QMA - Quant Memory Agent

@hoanlv214·Hoan Lai Van·6 followers·added 2026-07-19·updated 2026-07-26

QMA is a pay-per-call marketplace for quantitative market intelligence. A modular FastAPI backend under backend/app computes each report with KMeans regime clustering and Ledoit-Wolf covariance estimation in qma_engine.py, issues an invoice bound to provider, tier, and query snapshot, and verifies every required payment leg before an entitlement and a short-lived access token unlock the exact purchased report. Settlement runs through a TypeScript sidecar in arc_gateway/server.ts over Circle Gateway and x402 with @circle-fin/x402-batching. Every user gets their own wallet — /api/wallet/create provisions a per-user Circle Developer-Controlled Wallet, deposits and withdrawals flow through the same sidecar, and Gateway auto-deposits convert on-chain USDC into spendable balance without manual steps. Autonomous buying is a queued system: the qma-cli package in agents/ opens a budget-bounded session, a background worker in agents/src/worker.ts claims it atomically from the queue and runs the discover, compare, buy, and unlock loop, and AutonomousAgentModal.tsx in the React and Vite frontend follows the session live with status, spend against budget, purchases, and unlocked reports. Scalar-rendered OpenAPI docs cover the whole API from /openapi.json.