Strategy Overview

This strategy integrates a Large Language Model (LLM) with a professional quantitative trading framework using a classic Perceive → Decide → Execute three-stage architecture. The system runs on a minute-level cycle, continuously analyzing multiple major cryptocurrencies, generating actionable entry, hold, and close signals, and automatically executing orders with real-time risk monitoring on perpetual futures markets.

---

Core Architecture

The system consists of seven functional modules forming a complete data pipeline:

① Schedule Trigger
The system's heartbeat, activating the entire decision chain at a minute-level interval to ensure continuous operation during active market sessions.

② Account State Management
Initializes baseline data and persists account state across cycles. Tracks invocation count, runtime duration, total account value, available cash, and cumulative return percentage — providing the AI with a complete financial context for each decision cycle.

③ Market Data Acquisition
The system's perception layer. For each target coin, it concurrently fetches two timeframes of candlestick data:

• Short timeframe (minute-level): Captures precise entry and exit timing
• Long timeframe (hourly-level): Identifies primary trend direction and market structure

Technical indicators are auto-calculated for both timeframes: EMA20/50, MACD (12/26/9), RSI (7/14), and ATR (3/14). The most recent 10 candles' full indicator series are passed to the AI, giving the model temporal awareness of evolving market conditions.

④ Position Data Acquisition
Tracks real-time position status for each coin: quantity, entry price, current price, unrealized PnL, leverage, liquidation price, and pre-stored exit plans (take-profit/stop-loss). The quantity field serves as the core status flag — null indicates no position (entry eligible), while non-null restricts the AI to hold or close only, architecturally preventing pyramiding.

⑤ Data Merge Node
Consolidates market data and position data into a unified JSON structure, serving as the standardized context input for the AI decision engine.

⑥ AI Decision Engine
The system's core brain, utilizing a two-layer prompt architecture:

• System Prompt defines the AI's trading role, risk boundaries, decision framework, and output format specifications. Core constraints include:

  • Maximum 6 concurrent positions across all coins
  • Maximum 3–5% account risk per trade
  • Minimum risk-reward ratio of 2.5:1
  • Every trade must include a take-profit price, stop-loss price, and a natural language invalidation condition
  • No adding to existing positions

• User Prompt dynamically injects current market state at each invocation: runtime stats, return percentage, available cash, full technical indicator data for all coins, and current position details.

The AI performs dual-timeframe analysis — using the long timeframe to determine trend direction and the short timeframe to identify optimal entry timing. Output is a strictly formatted JSON trading signal per coin, containing: signal type (entry/hold/close), take-profit and stop-loss levels, recommended leverage, risk amount in USD, confidence score, and trade justification.

⑦ Trade Execution & Monitoring
After signal parsing, the execution layer handles:

• Precision management: Auto-fetches exchange-specific tick size and minimum order quantity for each pair
• Position sizing: Derives contract quantity from risk amount and stop distance (Risk Amount ÷ Stop Distance), capped by account leverage limits
• Direction detection: Automatically identifies long vs. short by comparing take-profit and stop-loss prices
• TP/SL monitoring: Actively polls all positions each cycle; closes immediately when price reaches target or stop level, then clears the corresponding exit plan record
• Visual dashboard: Displays a real-time tabular summary of AI signals and current positions for human review

---

Risk Management Framework

Three layers of risk protection are built into the system:

Layer	Mechanism	Description
1st	Capital Allocation	Account equity divided equally across coins; total portfolio risk is bounded
2nd	Stop-Loss Placement	ATR-based dynamic calculation; minimum stop distance ≥ 2.5%
3rd	Invalidation Condition	Active exit when technical structure breaks, independent of stop-loss price level

---

Suitable For

• Quantitative researchers looking to leverage AI for technical analysis and trade decision-making
• Traders requiring fully automated execution with 24/7 unattended monitoring
• Users interested in multi-coin rotation strategies

Risk Disclaimer: This strategy involves leveraged trading. Cryptocurrency markets are highly volatile, and AI-generated decisions carry inherent uncertainty. Please conduct thorough backtesting and allocate only appropriate capital before live deployment. This strategy is intended for technical research and educational purposes only and does not constitute investment advice.

Vedio Link：AlphaArena Replica Trading System