🏁 Post-Installation Guide: Next Steps

So you've just run npm install safeguard-ai. What now? This guide walks you through the first 10 minutes of using the library to ensure your implementation is robust and safe.

🏎️ The 1-Minute Verification

The quickest way to verify everything is working is to run a small test script.

1. Create a file named safeguard-test.js.
2. Paste this code:

const SafeguardAI = require('safeguard-ai');

async function verify() {
  console.log('🔍 Initializing SafeguardAI...');
  const moderator = new SafeguardAI();
  
  // Test 1: PII Detection (Doesn't require API key)
  console.log('Test 1: Local PII Detection');
  const result = await moderator.checkText("Email me at test@example.com");
  if (result.piiDetected.length > 0) {
    console.log('✅ PII Detection working: Found', result.piiDetected[0].type);
  }

  console.log('\n🚀 Installation verified! Next: Add your OpenAI API key to test full moderation.');
}

verify().catch(err => console.error('❌ Error:', err.message));

3. Run it: node safeguard-test.js

🛠️ Essential Setup Phase

1. Secure Your API Key

SafeguardAI uses OpenAI's high-performance moderation models by default. You MUST secure your key.

Don't do this:

const moderator = new SafeguardAI({ apiKey: 'sk-12345...' }); // ❌ UNSAFE

Do this:

1. Install dotenv: npm install dotenv
2. Create a .env file: OPENAI_API_KEY=your_key_here
3. Load it:

require('dotenv').config();
const moderator = new SafeguardAI({ apiKey: process.env.OPENAI_API_KEY }); // ✅ SAFE

🏗️ Choose Your Implementation Pattern

Depending on your application type, follow one of these patterns:

A. The "Gatekeeper" (Input Validation)

Use this if you want to block messages BEFORE they hit your database or LLM.

async function handleUserInput(input) {
  const check = await moderator.checkText(input);
  if (!check.safe) {
    throw new Error("Sorry, your message contains restricted content.");
  }
  return processMessage(check.cleanText); // Processes redacted version
}

B. The "Sanitizer" (Output Cleaning)

Use this to ensure your AI's responses don't leak sensitive info or generate toxic content.

const rawResponse = await openai.createChatCompletion(...);
const safeResponse = await moderator.checkText(rawResponse.choices[0].message.content);

displayToUser(safeResponse.cleanText); // Shows redacted/cleaned output

C. The "Compliance Officer" (Automated Audit)

Use this to scan batch data for GDPR/HIPAA compliance.

const logs = await db.getLogs();
for (const log of logs) {
  const audit = await moderator.checkText(log.content);
  if (audit.piiDetected.length > 0) {
      await db.markForRedaction(log.id, audit.cleanText);
  }
}

🛡️ Best Practices for Production

1. Fail-Safe Strategy: Decide what happens if the API fails.
   • Restrictive: Block the user if the safety check fails to run.
   • Permissive: Allow the user but log the failure for manual review.
2. Strictness Tuning: Start with medium. Switch to high if you're building a kid-friendly app or a corporate bot.
3. Custom Blocklists: The AI might miss your specific company's "no-no" words. Use moderator.rules.addBlockedWords(['competitor_x', 'secret_project_y']).

❓ Troubleshooting Common Errors

🔗 Deep Dives

• Full API Reference
• Configuration Guide
• Example: Express.js Integration

Need help? Open an issue on GitHub Issues.