How it works

The full method behind an autonomous AI trading a real $1,000 account.

Every market day, an AI agent (Claude, driving the Robinhood MCP) runs the same disciplined loop. Nothing is hand-traded. Every decision and its reasoning is logged and published.

1. Gather signals

The bot pulls market news and RSS headlines, extracts cashtags, and scores sentiment across a watchlist of tickers. Headlines are deduplicated and passed through a relevance gate so only material news reaches the decision step.

2. Decide

Claude reasons over the signal brief and the current account snapshot. It proposes long-only trades — buy, hold, or skip — and writes a plain-English rationale for each. Those rationales are what you read on every day page.

3. Check guardrails

Guardrails are enforced in code, not by the model’s goodwill. An order is rejected before submission if it would breach any limit:

• Long-only — no shorting, no options, no leverage.
• ≤ 20% per position — concentration is capped.
• Daily cadence — one decision pass per market day.
• Stop / take rules — predefined exit logic on positions.

4. Execute & publish

Approved orders are submitted through the Robinhood MCP. The day is then rendered into a short vertical video and a public scorecard — equity, P&L, trades, holdings, and an end-of-day reflection — all posted here.

Data sources

Market data and prices come from the brokerage and public market feeds; news comes from financial RSS feeds and headline capture. Every cited headline links back to its source.

What it can’t do

This is a small, honest experiment — not a hedge fund. It can’t predict the market, it has no inside information, and it will have losing days. It trades a real $1,000 account so the stakes are real, but the goal is transparency, not returns. Nothing here is financial advice.