Aegis-OptionAI: Risk-Governed Options Agent

Aegis-OptionAI is an autonomous US options trading system designed around a strict separation of powers: AI proposes, but deterministic software decides whether a trade can execute. A Featherless-hosted language model analyzes live Alpaca market context, including spot price, option-chain quotes, Greeks, implied volatility, recent trend, and portfolio value. It returns a structured BUY or HOLD decision, a bullish or bearish direction, a real contract pair, a conviction score, and plain-language reasoning. The model cannot access the broker, invent prices, or bypass risk controls. The strategy uses defined-risk bull-call and bear-put debit spreads rather than naked options. Every proposal enters an independent Chaos Sandbox that validates the option legs, checks bounded payoff behavior, computes net delta, and reprices the spread under three adverse scenarios: 6x bid-ask widening, implied volatility reduced to 20% of its input value, and a 10% adverse underlying move. A separate non-LLM risk engine then applies four hard vetoes covering chaos safety, maximum allocation, minimum conviction, and delta-dollar exposure. Alpaca's official CLI is the system's only broker interface. It provides live market data, account state, market clock, atomic multi-leg paper orders, positions, and exits on a dedicated $100,000 competition account. An autonomous scheduler operates every 15 minutes during market hours, manages existing positions before evaluating new opportunities, and applies deterministic take-profit, stop-loss, and expiry rules. Redis persists pipeline runs and trade history, while the React/TypeScript dashboard exposes AI reasoning, stress results, veto reasons, execution status, and forward P&L. The Chaos Sandbox test suite passes 161/161 tests, and a real two-leg Alpaca paper spread was filled, verified, reconciled, and closed.