Shield of Aeris TypeScript SDK

Protect your LLM applications from prompt injection, data leakage, and other AI security threats.

Installation

npm install @shield-aeris/sdk
# or
yarn add @shield-aeris/sdk
# or
pnpm add @shield-aeris/sdk

Quick Start

import { Shield } from '@shield-aeris/sdk';

// Initialize the client
const shield = new Shield({ apiKey: 'sk_test_your_api_key' });

// Scan a prompt
const result = await shield.scanPrompt('Hello, how can you help me today?');
console.log(`Safe: ${result.safe}`);
console.log(`Risk Score: ${result.riskScore}`);

// Handle unsafe prompts
if (!result.safe) {
  console.log(`Threats detected:`, result.threats);
}

OpenAI Integration

Automatically protect all your OpenAI API calls:

import OpenAI from 'openai';
import { Shield, UnsafePromptError } from '@shield-aeris/sdk';

const shield = new Shield({ apiKey: 'sk_test_your_api_key' });

// Wrap your OpenAI client
const client = shield.wrapOpenAI(new OpenAI());

// All calls are now automatically protected
try {
  const response = await client.chat.completions.create({
    model: 'gpt-4',
    messages: [
      { role: 'user', content: 'What is the weather like?' }
    ]
  });
  console.log(response.choices[0].message.content);
} catch (error) {
  if (error instanceof UnsafePromptError) {
    console.log('Blocked:', error.scanResult.threats);
  }
}

Manual Scanning

For more control, scan prompts and outputs manually:

import { Shield, UnsafePromptError } from '@shield-aeris/sdk';

const shield = new Shield({ apiKey: 'sk_test_your_api_key' });

async function handleUserMessage(userInput: string) {
  // Scan before sending to LLM
  const promptResult = await shield.scanPrompt(userInput);
  if (!promptResult.safe) {
    return { error: 'Your message was blocked for security reasons' };
  }

  // Call your LLM
  const llmResponse = await callYourLLM(userInput);

  // Scan the output before returning to user
  const outputResult = await shield.scanOutput(llmResponse);
  if (!outputResult.safe) {
    return { error: 'Response contained sensitive information' };
  }

  return { response: llmResponse };
}

Express/Hono Middleware

import express from 'express';
import { Shield } from '@shield-aeris/sdk';

const app = express();
const shield = new Shield({ apiKey: 'sk_test_your_api_key' });

// Add middleware to scan all incoming messages
app.use(express.json());
app.use('/api/chat', shield.middleware({ fieldName: 'message' }));

app.post('/api/chat', (req, res) => {
  // Request is already scanned - safe to process
  res.json({ response: 'Hello!' });
});

Configuration

const shield = new Shield({
  apiKey: 'sk_test_your_api_key',
  baseUrl: 'https://api.shieldofaeris.com', // Custom endpoint
  timeout: 10000, // Request timeout in milliseconds
});

Threat Types

The scanner detects the following threat types:

• prompt_injection - Direct prompt manipulation attempts
• indirect_injection - Injection via external content
• jailbreak - Safety bypass attempts (DAN, etc.)
• pii_detected - Personal identifiable information
• data_leakage - Sensitive data exposure
• toxic_content - Harmful or inappropriate content

Error Handling

import { 
  Shield, 
  ShieldError, 
  UnsafePromptError, 
  UnsafeOutputError 
} from '@shield-aeris/sdk';

try {
  const result = await shield.scanPrompt(userInput);
} catch (error) {
  if (error instanceof UnsafePromptError) {
    // Handle blocked prompts
    console.log('Risk score:', error.scanResult.riskScore);
    console.log('Threats:', error.scanResult.threats);
  } else if (error instanceof ShieldError) {
    // Handle API errors
    console.log('API error:', error.message);
  }
}

License

MIT License - see LICENSE for details.

Links

• Documentation
• Dashboard
• API Reference