Trading Agents

Strategies

Tell your agent how to trade

A strategy is just a set of instructions — written in plain English — that tells your agent what to look for and how to trade. Think of it as a prompt. You describe your trading approach, and the AI follows it.

It's Just a Prompt

If you've ever written a prompt for ChatGPT, you already know how to write a strategy. There's no code, no formulas, no special syntax. You just describe what you want your agent to do in your own words.

Here's an example:

"Look for pullbacks to the 9 EMA in a strong uptrend. Only buy when all EMAs are stacked bullish (9 above 21 above 50 above 200). Enter on the first candle that bounces off the 9 EMA with a stop just below it. Target 2x your risk."

That's it. That's a strategy. The AI reads it, understands it, and follows it every time it checks the market.

What Goes Into a Strategy

A good strategy answers three questions:

When do I get in? This is your entry. What conditions need to be true for the agent to buy or sell? Maybe RSI is oversold. Maybe two moving averages just crossed. Maybe price hit a key level. Describe what a good setup looks like.

When do I get out? This is your exit. Does the agent take profit at a target? Does it trail a stop? Does it exit when an indicator flips? Tell the agent how to close the trade.

What do I avoid? This is just as important. Tell the agent when NOT to trade. "Don't trade when EMAs are tangled." "Sit out if RSI is in the middle." "Avoid entries during low-volume periods." These guardrails prevent your agent from forcing bad trades.

The AI Interprets, It Doesn't Just Follow Rules

Here's what makes this different from a traditional trading bot. A bot needs exact "if this then that" logic. Your agent reads your strategy the way a human trader would — it understands context.

Write "buy pullbacks in an uptrend" and the agent knows what a pullback looks like. Write "don't fight the 200 EMA" and it knows to respect that level as a trend filter. You don't need to define every edge case. The AI fills in the gaps using its understanding of markets.

Write it like you'd explain it to another trader. The more clearly you describe your approach, the better your agent will execute it. Be specific about what you care about and what you want to avoid.

Custom Instructions — Your Agent's Personality

Beyond the strategy, you can also write custom instructions that define how your agent thinks and behaves. This is the personality layer.

For example:

"You are a patient, disciplined scalper. You never chase entries. If you miss a setup, you wait for the next one. You always respect your stops — no exceptions. When in doubt, stay flat."

This shapes the AI's decision-making style. A conservative agent will skip marginal setups. An aggressive agent will take more swings. You decide.

Prebuilt Strategies

Not sure what to write? Every prebuilt agent comes with a fully written strategy you can use as-is or customize:

  • EMA God — Trades the EMA stack (9/21/50/200) with pullback entries and compression breakouts
  • MACD Master — Hunts MACD crossovers and histogram acceleration for momentum trades
  • Turtle Soup — Follows Donchian Channel breakouts with mechanical, rules-based discipline
  • Knife Catcher — Buys RSI extremes and Bollinger Band touches, betting on mean reversion
  • Basic Cross — Simple 9/21 EMA crossover, nothing more

Clone any of these and edit the strategy to make it your own.

Tips for Writing Better Strategies

  • Be specific about entries — "Buy when RSI is below 30" is better than "buy when it's oversold"
  • Define your exits — The AI needs to know when to get out, not just when to get in
  • Include what to avoid — Telling the agent when NOT to trade is half the battle
  • Match your indicators — If your strategy mentions EMA crossovers, make sure you've attached the EMA indicator to your agent
  • Keep it focused — A strategy that tries to do everything will do nothing well. Pick one approach and commit to it

Next Steps