Questions, answered plainly

A way to test trading ideas on years of real crypto market data by describing them to the AI you already use. Your AI writes the strategy; we run it against the past and show exactly how it would have done — every trade, every fee, and how it compares with simply holding the coin.

No. You describe the idea in a sentence (“buy when the 50-day average crosses above the 200-day, sell on the reverse”) and your AI does the rest. If you’d rather click, the Build page makes a strategy by hand.

Claude (on the web, the desktop app and Claude Code), ChatGPT, Cursor, VS Code, Windsurf and Codex. Claude and ChatGPT connect with a click; the others use a key from your account page. How it works has the steps for each.

You do, through your own AI subscription — we never charge for it and never run an AI ourselves. Our side is the backtesting.

50 backtests a month, on daily candles for BTC, ETH, SOL, and 1 forward test. Everything you save on the free plan is public. See pricing.

Pro isn’t on sale yet. While the site is in beta, testers get a month of it free, then go back to the free plan by themselves — ask on the pricing page. Pro has 1,000 backtests a month, every market and timeframe, private strategies, the full report, Optimize, portfolios and 20 forward tests.

Each run of a strategy over a period. Checking a strategy for mistakes is free. Running exactly the same thing again, over months that are over, reuses the saved result and costs nothing. One Optimize search counts as 20.

Binance’s public archive of USDⓈ-M perpetual futures: candles, the funding rates every position paid or earned, and the mark prices that decide liquidations. About 500 markets, most going back years. Every market is listed.

Any whole number of minutes, hours or days, from one minute to seven days — the standard ones (5m, 1h, 4h, 1d…) and custom ones like 27m or 3h on Pro. Nothing shorter than a minute: that’s the smallest candle Binance publishes. The free plan uses daily candles.

A signal is only known once its candle has closed, so by default the order fills at the next candle’s open. A rule that says “crosses above” fills during the candle, at the price where the lines met. Stops fill at their level — or at the open if the price jumped past them. When one candle touched both the stop and the target, we assume the stop: the candle can’t tell us which came first. Fees, slippage and funding are all charged. Methodology has the detail.

Usually because TradingView fills orders at the signal candle’s close by default, which is a price you couldn’t have traded at, and leaves out funding and liquidation. Small differences in data and in how stops are handled add up too. Ours are the less flattering choices, on purpose.

No. It shows how rules would have done in the past, and the best-looking result among many tries is partly luck — the report says how much. A forward test, on candles the strategy has never seen, is the honest check. Nothing here is financial advice.

No. A backtest is only saved when you (or your AI, when you ask it to) save it. Unsaved runs aren’t stored — they only count towards your monthly limit.

A public strategy shows its name, description and results to everyone on Browse and your profile; its rules are visible to Pro members. A private strategy (Pro) is seen by nobody but you. Your email is never shown.

A saved strategy run on new candles from the moment you start it, brought up to date a few times a day as each day’s prices are published. It can’t be edited or restarted into a better-looking record, so the gap between its backtest and its forward result is the most honest number on the site.

Several of your strategies run together, each with a share of the money (Pro). You see the combined result, each strategy’s part in it, and how closely they move together — which is the point of combining them.

No. It holds no money and no exchange keys, and places no orders. It simulates.

What you ask it to fetch through your connection: your own strategies and results, and the public ones on Browse. Text written by other users reaches your AI marked as someone else’s data, and it’s told never to follow instructions inside it.

Only one, to keep you signed in. Visit counts use no cookies, so there’s nothing to accept. The privacy page lists everything we keep.

Reset it here. If you signed up with Google, just continue with Google.

Email support@aibacktester.app from the address on the account; it’s done within 30 days. You can delete single strategies yourself, from their cards.