GlassBox AI Quant

GlassBox AI Quant is an autonomous, news-driven options trading agent built on Alpaca’s developer stack. Its strategy is simple: compare the move implied by news with the move the options market is already pricing. Alpaca’s News API streams headlines in real time. A deterministic filter removes stale, repetitive, or weak stories before any model is used. Survivors go to an LLM that returns schema-validated estimates of expected move, horizon, confidence, and materiality. The LLM never places orders. GlassBox is driven by magnitude, not directional prediction. It compares the forecast move with the ATM straddle’s implied move and adjusts for movement that may already have occurred. If expected movement is materially above implied, it buys defined-risk convexity. If materially below, it sells defined-risk premium. If they broadly agree, it does nothing. Entry discipline is restrictive. A signal must survive sizing, liquidity, session timing, Greeks, concentration, correlation, macro-event, corporate-action, and other checks. The gate contains 19 named vetoes; any one can refuse the trade. Positions are defined-risk, submitted as atomic multi-leg limit orders, with deterministic IDs preventing duplicate fills. Models may influence selection and size, but deterministic code owns the limits: 1% base risk, 1.5% maximum loss per position, 6% portfolio heat, and bounded net delta. A separate supervisor can flatten and halt the system if the trader process fails. Open positions are managed through an ordered 11-barrier exit ladder covering deadlines, expiry and macro risk, thesis completion or break, stops, bell decisions, trailing protection, breakeven, profit, and time. Every decision, including every refusal, is written to a hash-chained audit log. Alpaca MCP and CLI provide independent inspection, while the dashboard shows why the system traded, why it refused, and which rule made the decision.