Worca - An Agentic Swarm to automate Trading

Most autonomous trading bots run one strategy and see one thing. A gamma bot sees gamma; a volatility bot sees volatility; neither sees what the other sees, and the interesting moves usually show up where both are true at once. Worca runs two independent specialist agents against live Alpaca options data: Gamma Scout, which computes signed gamma exposure per strike to find dealer gamma walls and the flip point where dealer hedging pressure flips direction, and Vol Surface Surfer, which tracks the front-to-back IV term structure and flags when it dislocates from its recent range. Neither agent talks to the other directly. Instead, both publish signals to a shared bus that a Featherless-hosted LLM reads on a timer, reasoning over the combination rather than just which signal looks stronger, and decides how much capital each agent gets. Position sizing, stop-losses, and drawdown limits are all enforced by plain deterministic Python — the LLM never touches risk, only allocation. Trades execute through alpaca-py, including multi-leg calendar spreads for the vol agent.