Name: AgentVeil
Author: AgentVeil

› How It Works

💻

Your App

Contains PII

PII →

AgentVeil

Mask ↔ Restore

Safe →

🤖

LLM API

Never sees real data

› Quick Start

bash

# Clone and run setup — builds native binary + background service

$ git clone https://github.com/vurakit/agentveil && cd agentveil

$ ./setup.sh

# Apply env vars, then use any AI tool as usual

$ source ~/.zshrc

$ claude # traffic now routes through AgentVeil

# Or run everything with Docker Compose

$ git clone https://github.com/vurakit/agentveil && cd agentveil

$ cp .env.example .env

$ docker compose up -d

# Point your AI tool to AgentVeil

$ export ANTHROPIC_BASE_URL=http://localhost:8080

$ claude

# Wrap any AI tool — one command

$ agentveil wrap -- claude

$ agentveil wrap -- cursor .

$ agentveil wrap -- aider --model gpt-4

$ agentveil wrap -- python my_agent.py

Works on macOS & Linux. Runs as a native background service — auto-starts on login, auto-restarts on crash.

› What It Does

PII & Secret Detection

39 detection patterns: CCCD, phone, email, passwords, API keys (OpenAI, AWS, Stripe...), JWTs, PEM keys, connection strings. Vietnamese & international formats.

Encrypted Vault

PII-to-token mappings stored in Redis with AES-256-GCM encryption. Session-isolated, auto-expires with configurable TTL. Role-based masking for viewer/admin.

Skill Auditor

Scan AI agent skill files for dangerous instructions. Behavior chain analysis, anti-evasion, and Vietnam AI Law 2026 compliance.

Prompt Guard

Block prompt injection, jailbreak attempts, and DAN attacks. Canary tokens detect data leaks in real-time.

Multi-Provider Router

Route to OpenAI, Anthropic, Gemini, Ollama simultaneously. Load balancing, fallback, and health checks built-in.

Compliance Reports

Pre-built compliance checks for Vietnam AI Law 2026, GDPR, and EU AI Act. Generate audit reports in JSON or HTML. Discord & Slack webhook alerts.

› Works With Everything

Claude Code

Cursor

OpenAI

Gemini

Ollama

LlamaIndex

GitHub Copilot

Python SDK

Go SDK

MCP Server

View all integrations →

› See The Difference

Without AgentVeil

# Your config.py — sent raw to AI ⚠️
DB_HOST = "192.168.1.100"
DB_PASSWORD = "Admin@123#"
ADMIN_EMAIL = "[email protected]"
CCCD = "012345678901"
API_KEY = "sk-live-abc123..."

With AgentVeil

# What AI actually sees — safe ✅
DB_HOST = "[IP_1]"
DB_PASSWORD = "[SECRET_1]"
ADMIN_EMAIL = "[EMAIL_1]"
CCCD = "[CCCD_1]"
API_KEY = "[OPENAI_KEY_1]"

› Why AgentVeil

Open Source & Free Forever

MIT licensed. Full source on GitHub, no vendor lock-in, no usage limits.

Vietnamese PII Support

Detects CCCD, SĐT, CMND, biển số xe, MST — built for Vietnamese developers.

Secret & API Key Protection

Auto-detects 23+ key formats — OpenAI, AWS, Stripe, GitHub, PEM keys, JWTs, and connection strings.

AI Agent Safety

Skill auditing with behavior chain analysis and anti-evasion. Blocks malicious instructions before they execute.

Transparent Proxy — Zero Code Changes

Plug & play. Just point your AI tool to AgentVeil and you're protected. Works with any LLM provider.

Vietnam AI Law 2026 Ready

Built-in compliance checks for Vietnamese AI regulations, GDPR, and EU AI Act. Export audit reports in JSON or HTML.