
Primuez Guard is the "Firewall" for Agentic Commerce. As AI agents begin to spend money autonomously, the risk of automated fraud increases. We built a Zero-Trust security layer that vets every transaction before it happens. The Problem: Invoice fraud and phishing are evolving. Standard wallets do not verify who you are paying, leading to lost funds. I personally lost money to a "verified-looking" phishing link, which inspired this solution. The Solution: Primuez Guard intercepts invoice data and performs a ruthless forensic investigation: Visual Analysis: Uses Mistral OCR to read the raw invoice/contract. Deep Forensics: Uses Tavily to crawl the vendor’s website and search for "scam" footprints or deep-path phishing patterns. Reasoning: Uses DeepSeek V3 and Gemini to act as a "Judge," assigning a risk score (Safe, Caution, Unsafe). Settlement: If safe, it unlocks the payment UI to execute a USDC transaction on the Arc network. Feedback on Circle Origin (Vibecoding): We utilized the Origin "vibecoding" workflow to prototype our Payment Terminal UI. Even though we encountered regional access issues with the live API, the tool's concept significantly accelerated our frontend development. It allowed us to generate a professional, dark-mode component that visually integrates with the Arc ecosystem without writing boilerplate React code. Tech Stack: AI: Google Gemini 2.5, DeepSeek V3, Mistral. Blockchain: Arc L2, Circle USDC. Orchestration: n8n. Links: Repository URL: https://github.com/Primuez/Primuez-Guard Demo URL: https://primuez-guard.vercel.app Video Demo URL: https://youtu.be/yM2_8j5PDkQ
24 Jan 2026