TraceMind
AI-powered onchain investigator that traces wallet activity with evidence.
Demo
About
TraceMind is an AI-powered onchain investigator that lets users ask natural-language questions about a crypto wallet and receive an evidence-backed investigation report. The current MVP focuses on Aave V3 on Ethereum. TraceMind retrieves structured onchain activity through The Graph, performs deterministic financial normalization and aggregation, builds traceable evidence records, and uses Claude to investigate the retrieved activity and identify unusual behavior. Findings are linked back to the underlying onchain evidence so users can see what supports each conclusion.
How it's made
TraceMind is built with Next.js, TypeScript, Tailwind CSS, The Graph, Anthropic Claude, Zod, and Vitest. The user enters a wallet address and investigation question through the frontend. The backend retrieves Aave V3 Ethereum activity from The Graph using cursor-based pagination. Token amounts are normalized using retrieved decimals and financial aggregates are calculated deterministically with exact decimal-safe arithmetic. The system then builds stable evidence records and sends the retrieved evidence, computed aggregates, and completeness metadata to Claude for investigation. Claude produces structured findings that must reference real evidence IDs, while the application validates those references before displaying the final report. This keeps blockchain facts, calculations, and AI interpretation clearly separated.