BullRun

BullRun is a multi-agent options trading system built for the Alpaca AI Trading Agents Hackathon. It runs a 4-stage pipeline: Scout detects market regime (BULLISH/BEARISH/NEUTRAL/VOLATILE), Quant selects defined-risk options spreads (bull call or bear put), a deterministic Risk Engine validates against 12 hardcoded safety gates, and CIO generates a plain-English trade thesis via GPT-4o-mini. The core insight: most trading agents optimize for autonomy. BullRun optimizes for accountability. The LLM never decides whether to trade — it only explains. Python does all math, validation, and execution. The human authorizes. Every decision is SHA-256 hash-chained for tamper-evident audit. During testing, the AI proposed using 18% of the portfolio despite a 2% allocation limit. Prompt instructions were not enough. We rebuilt it so safety is architectural, not behavioral — 12 deterministic gates the AI cannot override, including a 2% per-trade risk rule, portfolio heat cap, correlation guard, time-of-day guard, daily loss circuit breaker, and drawdown halt. Missing data defaults to REJECT, not PASS. BullRun connects to Alpaca's Trading API via alpaca-py for real multi-leg options execution with atomic fill verification. It includes an MCP server for AI tooling integration, a Next.js dashboard with TradingView charts, automated position monitoring with 7 exit rules, and a teaching engine that explains every decision in plain English.