AI Trader

An automated trading bot that uses machine learning to predict price movements and execute trades on the NEAR blockchain using the REF Finance DEX.

Overview

The bot combines several components to make trading decisions:

• Cloudflare D1 (SQLite) for storing historical market data
• TimeGPT (Nixtla) for price predictions
• Technical indicators from TAAPI
• Market data from Binance (via proxy)
• REF Finance DEX for executing trades
• Cloudflare Workers for serverless deployment
• Cloudflare KV for state management

How it Works

1. Data Collection (Every 5 minutes)

   • Gets technical indicators from TAAPI (VWAP, ATR, Bollinger Bands, RSI, OBV)
   • Fetches orderbook depth and liquidation zones from Binance
   • Stores everything in Cloudflare D1

2. Price Prediction

   • Uses TimeGPT to forecast next 24 5-minute price points
   • Applies time-decay weighting to prioritize near-term predictions
   • Uses fine-tuning with MAE loss for better accuracy

3. Trading Strategy The bot uses a dual-signal approach, combining AI predictions with technical analysis:

   AI Signal:

   • Applies time-decay weighting to short-term predictions
   • Uses tighter thresholds when position is open
   • Generates buy/sell/hold signal based on weighted forecast vs current price

   Technical Analysis Signal:

   • Calculates a score based on multiple indicators:
     • VWAP crossovers and divergence
     • Bollinger Bands breakouts
     • RSI oversold/overbought levels
     • OBV momentum (using square root scaling)
   • Generates buy/sell/hold signal based on final score

   Trading Decision:

   • Only executes trades when both signals agree (both buy or both sell)
   • Uses REF Finance Smart Router API for best swap prices
   • Implements stop loss and take profit thresholds
   • All positions are unidirectional (no shorts)

4. Position Management

   • Paper trading with simulated USDC balance
   • Tracks PnL, win rate, and other statistics
   • All positions are unidirectional (no shorts)
   • Uses actual DEX prices/liquidity for realistic simulation
   • State stored in Cloudflare KV

Configuration

The main configuration is in trading.ts:

const TRADING_CONFIG = {
    DECAY_ALPHA: 0.92,              // Exponential decay for new positions
    DECAY_ALPHA_EXISTING: 0.9,      // More conservative decay for existing
    UPPER_THRESHOLD: 0.002,         // +0.2% threshold for buying
    LOWER_THRESHOLD: -0.002,        // -0.2% threshold for selling
    UPPER_THRESHOLD_EXISTING: 0.0005, // +0.05% when position exists
    LOWER_THRESHOLD_EXISTING: -0.0005, // -0.05% when position exists
    STOP_LOSS_THRESHOLD: -0.02,     // -2% stop loss
    TAKE_PROFIT_THRESHOLD: 0.05,    // +5% take profit
    INITIAL_BALANCE: 1000           // Starting USDC balance
}

Setup

1. Environment Variables:

NODE_URL=<NEAR RPC URL>
NIXTLA_API_KEY=<TimeGPT API Key>
TAAPI_SECRET=<TAAPI API Key>
BINANCE_API_URL=<Binance API Proxy URL>

2. Database:

• Uses Cloudflare D1 (SQLite) for market data
• Uses Cloudflare KV for positions and state
• Schema includes tables for market data and indicators

3. Deploy:

yarn install
wrangler d1 create ai-trader-db
wrangler kv:namespace create ai-trader-kv
wrangler deploy

API Endpoints

• /history/:symbol - Get historical market data
• /latest/:symbol - Get latest market data
• /forecast/:symbol - Get current price forecast
• /position/:symbol - Get current position
• /stats/:symbol - Get trading statistics
• /portfolio - Get overall portfolio status

Monitoring

The bot logs detailed information about:

• Data collection status
• Forecast accuracy metrics (MAE, MAPE, R²)
• Technical analysis scores
• Trade decisions and reasoning
• Position updates and PnL
• Error conditions and recovery

Supported Markets

Currently supports NEAR/USDT on REF Finance with the following features:

• Real-time price data via Binance API proxy
• Full orderbook depth
• Smart Router API for best swap prices
• Position tracking

Development

To run locally:

1. Clone the repository
2. Install dependencies: yarn install
3. Set up environment variables
4. Create local D1 database: wrangler d1 create ai-trader-db --local
5. Start local development: yarn dev

Architecture Notes

• Uses a separate Binance API proxy service since Cloudflare Workers IP ranges are blocked by Binance
• Trading decisions use current market price for signals but actual DEX prices for execution
• All state is maintained in Cloudflare KV for serverless operation
• Uses REF Finance Smart Router API with fallback to single pool for best prices

Contributing

Feel free to submit issues and pull requests for:

• New trading strategies
• Additional technical indicators
• Market support
• Performance improvements
• Documentation updates