Magno stops you bleeding capital on wide spreads and sudden market swings. It continuously identifies mispriced volatility while hardwired mathematical gates and automated rebalancing protect your book — even while you sleep.
Alpaca paper trading · $100,000 baseline · nothing to install
A live desk, not a mockup. The market underneath is walking on its own and the option is re-priced on every tick — the drift is real gamma, and the engine hedges it back without being asked.
Two clocks run side by side. Trade decisions every sixty seconds, risk correction every five. The fast one is deliberately deterministic — if the model is slow or unreachable, exposure is still being neutralised.
Pulls live option chains for SPY, QQQ, NVDA and AAPL, solves implied volatility from the NBBO mid, and ranks each name against a year of realised movement.
Three failures kill most autonomous options books. Magno is built around refusing each one.
Most bots buy into illiquid options and give up 5–15% the moment they enter. Magno rejects any contract whose bid/ask spread exceeds 5% of the mid price — before the order is ever constructed.
SPREAD_TOO_WIDE → rerouted to the tightest book
When a market swing pushes the book into directional risk, Magno submits fractional stock orders on Alpaca to snap exposure back to zero — every five seconds, without waiting on the model.
Sized to three decimal places. Rounding to whole shares would strand up to half a delta per name — most of the trigger threshold on a four-name book.
Naive systems net total portfolio delta to zero and quietly hide two opposing directional bets inside it. Magno buckets Greeks strictly per underlying.
Reads neutral. Is actually a live SPY/QQQ basis bet.
Each row is a specific failure mode, and a specific piece of code that refuses it.
| Risk dimension | Standard trading bots | Magno autonomous desk |
|---|---|---|
| Execution slippage | Pays heavy markups on illiquid contracts | Hard 5% spread gateVetoes bad pricing before the order is constructed |
| Market volatility | Ignores delta drift; bleeds on sharp moves | Continuous dynamic hedgingRebalances exposure back to 0.00Δ every five seconds |
| Cross-asset contagion | Nets SPY and QQQ together, creating a basis bet | Per-underlying isolationIndependent Greek buckets, hedged per name |
| Drawdown circuit breakers | Freezes the account, leaving open options trapped | Non-trapping breakerHalts new risk while hedging stays live |
| Model reliability | Unconstrained LLMs hallucinate tickers | Closed-menu schemaReasoner picks only from pre-cleared candidates |
The model proposes. Arithmetic disposes. Handing an agent a trading account is only reasonable if the model cannot be the last word.
Newton-Raphson implied-volatility solver with a bisection fallback in the wings. Never stalls when the broker returns null Greeks — which on paper accounts is most of the time.
One tool implementation, published over the Model Context Protocol. External agents drive the same code path as the autonomous loop, gates included.
Nine mathematical limits enforced in Python before any order is constructed. Pure functions over plain data — no network, no clock, no model.
Every gate verdict and fill streams to an append-only audit log with the observed value and the limit it was measured against.
Drag the market and watch exposure move. This is the same computation the terminal’s shock simulator runs against a real Alpaca book.
Drag the handle, or focus it and use the arrow keys. Every figure is re-priced through Black-Scholes at the shocked spot — the delta drift is genuine gamma on a 30-day SPY call, not a scripted curve.
Credentials are held in memory for the life of the session — never written to disk, never logged. The Alpaca client is constructed with paper=True unconditionally, so there is no code path to a funded account.