a continual learning agent

A multi-agent options trading system that combines price-based neural networks with a continual stock-selection learning agent. The PX5000 hourly neural network ranks thousands of stocks by predicted realized move across multiple horizons, without relying on directional alpha. A live candidate producer converts these volatility forecasts into near-ATM long straddle opportunities by comparing predicted move against option-implied move from Alpaca chains and quotes. The continual learning stock-selection agent adds a memory layer: it tracks prior research, catalysts, event quality, priced-in risk, confirmations, rejections, and symbol-level context over time. This memory is joined point-in-time with live option candidates, so the strategy can favor volatility setups backed by fresh catalysts and avoid names the agent has learned are weak, stale, or already priced in. Separate agents handle model inference, candidate production, option contract selection, execution gating, sizing, paper trading, exit management, monitoring, and strategy evaluation. The result is a modular contest-ready system where neural volatility forecasts provide the quantitative edge, while continual agent memory improves selection quality and adapts the strategy as new market information arrives.