AI doesn't know what
it doesn't know
AI coding assistants confidently generate non-existent APIs, phantom npm packages, incorrect function signatures, and invented syntax. These hallucinations look correct but introduce subtle bugs that waste hours to debug.
How it works
Multi-Model Fact Checking
Submit to Multiple LLMs
Your AI-generated code is sent to up to 6 language models simultaneously for independent verification and cross-checking.
Compare for Consensus
Responses are analyzed for agreement and divergence across models. Areas of consensus strengthen confidence; disagreements flag potential issues.
Flag with Evidence
Hallucinations are flagged with confidence scores, specific evidence, and citations showing why a claim is likely fabricated.
What We Catch
Common Hallucination Types
Fabricated APIs
Detects when AI invents non-existent API endpoints, methods, or function signatures that look plausible but don't exist.
Phantom Dependencies
Catches references to npm packages, libraries, or modules that sound real but have never been published.
Invented Syntax
Flags made-up language features, incorrect method chains, or syntax patterns that won't compile.
False Facts
Identifies confidently stated but incorrect claims about frameworks, version capabilities, or technical specifications.
Scoring System
Weighted Analysis
Every check produces a composite score across four dimensions, weighted by their impact on code reliability. The default analysis model is sonnet-4.5.
Status Thresholds
Confidence Levels
Excellent
High confidence in output accuracy
Good
Minor concerns, generally reliable
Warning
Significant divergence detected
Critical
High hallucination probability
Detection by Plan
Higher tiers unlock more models and deeper analysis
| Free | Pro | Ultra | Enterprise | |
|---|---|---|---|---|
| Checks/day | 3/day | 15/day | 40/day | 100+/day |
| Models/check | 1 | 2 | 4 | 6 |
| Max context | 5K | 12K | 32K | 100K |
Explore More
Related Features
Stop hallucinations before they ship
3 free checks per day. No credit card required.