How it works

1 Build

Open the builder. It starts on a complete worked example — or pick a template, or clear it and start empty. The builder is laid out in sections, each with a short explanation and a “More” link:

Strategy

A name, the symbol and the base timeframe the rules are checked on. The symbol list shows only symbols that actually have data, grouped by market, each with the date range stored. One symbol per backtest.

Indicators

Pick from 115 types. Each one gets an id, its inputs (pre-filled with defaults), an optional higher timeframe, and a source — a price, or another indicator’s output. The description under each block is generated from the engine itself.

Gates (Advanced)

A condition that switches on and stays on — for example “the 4h close is above its 200 EMA” — which entries can require. It can close the position when it switches off.

Entry and Exit

Long, short or both. Each rule compares a price, an indicator or a number with an operator (>, <, crosses above, crosses below…), and rules combine with ALL / ANY groups.

Risk & sizing

Stop loss, take profit (including an R-multiple of the stop) and a trailing stop — as a percent, an ATR multiple or a fixed distance — plus break-even in Advanced. Sizing is a fixed quantity, a percent of equity, or a risk-based size from the stop.

Execution and Limits (Advanced)

The execution model is shown as locked by the engine — signal at bar close, fill at the next bar’s close — with one editable cooldown. Limits are circuit breakers: entries per day, daily loss, consecutive losses.

For example, the RSI + MACD confluence template enters long when the MACD line crosses above its signal while RSI is between 50 and 70. In the spec pane that entry rule is:

{
  "entry": {
    "long": {
      "all": [
        {
          "op": "crosses_above",
          "left": "macd.macd",
          "right": "macd.signal"
        },
        {
          "op": "gt",
          "left": "rsi14",
          "right": 50
        },
        {
          "op": "lt",
          "left": "rsi14",
          "right": 70
        }
      ]
    },
    "short": null
  }
}

2 Validate

Press Validate. The spec is checked by the same Python validator the backtest worker uses — one definition of what a valid strategy is, not a browser-side imitation of it. You get either “✓ valid” with the timeframes the strategy uses, or a list of exactly what is wrong and where.

A spec that validates cannot fail spec validation when it runs. It can still fail for other reasons — a window with no data, say — which is why the next step checks the window too.

Validation needs no account. Validate & create saves the strategy, and that is when you are asked to sign in; the server validates it again on save.

3 Run a backtest

On the saved strategy’s page, choose a window. Simple offers last week, last 30 days, last 3 months, last year — each ending at the most recent UTC midnight, so no half-formed candle is included. Advanced takes an exact start and end (UTC, echoed back so there is no time-zone guesswork) and a starting balance.

Windows your plan’s history depth does not reach are marked, and a window outside the data stored for the strategy’s symbol is refused before launch with the real range shown. Backtesting perpetual futures needs the Pro plan.

4 Read the results

• Performance metrics — Sharpe, Sortino, max drawdown, CAGR, Calmar, win rate, profit factor and exposure up top; expectancy, average and largest win and loss, recovery factor and more below. A metric that is undefined for a run says “n/a” rather than guessing.
• The chart — the run’s own candles with every fill marked, the time in position shaded, and each indicator drawn on the price or in its own pane.
• The equity curve — realized and mark-to-market, side by side.
• Fills, and for perpetuals the funding and liquidation ledger.
• The immutable snapshot — the exact spec and run configuration, and the engine build that produced the result, so a run can be reproduced.

5 Build with AI

The AI assistant (signed in) has three modes: Build — describe a strategy in words and get a validated spec; Optimize — improve a saved strategy, running a few backtests along the way; and Discuss a result — interpret one of your runs. Some models are free; paid ones use credit, or your own provider key (BYOK) from Settings. Each message is sent to a third-party model provider, and the chat says so.

Prefer your own assistant? Connect it over MCP: mint a key, paste one config snippet into Claude, Cursor or any MCP client, and it can build, validate, launch and read backtests for you.

6 Or start from a template

Four ready-made spot strategies, from a two-indicator rule to the full risk kit, each with a read-only preview. Open the templates — loading one gives you an editable draft and nothing more.