How does it work inside?
No black boxes. A simple look at the tech stack and algorithms that determine the Truth Score.
1. Vector Memory (Supabase + pgvector)
Information isn't saved as plain text, but as 'Vectors' (mathematical coordinates). This allows the system to understand semantics and find hidden relationships instantly.
2. Autonomous Agent (Gemini 3 + Vercel AI)
When you search for something new, there's no human typing the answer. The system awakens an AI agent that investigates the internet in real time, reads scientific articles, news, and extracts evidence.
3. Mathematical Engine (Fuzzy Logic)
Truth is rarely 100% black or white. Fuzzy Logic takes AI evidence and assigns weights: a university paper (.edu) is mathematically worth more than an anonymous opinion blog.
4. Human Consensus (Bayesian Truth Serum)
The Oracle allows users to vote. Through the 'Bayesian Truth Serum' formula, the system punishes obvious answers or trolls, statistically rewarding expert knowledge that is 'surprisingly common'.
5. Forensic Transparency (The Glass Box)
Every logical deduction of the Agent (Chain of Thought) and source analysis is immutably recorded and audited, accessible in real-time through a terminal interface exposing structured reasoning.