Quick Start Guide

Let’s get you up and running with Overseer’s AI safety checks in minutes. This guide will show you how to integrate our safety engine with your existing AI system.

5-Minute Integration

1. Set Up Your Environment Create a .env file to store your API keys. Each AI system you want to validate needs its own API key: You can find your api keys in the dashboard in the connections section after you’ve defined your systems

   # API keys for different AI systems
   OPENAI_SYSTEM_API_KEY=your_openai_system_key
   ANTHROPIC_SYSTEM_API_KEY=your_anthropic_system_key
   CUSTOM_SYSTEM_API_KEY=your_custom_system_key
   
   # You can add more systems as needed

   Multiple Systems

   Each system connection gets its own API key. This allows you to:

   • Track safety metrics per system
   • Apply different safety policies per system
   • Monitor usage separately

2. Install the SDK

   npm install @overseerai/sdk

3. Initialize Overseer

   import { Overseer } from '@overseerai/sdk';
   
   // Initialize clients for different systems
   const openaiValidator = new Overseer({
     apiKey: process.env.OPENAI_SYSTEM_API_KEY
   });
   
   const anthropicValidator = new Overseer({
     apiKey: process.env.ANTHROPIC_SYSTEM_API_KEY
   });

4. Implement Safety Checks

   // Example with OpenAI
   async function getOpenAIResponse(prompt) {
     const completion = await openai.chat.completions.create({
       model: "gpt-4",
       messages: [{ role: "user", content: prompt }]
     });
   
     // Validate using the OpenAI system validator
     const result = await openaiValidator.validate(completion.choices[0].message.content);
   
     if (result.isAllowed) {
       return result.text;
     } else {
       throw new Error("Response failed safety checks: " + result.details?.reason);
     }
   }
   
   // Example with Anthropic
   async function getAnthropicResponse(prompt) {
     const message = await anthropic.messages.create({
       model: "claude-2",
       messages: [{ role: "user", content: prompt }]
     });
   
     // Validate using the Anthropic system validator
     const result = await anthropicValidator.validate(message.content);
   
     if (result.isAllowed) {
       return result.text;
     } else {
       throw new Error("Response failed safety checks: " + result.details?.reason);
     }
   }

Common Integration Patterns

REST API

// Check AI response before sending to user
app.post('/api/chat', async (req, res) => {
  try {
    // Use the appropriate validator based on the AI system
    const validator = req.body.system === 'anthropic' ? anthropicValidator : openaiValidator;

    // Get response from appropriate AI system
    const aiResponse = await getAIResponse(req.body.prompt);
    const result = await validator.validate(aiResponse);

    if (result.isAllowed) {
      res.json({ response: result.text });
    } else {
      res.status(400).json({ error: 'Response failed safety checks: ' + result.details?.reason });
    }
  } catch (error) {
    res.status(500).json({ error: error.message });
  }
});

Error Handling

// Example with proper error handling
async function validateAIResponse(content, system = 'openai') {
  try {
    const validator = system === 'anthropic' ? anthropicValidator : openaiValidator;
    const result = await validator.validate(content);

    if (result.isAllowed) {
      return result.text;
    } else {
      console.warn('Content rejected:', result.details?.reason);
      return 'I apologize, but I cannot provide that response.';
    }
  } catch (error) {
    console.error('Validation error:', error);
    throw new Error('Failed to validate response');
  }
}

Best Practices

API Key Management

Keep API keys separate for each AI system to maintain clear usage tracking and system-specific policies.

Error Handling

Always implement proper error handling for safety check failures. Consider having fallback responses ready.

Safety Policies

Configure system-specific safety policies in the Overseer dashboard to match each use case.

---