Designing hood-agent: An Agentic Workflow That Makes Smart Choices

ai-agentssystem-designtrading

hood-agent is an always-on agent that trades for me. It listens to a Discord trading room, watches the market on its own, and decides whether to trade, short term or long term, through a Robinhood agent account. Every trade has to pass hard limits written in code, and doing nothing is always an option.

This is a design, not a finished system, and nothing here is financial advice. The diagram opens at full size.

Every component of Hood Agent. Sources feed collect workers that write a raw record by day. Think workers read the record: the Discord view and the agent's own eyes, with levels-watch computing support and resistance. They write the mind. The fast and slow agent loops read the mind, pass a sizer and a risk guard, and send orders to a paper broker and later Robinhood; every fill lands in the trades ledger. A learning loop after the close reviews, scores, dreams and promotes lessons into playbooks. A desk shows everything, and an iMessage remote control sends instructions in and alerts out.Every component of Hood Agent. Sources feed collect workers that write a raw record by day. Think workers read the record: the Discord view and the agent's own eyes, with levels-watch computing support and resistance. They write the mind. The fast and slow agent loops read the mind, pass a sizer and a risk guard, and send orders to a paper broker and later Robinhood; every fill lands in the trades ledger. A learning loop after the close reviews, scores, dreams and promotes lessons into playbooks. A desk shows everything, and an iMessage remote control sends instructions in and alerts out.

Data flows top to bottom: sources, collect, record/, think, mind/, act, trades/, learn, and back into mind/. Yellow boxes are model steps, blue boxes are plain code, and grey boxes are stores.