Quick Start
Get LLMGuard Pro running in your application in under 5 minutes. This guide covers the essential steps to start protecting your LLM interactions.
1. Installation
Install the LLMGuard Pro SDK using your preferred package manager:
# npm
npm install @llmguard/pro
# yarn
yarn add @llmguard/pro
# Python
pip install llmguard-pro
2. Authentication
Get your API key from the dashboard and set it as an environment variable:
# .env file
LLMGUARD_API_KEY=lgp_sk_your_api_key_here
LLMGUARD_PROJECT_ID=your_project_id
3. Basic Configuration
Initialize the client and configure your security scanners:
import { LLMGuard } from '@llmguard/pro';
const guard = new LLMGuard({
apiKey: process.env.LLMGUARD_API_KEY,
scanners: {
promptInjection: { enabled: true, sensitivity: 'high' },
piiAnonymization: { enabled: true },
dataLeakage: { enabled: true },
toxicity: { enabled: true, threshold: 0.8 },
}
});
4. Scanning Prompts
Wrap your LLM calls with the guard to automatically scan inputs and outputs:
// Scan input before sending to LLM
const inputResult = await guard.scanInput(userPrompt);
if (inputResult.isClean) {
// Safe to send to LLM
const response = await openai.chat.completions.create({...});
// Scan output before returning to user
const outputResult = await guard.scanOutput(response);
return outputResult.sanitizedText;
} else {
console.log('Threat detected:', inputResult.threats);
return 'Request blocked for security reasons.';
}
Input Scanners
Input scanners analyze user prompts before they reach the LLM. Available scanners include:
Output Scanners
Output scanners inspect LLM responses before they reach the end user, preventing data leakage and ensuring content quality.
Custom Rules
Define custom security rules using our rule engine:
guard.addRule({
name: 'block-competitor-mentions',
type: 'output',
pattern: /competitor_name/gi,
action: 'redact',
replacement: '[REDACTED]'
});
REST API
Use the REST API directly if you prefer HTTP calls over the SDK:
POST https://api.llmguardpro.com/v1/scan
Headers:
Authorization: Bearer lgp_sk_...
Content-Type: application/json
Body:
{
"text": "User prompt to scan",
"direction": "input",
"scanners": ["prompt_injection", "pii"]
}
Webhooks
Configure webhooks to receive real-time notifications when threats are detected. Set up webhook endpoints in your dashboard settings.