AI 챗봇을 위한 클라이언트 측 쉴드를 구축한 방법 (오픈 소스)

I was burning through API tokens on off-topic chatbot queries.

Every “Who are you?” = brand leak + wasted money.

So I built Shield by Vouch — a zero-cost, client-side filter that blocks prompt injections and off-topic queries before they hit your LLM.

The Problem

You build an AI chatbot for your business. Users ask:

• “Who are you?” → Reveals it’s ChatGPT
• “Python code for calculator” → Wastes API tokens
• “Ignore previous instructions” → Prompt injection

Every off-topic query = $0.002 gone. 1000 queries = $2 burned.

Why Client-Side?

Server-Side Client-Side Latency: ~100-500ms Latency: 0ms Cost: API call already made Cost: Zero if blocked

Auto-Healing Backend

• Paste any API key → Auto-detects provider (OpenAI, Groq, Gemini, Anthropic, OpenRouter)
• Model fails → Auto-switches to fallback
• Cross-provider failover → Zero downtime

Code Example

const shield = new VouchShield({
  businessType: 'restaurant',
  businessDesc: 'Italian food in Mumbai'
});

const result = shield.check("Who are you?");
// { blocked: true, response: "I can only help with our restaurant services..." }

Enter fullscreen mode Exit fullscreen mode

Results
80% off-topic queries blocked client-side
Zero identity leaks
$2 saved per 1000 blocked queries
Try It
🚀 Live Demo
⭐ GitHub
MIT License. No signup. Single file drop-in.