌 3. No Root CA Installation + Native Fallback Passthrough (Never Breaks Your API) - Multi-port Reverse Proxy: Dedicated local ports per model channel (e.g. 18701 for OpenAI, 18703 for Anthropic). Change base_url to local port without installing untrusted self-signed root CAs; - Fallback Passthrough Guarantee: If the proxy is stopped or closed, ports automatically fallback to raw transparent passthrough. Your coding tools will never experience unexpected connection dropouts! 📊 4. Real-time Logs, Security Audit & Cost Tracking - Inspect full request/response diffs with one-click highlight mode; - Detect prompt leaks, model-swapping, and destructive commands; - Live token usage & model pricing cost estimation. 🔒 5. 100% Local Execution, Zero Telemetry - All masking and unmasking happen inside your local process. No analytics, tracking SDKs, or cloud telemetry.