Hallucination Detector: Catch AI Fabrications with Multi-Model Consensus

The Hallucination Problem

AI models hallucinate — they generate plausible-sounding but incorrect information. Fake citations, invented statistics, non-existent API methods. This is especially dangerous in code generation, where a hallucinated function name compiles but fails at runtime. The Hallucination Detector catches these before they reach production.

Detection via Consensus

The detector works by sending the same query to multiple models and comparing their factual claims. When models disagree on specific facts, those claims are flagged as potential hallucinations. Claims that all models agree on are marked as high-confidence.

import vincony

client = vincony.Client(api_key="YOUR_KEY")

result = client.consensus.hallucination_check(
    text="The React useEffect hook was introduced in React 16.8...",
    models=["gpt-5", "claude-opus-4.6", "gemini-3-pro"]
)

for claim in result.claims:
    status = "✓" if claim.verified else "⚠ HALLUCINATION"
    print(f"{status}: {claim.text}")
    print(f"  Confidence: {claim.confidence}%")

Common Hallucination Types Detected

• • Invented API methods or parameters that don't exist
• • Fabricated research citations and statistics
• • Incorrect version numbers or release dates
• • Made-up library names or package imports
• • Wrong default values or configuration options

Pricing

Hallucination detection costs 5-10 credits per check. Available on Power ($54.99/mo) and Business ($199/mo) plans.