ConvictionOS

ConvictionOS is a paper-only AI trading operating system built for the Alpaca AI Trading Agents Hackathon. Instead of letting an LLM freely place trades, ConvictionOS separates intelligence from execution: the AI can research catalysts, summarize market context, and form a thesis, but every trade must pass deterministic risk checks before it can reach the paper broker. The system focuses on options strategies across three mandates: catalyst trades over 1–10 days, swing trades over 2–8 weeks, and thematic positions over 3–12 months. Each mandate has its own risk budget, expiry logic, exposure limits, and abstention rules. If the quote is stale, spreads are too wide, broker state is missing, portfolio exposure is too high, or AI reasoning disagrees with quantitative checks, the agent abstains instead of forcing a trade. ConvictionOS includes a SaaS-style dashboard for judges, builders, and portfolio managers to inspect readiness, worker status, paper account state, strategy mandates, trade receipts, and risk decisions. A separate 24/7 worker handles scheduling and market-hours-aware operation, while broker reconciliation keeps the system honest about positions, orders, and account state. The core idea is “grounded AI plus deterministic execution.” ConvictionOS does not claim proven live P&L edge yet; it demonstrates the infrastructure required to safely evaluate and improve an AI trading agent in a real paper-trading environment: provider-agnostic AI, Alpaca paper trading integration, options-focused risk controls, transparent receipts, and a commercial-grade dashboard.