A bot is just a rule that runs itself
Strip away the mystique and a trading bot is a rule executed automatically: 'buy when this happens, sell when that happens,' applied without sleep, hesitation, or second-guessing. The bot brings the discipline; the rule decides whether you make or lose money.
That's the part most bot guides get backwards. They spend chapters on plumbing - data feeds, order execution, hosting - and a paragraph on the strategy. But the plumbing is a solved problem. The open question, always, is whether your rule actually works.
The traditional route: Python, ccxt, and API keys
Building a bot by hand means writing Python (usually with a library like ccxt), fetching and cleaning market data, handling partial fills, retries, and edge cases - weeks of work before the first trade, and a subtle bug can quietly cost real money.
The shortcut most people take is a bot platform, which trades through your exchange account. That means handing a third party your exchange API keys - and key security is the perennial fear in every bot community, because a leaked key or a rogue platform has direct access to your funds.
The no-code route
The modern alternative removes both the coding and the key handover:
- Describe the rule in plain English - 'hold bitcoin while it's above its 100-day average, move to cash below.'
- The AI writes the strategy as real Python - code you can read and audit, not a black box of settings.
- Backtest it on years of real candles first, so the rule earns its place before it runs.
- Run it forward on live market data, tracking every trade it would make - without connecting your exchange account.
Backtest before you let anything trade
Most 'my bot lost money' stories are really 'my rule never worked' stories. A bot executes its rule faster and more consistently than you would - which means a bad rule loses money faster and more consistently too.
So the backtest isn't a preliminary; it's the point. The same rule that will drive the bot should first be run over years of history, across different market conditions, with fees included. If it can't survive the past, it has no business trading the future.
How Premiss does it
Premiss is the harness for exactly this loop: you type the rule in plain English, it writes the real Python, backtests it on years of real market data, and shows you the code, the trades, and a verified result. A strategy that survives can keep running forward on live data, tracking every trade it would make - and Premiss never holds your funds or your exchange keys.