GX CoreVaultsAI Trading Vaults

AI Trading Vaults

GX AI Trading Vaults allow users to deposit USDC and let an AI agent trade perpetual futures autonomously on their behalf. The AI uses a hybrid Quant + LLM architecture: a Rust-based indicator engine computes technical signals in real time, while an LLM (currently Claude by Anthropic) provides strategic reasoning and trade decisions.

Important Disclaimer: AI return targets are illustrative only. Only deposit what you can afford to lose entirely. Past performance does not guarantee future results. AI trading involves significant risk including the potential loss of your entire deposit.


How It Works

1. Connect Wallet
2. Select a Strategy Tier (Capital Guard / Growth / High Octane)
3. Deposit USDC
4. AI trades perpetual futures 24/7
5. Withdraw anytime

Execution Flow

  1. User deposits USDC to the GXVault smart contract on Arbitrum
  2. GXVault bridges USDC to GX Chain via the native bridge
  3. Vault approves the AI agent as an API Wallet (can trade but cannot withdraw)
  4. AI agent signs orders targeting the user’s account
  5. Orders execute with real liquidity and real PnL
  6. User withdraws: performance fee deducted from profits only, USDC returned

Custody

The AI agent uses GX Core’s native API Wallet mechanism:

  • The AI can place and cancel orders on behalf of the user
  • The AI cannot withdraw funds — this is enforced at the GX Chain protocol level
  • Each vault gets a separate API wallet for isolation
  • Withdrawal requires the user’s master wallet key

Strategy Tiers

Users choose one of three risk tiers when creating a vault.

Capital Guard (Low Risk)

ParameterValue
Target return1-3% per month
Max leverage3x
Max position size20% of equity
Max drawdown10% (auto-pause)
Daily loss limit2%
Trading frequency~10 trades/day
Min deposit50 USDC
Performance fee12% of profits above HWM

Behavior: DCA with trailing stops in trends. Tight 5-level grid when ranging. Goes to cash (no trading) in volatile markets.

Growth (Medium Risk)

ParameterValue
Target return3-8% per month
Max leverage5x
Max position size40% of equity
Max drawdown15% (auto-pause)
Daily loss limit5%
Trading frequency~30 trades/day
Min deposit100 USDC
Performance fee15% of profits above HWM

Behavior: Momentum (MA crossovers) with trailing stops in trends. Grid + mean reversion when ranging. Small mean reversion in volatile markets. Momentum entry on confirmation during breakouts.

High Octane (High Risk)

ParameterValue
Target return8-15%+ per month
Max leverage10x
Max position size60% of equity
Max drawdown25% (auto-pause)
Daily loss limit10%
Trading frequency~100 trades/day
Min deposit250 USDC
Performance fee18% of profits above HWM

Behavior: Leveraged momentum with pyramiding in trends. Wide 15-level grid + mean reversion when ranging. Quick scalps in volatile markets. Early aggressive entry on breakouts.


Strategy by Market Regime

RegimeCapital GuardGrowthHigh Octane
Trending UpDCA long, trailing stopMomentum (MA cross) + trailingLeveraged momentum, pyramiding
Trending DownDCA short bias, tight stopMomentum short + hedgingLeveraged short, scale-in
RangingTight grid (5 levels)Grid + mean reversionWide grid (15 levels) + MR
VolatileCash — no tradingSmall mean reversionMean reversion, quick scalps
BreakoutWait for confirmationMomentum entry on confirmEarly aggressive entry

AI Architecture (Five Layers)

Layer 1: Data Pipeline

Aggregates data every 30 seconds from five sources:

  • Market data — OHLCV candles, tick trades, funding rates, open interest via WebSocket
  • Orderbook microstructure — Bid/ask spread, order imbalance, whale wall detection (>$100K), depth analysis
  • News and sentiment — Crypto news feeds, social sentiment, Fear and Greed Index, macro calendar (FOMC, CPI)
  • On-chain data — Funding rates, OI changes, liquidation cascades, whale wallet movements
  • Cross-market signals — BTC dominance, ETH/BTC ratio, DXY correlation, S&P 500 futures

Layer 2: Quant Engine (Rust)

Computes technical indicators and classifies market regimes:

  • Indicators: SMA(9,21,50,200), EMA(12,26), RSI(14), MACD(12,26,9), Bollinger Bands(20,2), ATR(14), ADX(14), OBV, VWAP
  • Regime detection: Trending (ADX > 25), Ranging (ADX < 20), Volatile (high ATR%, low ADX), Breakout (BB squeeze + 2x volume)
  • Each classification includes a confidence score (0-100)

Layer 3: LLM Brain (Every 2-3 Minutes)

The LLM receives a structured JSON payload with all data from Layers 1-2 plus vault state (equity, positions, PnL, drawdown). It returns structured trade actions:

{
  "actions": [{
    "action": "open_long",
    "market": "BTC-USD",
    "size_pct": 15,
    "leverage": 5,
    "stop_loss": 66200,
    "take_profit": 69500,
    "confidence": 0.72,
    "reasoning": "BTC trending up, ADX 32, momentum confirmed"
  }],
  "market_outlook": "bullish_short_term"
}

“Do Nothing” is first-class. If confidence is below 0.3, the AI holds and logs its reasoning. Every action includes a confidence score and reasoning annotation for user transparency.

Layer 4: Risk Validator (Rust)

Every LLM output passes through a hard-coded Rust risk validator before execution:

  • Leverage within tier maximum
  • Position size within equity limit
  • Stop-loss present and ATR-reasonable
  • Daily loss budget not exceeded
  • Drawdown threshold not breached

If any rule is violated, the trade is rejected. This layer cannot be bypassed by the LLM.

Layer 5: Smart Router

Routes validated orders to the optimal venue:

ConditionRouting
All ordersExecute natively on GXCore matching engine

Revenue Model

Revenue SourceRateDescription
Performance fee12-18%Only charged on profits above the high-water mark
Builder code1 bpsCollected on every trade via GX builder code
Management fee0%No management fees on any vault

Users only pay when the AI makes money.


Supported Markets

All tiers trade the same markets: BTC-USD, ETH-USD, SOL-USD, DOGE-USD, XRP-USD, and BNB-USD. Additional markets may be added through governance.