ArgusTrader runs an eight-agent AI trading desk — analysts, a bull/bear debate, three risk officers — against a $100,000 paper wallet. It does the research, shows its work, and hands you the decision.
No card. No real money. The risk engine is written in code, not in a prompt.
RSI reset to 54 inside a healthy uptrend; FinBERT news flow positive (+0.28). 50-day SMA acting as support.
Every setup survives the same gauntlet your favorite hedge fund runs — compressed into minutes instead of a Monday memo.
20 mega-caps screened hourly: trend, momentum, ATR, volume spikes — plus FinBERT news sentiment and StockTwits crowd pulse.
A bull and a bear agent fight over the setup with identical data. A judge scores the evidence, not the rhetoric.
Three risk officers (aggressive, defensive, neutral) stress the trade. The PM only forwards setups that survive.
Risk-scored offers land in your desk with entry, stop and targets pre-sized to your wallet. Accept, resize or decline.
The same building blocks the pros pay for — running on paper money.
Eight LLM agents in a structured graph — analysts, researchers, trader, risk officers, portfolio manager — with the final word going to a human.
Every headline scored by a finance-trained transformer, recency-weighted before it touches the screener.
StockTwits posts with bullish/bearish tags plus sentiment scoring — the retail crowd, measured rather than followed.
1% risk per trade, ATR-based stops, exposure caps, and a daily-loss circuit breaker — pure code the AI cannot talk its way past.
A live timeline shows exactly which tool the AI is using at every second — and you can open any agent's written report.
Realistic fills with slippage and commissions, OCO-style brackets, full P&L history. The money is imaginary. The learning isn't.
“LLMs propose, code disposes. The model writes the thesis — but position size, stops and circuit breakers are arithmetic. That's why the risk engine never sees a prompt.”// the only rule in the building