@ny-squared/guard

Try PromptSentry Dashboard | Security Benchmark | Documentation

Unified LLM Security SDK — One-liner protection for OpenAI / Anthropic / Gemini

Features

• 孱・・One-liner protection 窶・guard.wrap(new OpenAI()) intercepts every LLM call
• 剥 OWASP Top 10 LLM scanning 窶・Prompt injection, jailbreaks, PII leakage, toxic content
• 笞｡ OSS mode 窶・Rule-based detection, no API key required, no network calls (Apache 2.0)
• ､・Pro mode 窶・ML-enhanced cloud detection (~95% accuracy)
• 迫 Multi-provider 窶・OpenAI, Anthropic, Google Gemini, and more

Installation

npm install @ny-squared/guard

Quick Start

import { guard } from '@ny-squared/guard';
import OpenAI from 'openai';

// Wrap your LLM client 窶・that's it!
const client = guard.wrap(new OpenAI());

const response = await client.chat.completions.create({
  model: 'gpt-4',
  messages: [{ role: 'user', content: 'Hello!' }],
});

Why zero dependencies?

Supply chain is exactly what you're protecting against. No transitive vulnerabilities. Ever.

Benchmarks

Input type	Latency (M1 Mac)
Short prompt	2ms
Long prompt (4KB)	5ms
Batch (100 items)	180ms

vs. alternatives

Feature	@ny-squared/guard	[others]
Zero dependencies	笨・	笶・
Local-first OSS	笨・	笶・
Risk score 0-100	笨・	笶・

Links

• 逃 npm: https://www.npmjs.com/package/@ny-squared/guard
• 倹 Dashboard: https://app.trypromptguard.com
• 当 Docs: https://app.trypromptguard.com/docs