NEXUS — Onchain Code Audit Protocol

AI-native onchain code audit protocol built on GenLayer. Submit any GitHub repository with a claim about your codebase and let decentralized AI validators independently fetch, analyze, and reach consensus on the audit result — checked against an independent metadata source, with a genuine dispute mechanic.

Live: https://nexus-genlayer.vercel.app (frontend redeploy pending — see below)
StudioNet contract: 0x62E1D57199FefF364A00DeE0059abdE3303A286c
StudioNet deploy TX: https://explorer-studio.genlayer.com/tx/0x58711bd190df3187abbae8de83777b9de3e6a40c64999bc548571b640888135a
Bradbury contract: 0x33bcb525b70E6fe59883Eeb7BC0Efa557df1463d
Bradbury deploy TX: https://explorer-bradbury.genlayer.com/tx/0xa680b72424a52429ad096a9e1987174f8048511df706c47009d3405b35fe78b8
Network toggle: StudioNet / Bradbury, switchable in the navbar

---

Post-review update

An earlier submission of a similar single-fetch-single-judgment concept (SourceChecker) was rejected: a caller-selected page fetched and judged by itself only proves the page repeats a claim — it doesn't establish anything independently. This version addresses the same structural gap directly:

1. Independent metadata source. Every audit fetches the repo's own README/source and the GitHub REST API's repository metadata (stars, issues, last-updated, license) — structured facts the submitter does not author as prose, checked separately from the README content their claim is written in.
2. Genuine dispute mechanic. Any second party (not the original submitter) can call dispute_audit() once per audit with a counter-claim and optional counter-evidence URL. This triggers a fresh leader/validator run that fetches the counter-evidence and weighs it against the original claim, the README/source, and the independent metadata — and can revise the verdict. That creates a real submitter-vs-disputer pair instead of a single party asking an oracle for an opinion.

---

What It Does

Submit a GitHub repository + a specific claim → GenLayer validators independently scrape your README, source files, and GitHub API metadata → AI analyzes the claim against all three → consensus verdict stored permanently onchain, contestable once by any second party.

Verdicts: PASS · FAIL · NEEDS_REVIEW
Risk Score: 1 (low risk) → 10 (high risk)

---

Tech Stack

• Frontend: React + Vite + TypeScript + Tailwind CSS + Framer Motion
• Smooth Scroll: Lenis.js + GSAP ScrollTrigger
• SDK: genlayer-js ^1.1.7
• Contract: Python GenLayer Intelligent Contract
• Chain: StudioNet (61999)
• Deploy: Vercel

---

Architecture

User submits repo URL + claim
       ↓
submit_audit() write tx
       ↓
leader_fn() fetches README + source via GitHub raw API
       ↓
gl.nondet.exec_prompt() → AUDIT=verdict|risk|findings|summary
       ↓
validator_fn() independently re-fetches + re-analyzes
       ↓
Verdict match + risk score ±2 → consensus
       ↓
AuditReport stored in TreeMap[u256, AuditReport]

---

Local Setup

git clone https://github.com/Siriron/NEXUS
cd NEXUS
npm install
cp .env.example .env
npm run dev

.env variables:

VITE_CONTRACT_ADDRESS=0xD2d27d4Ca5cD0F38461a991E1F290173c30C292f
VITE_GENLAYER_RPC_URL=https://studio.genlayer.com/api
VITE_EXPLORER_URL=https://explorer-studio.genlayer.com

---

Contract

See contracts/nexus.py and docs/contracts.md.

No ETH handling. No token transfers. No financial rewards. Pure neutral utility registry