Skip to content

Reliability quick start＃

AI models can generate convincing but factually incorrect responses, known as hallucinations. Traditional approaches to validation often require manual review or complex rule-based systems that are time-consuming and difficult to scale.

The kluster.ai Reliability service addresses these challenges by providing real-time validation of AI-generated responses. It automatically detects hallucinations and ensures accuracy by analyzing the original prompt and the AI's response to determine if the output contains unreliable or fabricated information.

This guide will walk you through setting up the Reliability service, demonstrate a quick example, and show you the different integration options available.

Prerequisites＃

Before getting started, ensure you have:

  • A kluster.ai account: Sign up on the kluster.ai platform if you don't have one.
  • A kluster.ai API key: After signing in, go to the API Keys section and create a new key. For detailed instructions, check out the Get an API key guide.

Integration options＃

You can use the Reliability service through three methods:

  • Verify API - direct REST API endpoint for maximum control.
  • Chat completion - OpenAI-compatible endpoint using the klusterai/verify-reliability model.
  • MCP integration - connect to Cursor or other AI assistants for interactive verification.

Quick example＃

Here's the simplest way to check if an AI response contains hallucinations:

from os import environ
import requests
from getpass import getpass

# Get API key securely
api_key = environ.get("INSERT_API_KEY") or getpass("Enter your kluster.ai API key: ")

# Check if a response is reliable
response = requests.post(
    "https://api.kluster.ai/v1/verify/reliability",
    headers={"Authorization": f"Bearer {api_key}"},
    json={
        "prompt": "What is the capital of France?",
        "output": "The capital of France is London."
    }
)

result = response.json()
print(f"Hallucination detected: {result['is_hallucination']}")
print(f"Explanation: {result['explanation']}")

Response format＃

The API returns:

{
    "is_hallucination": true,
    "explanation": "The response incorrectly states that London is the capital of France. The capital of France is Paris, not London.",
    "usage": {
        "completion_tokens": 42,
        "prompt_tokens": 28,
        "total_tokens": 70
    }
}

Next steps＃