AlphaDesk: Bounded-Risk AI Options Trading Desk

AlphaDesk: Institutional-Grade AI Options Research & Paper Execution Desk AlphaDesk is an institutional-grade, paper-only options research and execution desk designed to solve the critical danger of modern AI trading: unconstrained language models hallucinating financial trades without deterministic risk parameters or safety controls. Rather than allowing an LLM direct access to broker execution endpoints, AlphaDesk establishes a rigorous, fail-closed separation of concerns. Trading decisions flow through an immutable, deterministic 5-stage pipeline: Quant Signal → Trade Idea → Structure Candidate → Risk Decision → Order Intent. Quantitative market indicators and options mathematics govern pricing, Greeks (Delta, Gamma, Theta, Vega), and mathematically bounded-risk spread selection (e.g., Bull Call Spreads with hard max-loss ceilings). Large Language Models (integrated via OpenRouter with BYOK support) serve strictly as read-only advisory analysts, producing structured qualitative assessments and catalyst evaluations with validated schemas and citations—without any authority to execute or modify risk parameters. AlphaDesk features an innovative dual-workspace architecture: 1. Public Demo Workspace (Green): A fully anonymous, zero-credential environment providing deterministic scenario replays (`TRADE`, `NO_TRADE`, and `VETOED`), an interactive 7-step guided tour, simulated CQRS positions/Greeks read models, and an operator kill-switch simulator. It is mathematically and architecturally barred from constructing broker adapters. 2. Connected Paper Workspace (Blue): A multi-tenant, Supabase-authenticated environment featuring AES-256-GCM encrypted credential vaults, live Alpaca market clock integration, real-time Alpaca market data and options chain scanning, broker account reconciliation, and explicit human-in-the-loop paper order confirmation.