An AI knowledge-health system for docs: it finds contradictory claims across your documentation and routes each fix through human approval.
TruthLayer answers a question most documentation teams are afraid to ask: what if your docs could argue with themselves? It's a "knowledge health" dashboard for Sanity-backed documentation that continuously scans your docs, detects contradictory claims, and routes each one through human review. The dashboard shows an overall health score (the demo reads 58%, high risk), totals for documents, conflicts, and items needing review, plus metrics on consistency, freshness, review backing, and source coverage.
The conflicts it finds are the kind that quietly break products: an onboarding guide that enforces an 8-character password minimum while the enterprise compliance doc demands 12; media specs saying 10 MB per file while the attachment policy says 50 MB. For each conflict, an AI-written summary explains the discrepancy, a side-by-side view shows both sources, and a review workflow lets a human approve the resolution, reject it, or flag it as needing investigation — every decision logged with the reviewer's name.
The build is documented in the author's dev.to post for the Sanity "Vibe-Code Something Strange" challenge, where they write: "I built TruthLayer using Google Antigravity" and describe using the coding agent to turn the idea into a working product. The declared stack is Next.js 16 (App Router), TypeScript, Tailwind CSS v4, Sanity CMS, and the OpenAI API. For anyone maintaining docs at scale, it's a working demo of a genuinely useful pattern: AI does the hunting, humans keep the gavel.