Assay is a spending brain for research agents. Given a research question and a fixed USDC budget, the Python engine in agent/assay/engine.py estimates the marginal value-of-information of each candidate source, treats allocation as a portfolio problem, and decides buy, skip, or cache — logging a machine-readable rationale for every decision, refusals included. Purchases pay creators per use over x402 with Circle Gateway nanopayments on Arc testnet, handled by backend/src/x402.ts and backend/src/arc.ts alongside mandate and paymaster modules. Repeat questions reuse owned sources free through cache hits that must clear the same relevance bar as a fresh buy. A Next.js dashboard in web/ surfaces the payout ledger, and deploy/ ships systemd units and an nginx config.
Web3 Developer + Content Creator
1 (Solo) — Emin Karagöz
The spending brain for autonomous agents — decides what paid information is worth buying, enforces bounded budgets, records BUY/SKIP/CACHE decisions, and anchors run receipts on Arc mainnet.