Build a strategy with ChatGPT or Claude

How it works

A Janorium strategy is a JSON config — rules, not code. We publish its complete format as one file written for AI assistants: every field, every indicator with its parameters, the rules the validator checks and worked examples. An assistant that reads it can turn “buy the dip in an uptrend, stop below the swing low” into a config that Janorium accepts.

Any capable assistant works — ChatGPT, Claude, Gemini or another. It needs to accept a file of about 200 KB, or to be able to read a web page from a link.

Three steps

1. Give the assistant the reference. Download the strategy reference and attach it to the chat. An assistant that can browse the web can read it from the link in the prompt instead.
2. Copy the prompt below into the same message and replace the bracketed line with your idea, in your own words and in any language.
3. Paste the JSON into Janorium. Open a new strategy with the AI box, paste the config the assistant returns into the JSON box and press Apply. The form fills itself in; save the strategy and run a backtest.

The prompt

You are helping me write a trading strategy for Janorium, a crypto backtesting platform. A Janorium strategy is a JSON config, and its complete format is in the attached file janorium-strategy-config-reference.md (also online: https://janorium.com/ai/strategy-config-reference.md). Read all of it before answering.

My idea:
[Describe it in your own words — spot or futures, symbol and timeframe, when to enter and when to exit, how much to risk. If you leave something out, the assistant will choose a sensible value and say so.]

Please reply with:
1. The complete strategy config as one JSON object in a single ```json code block. Inside it, follow §18.14 of the reference: only the fields, indicator codes and operators the reference defines, with every indicator's params spelled out.
2. After the code block, a short plain-language explanation of the rules and of every value you chose for me.

Before answering, walk the validation checklist in §18.11. If part of my idea cannot be expressed in the config, say so instead of inventing fields. If I send you an error message from Janorium, fix the config and send the whole JSON again.

It asks for the config in one code block — chat apps put a copy button on it — followed by a plain-language explanation, so you can check that the assistant understood you before you run anything. Already have a strategy? The editor’s AI box copies a version of this prompt that carries your current config, for asking the assistant to change it.

How to describe an idea

• The market. Spot or futures, the symbol (BTCUSDT, ETHUSDT…) and the timeframe (15m, 1h, 4h).
• When to enter. Indicators, levels, crossings, filters — as much detail as you have. “When it looks cheap” is not a rule; “when RSI(14) crosses back above 30” is.
• When to exit. Take-profit, stop-loss, a trailing stop, an exit signal or a time limit.
• How much to risk. Position size or risk per trade, and leverage on futures.
• One idea at a time. A strategy that tries to do everything is hard to read and harder to trust.

When Janorium finds a mistake

Janorium checks every config before it runs. Apply in the editor checks the format; saving checks the rules that span several fields, such as a take-profit in R that needs a stop. If either reports an error, copy the message and send it back to the assistant — the prompt already asks it to answer with the whole corrected config. One or two rounds are normal for a complex strategy.

What the assistant cannot do for you

An assistant writes the config you asked for. It does not know whether the idea is any good, and a config that passes every check can still lose money. Assistants also misread requests now and then, so read the explanation it gives and look over the filled-in form before trusting a result.

Then judge the backtest like any other: on more than one symbol and period, with costs, and with suspicion of anything that looks too good. What backtesting is and how it misleads covers the traps.