How to build an AI trading bot (with real code)

Prerequisites

• Basic Python or JavaScript.
• An exchange account with API keys (use the testnet/sandbox first — never live keys while learning).
• pip install ccxt pandas for the Python path.

1. Fetch market data

Every bot starts by pulling OHLCV (open, high, low, close, volume) candles. ccxt gives one API across 100+ crypto exchanges:

python · fetch_data.pyimport ccxt, pandas as pd

exchange = ccxt.binance()                  # or bybit(), kraken(), ...
ohlcv = exchange.fetch_ohlcv('BTC/USDT', timeframe='1d', limit=365)

df = pd.DataFrame(ohlcv, columns=['ts','open','high','low','close','volume'])
df['ts'] = pd.to_datetime(df['ts'], unit='ms')
print(df.tail())

2. Compute the signal

Here is the exact SMA-crossover logic our backtester uses: go long when the 20-period average rises above the 50-period average, flat otherwise.

python · signal.pydf['sma_fast'] = df['close'].rolling(20).mean()
df['sma_slow'] = df['close'].rolling(50).mean()

# position: 1 = long, 0 = flat. Shift(1) avoids look-ahead bias.
df['position'] = (df['sma_fast'] > df['sma_slow']).astype(int).shift(1).fillna(0)

3. Backtest before risking money

Replay history and compute equity, return and drawdown — fees included:

python · backtest.pyfee = 0.001                                   # 0.10% per side
df['ret']      = df['close'].pct_change().fillna(0)
df['turnover'] = df['position'].diff().abs().fillna(0)
df['pnl']      = df['position'] * df['ret'] - df['turnover'] * fee
df['equity']   = (1 + df['pnl']).cumprod()

total_return = (df['equity'].iloc[-1] - 1) * 100
max_dd = ((df['equity'] / df['equity'].cummax()) - 1).min() * 100
print(f"Return: {total_return:.1f}%  Max DD: {max_dd:.1f}%")

This is exactly the math powering our in-browser backtester — try the same strategy there first to sanity-check your numbers.

4. Add risk management (the part that matters most)

Never size a trade by gut. Risk a fixed small fraction of the account per trade and let the stop define the size:

python · risk.pydef position_size(account, risk_pct, entry, stop):
    """Units to trade so hitting the stop loses exactly risk_pct."""
    risk_amount = account * (risk_pct / 100)
    per_unit    = abs(entry - stop)
    return 0 if per_unit == 0 else risk_amount / per_unit

# risk 1% of a $10k account, entry 42000, stop 40000 -> 0.05 BTC
units = position_size(10000, 1, 42000, 40000)

That's the same formula behind our position sizing calculator — open it side-by-side to verify your bot's sizing.

5. Paper trade, then go live (carefully)

1. Run on the exchange testnet for weeks. Confirm orders, fills and edge cases behave.
2. Add logging + alerts so you know the instant it breaks.
3. Go live with the smallest possible size. Scale only after weeks of live results match the backtest.

python · live_order.pyexchange = ccxt.binance({'apiKey': KEY, 'secret': SECRET})
exchange.set_sandbox_mode(True)              # TESTNET — always start here

if signal == 'buy':
    exchange.create_market_buy_order('BTC/USDT', units)

JavaScript version (browser-style signal)

If you prefer JS, the signal logic is identical. This is the core of the engine running on this very site:

javascript · signal.jsfunction sma(closes, p, i) {
  if (i < p - 1) return null;
  let s = 0;
  for (let k = i - p + 1; k <= i; k++) s += closes[k];
  return s / p;
}
// position[i] = 1 when fast SMA > slow SMA, else 0
const pos = closes.map((_, i) => {
  const f = sma(closes, 20, i), s = sma(closes, 50, i);
  return (f && s && f > s) ? 1 : 0;
});

Frequently asked questions