Build a strategy with AI-assisted formula writing
Build an inspectable strategy by defining your trading idea, using the supported calculator guide to request formulas, and reviewing BUY, SELL and STOPLOSS rules before testing and saving. Understand visibility, versioning and the eligibility and cost of evidence-based AI review. The formulas in this guide are educational examples, not a proven strategy or a promise of returns.
Build a strategy with AI-assisted formula writing — MP4 · SRT
Define a trading idea first
On the Strategies page, start with an understandable trading idea, a market and a risk budget. A strategy contains explicit BUY, SELL and STOPLOSS rules.
Treat the example formulas as material to inspect and test, not as evidence of a profitable trading system.
Name, describe and choose visibility
Give the strategy a meaningful name and describe its intended behavior and limits. The description should explain the trading idea rather than advertise guaranteed performance.
Choose Private for your own use, Limited for listed referrals or Public for community access.
Copy the indicator guide for your AI
Open Available calculators and use its clipboard button. Verify that the calculator definitions and AI-ready formula-writing guide have been copied.
Give this material to your chosen AI assistant along with your market, timeframe, entry, exit and risk requirements. Using the supported catalogue helps avoid invented indicator names and unsupported operators.
Ask for formulas you can inspect
Ask your assistant for separate BUY, SELL and STOPLOSS formulas using only the supported catalogue. Request a plain-language explanation and review the assumptions.
An illustrative response is not evidence that an AI has produced a profitable trading system. Inspect each expression before using it.
Paste the BUY formula
Paste only the BUY expression into BUY strategy, without explanatory prose or code fences. The educational example is: CurrentPrice > 5m-EMA-20 AND 5m-RSI > 52 AND 5m-MACD-hist > 0.
This expression combines price above a short-term moving average with momentum confirmation. Review the exact thresholds and missing-data behavior before testing.
Paste the ordinary SELL formula
Paste the ordinary exit rule into SELL strategy. The educational example is: ProfitSinceBuyPct >= 2 AND 5m-MACD-hist < 0.
This example waits for a profit threshold and momentum deterioration. An ordinary SELL may have different eligibility conditions from protective STOPLOSS, so do not assume it covers every losing position.
Paste and understand STOPLOSS
Put the protective rule into StopLoss strategy. The educational example is: ProfitSinceBuyPct <= -2.5.
The negative 2.5 percent position threshold is a rule condition, not a guaranteed maximum fill loss. The Plan must also enable StopLoss for the rule to act. Execution timing, gaps and costs still matter.
Test before relying on a formula
Use the formula-test controls to check evaluation against the selected symbol and review the result. A valid expression or one successful evaluation does not prove profitability.
Backtest and review behavior before risking capital. When you are ready to keep the strategy, use Save strategy deliberately.
Keep versions and sharing deliberate
The strategy list shows the strategy’s name, ID, version and usage. Review which version a Plan follows before changing rules, and check the Plan usage count.
The sharing control can produce an invitation linked to a strategy. Public visibility alone does not prove quality.
Check AI review eligibility and scope
The strategy owner can request AI review once per strategy family per UTC day when Plan evidence is available. A new strategy with no usage evidence may not be eligible.
Open AI review to inspect its scope and cost before confirming a request.
Understand the review charge and follow-up
The current review costs 100 credits after a successful AI analysis, even if the conclusion is to leave the formulas unchanged. Queueing alone does not charge, and a transport failure or unavailable response is not charged.
Requests are processed asynchronously, and the result arrives by email. Review any adopted change against later evidence.