Research

AI strategy assistant

The AI strategy assistant authors a 0DTE SPX strategy from a plain-English description — you chat with it the way you’d brief a colleague, and it produces a complete, valid strategy that drops into the same saved-strategies pipeline as the form builder.

Plain language is the default — a description, a structured summary, and the strategy’s risk scenarios. The assistant handles the rest, and shows you the code behind it whenever you ask.

How it works

Authoring is a short loop you repeat until the strategy is one you’d trust with a live account:

Interrogate and refine loop back through the live preview as many times as you like — nothing is committed until you save.

  1. 1
    Describe your idea

    Brief it the way you’d brief a colleague — "16-delta iron condor on SPX 0DTE, exit at 50% of the credit or at 4:00 PM ET, no stop." Even a bare fragment like "a long call" works: your first message always becomes a complete strategy, with anything you left out filled in by convention (entered at the open, held to expiration). If your opening message doesn’t describe a position at all, you get the default — a short iron condor with $50-wide wings.

  2. 2
    The assistant authors it

    It streams its reply and asks one clarifying question at a time when something is ambiguous. By default you work in plain English: a natural-language description, a structured summary, and the strategy’s risk scenarios — each with its severity and worst-case loss. Ask to see the code behind the strategy and the assistant shows you that too.

  3. 3
    Live preview on every edit

    A real backtest re-runs beside the chat each time the strategy changes, so the equity curve, metrics, and daily P&L respond to every refinement — exactly the preview the form builder shows.

  4. 4
    Ask about the results

    Once a strategy is on screen, ask how it performed — "why did this session open no position?", "which sessions lost the most?", "was it ever stopped out?" — or click a day in the P&L calendar to ask about it directly. It reads the same results you’re looking at and explains them in plain trading terms, grounded in the actual numbers. When the cause can’t be pinned down from the data, it says so rather than inventing one — and if it turns up a real bug, it offers to fix it.

  5. 5
    Always valid, always on topic

    There is always a complete, backtestable strategy on screen. Your first message never blocks — vague or off-topic input falls back to documented 0DTE conventions and tells you what it chose. Once a strategy exists, a follow-up clearly unrelated to it gets a brief redirect back to your work instead of changing anything.

  6. 6
    Save when you’re happy

    There is no separate "submit" step. Publishing works exactly like the builder: the finished strategy is saved as an immutable strategy with its results attached, and you land on its detail page.

Good to know

Backtest it, then run it live

An assistant-authored strategy is a saved strategy like any other: evaluate its backtest, then create a bot from it to run it live, hands-free. What runs live is precisely the source you saved.

One trade per session

Each strategy opens a single trade per day — one entry, managed to its exits. Want several entries in a day? Make each its own strategy and combine them in a portfolio, which backtests each independently and aggregates the results.

Sometimes it speaks first

Once a strategy’s historical results are complete, the assistant may occasionally come back on its own with a short note — most often when the strategy turned out never to enter a trade in any of those trading sessions, which usually means an entry condition it could never satisfy. It either corrects the strategy or tells you what it found, in the same conversation. Notes like these are rare, and they count toward the same daily assistant limits as the messages you send.

The strategy builder remains available for point-and-click authoring of leg structures, relative strike selection, the entry time, and profit/stop/time exits.