Vector52

Demo

About

Vector52 is an evidence-first onchain investigation platform for humans and AI agents. It transforms a wallet, transaction, or claim into a reproducible investigation where every graph connection can be traced back to observable blockchain evidence. Most blockchain explorers show transfers but do not explain what each connection actually proves. They often turn interactions with routers, pools, and smart contracts into misleading wallet attributions. AI-generated reports can amplify this problem by producing confident conclusions without preserving the original query, raw data, methodology, or limitations. Vector52 separates evidence from interpretation. Its forensic Core acquires and normalizes onchain activity, preserves source data with SHA-256 integrity, reconstructs wallet flows, and classifies every result as evidence, claim, hypothesis, assertion, or unknown. When the available information is insufficient, the system returns `UNKNOWN` instead of inventing an explanation. The Graph provides indexed protocol and event data for efficient investigation and reproducible queries. Uniswap activity is analyzed with protocol-aware rules that distinguish total pool or router volume from the value that can actually be attributed to the investigated address. This prevents a wallet interaction from being incorrectly presented as responsibility for an entire protocol transaction. AI agents access Vector52 through structured tools and ARC-oriented agent workflows. Using x402, an agent can discover a forensic capability, receive its price, pay for a single request, and obtain structured evidence without requiring a subscription or privileged API account. The agent may explain the findings, but deterministic services perform the calculations and preserved evidence supports the final result. Investigations can be explored through an interactive web application and exported as portable `.v52` evidence files containing transfers, provenance, query limits, warnings, the generated report, and a SHA-256 integrity digest. This allows another researcher, application, or agent to independently inspect and verify the investigation.

How it's made

Vector52 is built as five modular layers: a React investigation interface, a FastAPI evidence Core, blockchain data adapters, an MCP server for AI agents, and an x402 payment boundary. The frontend is an installable PWA built with React, TypeScript, Vite, Reown AppKit, Wagmi, and viem. Investigators can enter a public address, define a date range, add optional context, and explore the resulting flow through an interactive graph. Every edge retains its transaction hash, block, timestamp, asset, value, direction, counterparty, and data source. The FastAPI backend performs validation, evidence acquisition, normalization, reconciliation, and report generation. Provider responses are preserved and hashed with SHA-256 before analysis. Token values remain integer strings with their decimals instead of being converted to floating-point numbers, preventing silent precision errors. The Graph supplies indexed protocol events and Uniswap activity. We reconcile indexed data with RPC receipts and logs, then apply protocol-aware normalization so interacting with a Uniswap router or pool is not automatically interpreted as ownership of, or responsibility for, the entire transaction volume. Provider disagreements and indexer delays produce explicit states such as `PARTIAL`, `DEGRADED`, or `UNKNOWN`. Our MCP server is built with Node.js, TypeScript, the Model Context Protocol SDK, and viem. It exposes the same forensic Core to ARC-oriented AI agent workflows, ensuring that agents and human investigators receive the same evidence and limitations. x402 provides machine-to-machine, pay-per-request access. A protected endpoint first returns an HTTP 402 challenge containing the price and payment requirements. The client validates those conditions, authorizes the exact payment, and retries the original request with proof. Only after verification does the backend execute the investigation. No private key is sent to the API or included in an agent prompt. One notable design choice is that the AI layer never calculates token amounts or creates forensic verdicts. Deterministic services compute the result; the agent only requests tools and explains structured evidence. Investigations can be exported as `.v52` files containing the evidence, provenance, limits, warnings, report, and SHA-256 integrity digest, allowing another application or agent to verify the result independently.