An AI-powered crypto futures trading bot for Hyperliquid DEX with a full terminal UI, built with Python and Claude AI.
┌──────────────────────────────────────────────────────────────┐
│ HYPERLIQUID AI BOT [PAPER] [TESTNET] HH:MM:SS│
├────────────────────┬────────────────────┬────────────────────┤
│ PRICES & MARKET │ OPEN POSITIONS │ TECHNICAL │
│ BTC $69,833 +1% │ BTC LONG 0.05 │ RSI MACD ADX BB │
│ ETH $2,148 -1% │ ETH SHORT 2.0 │ 65 +50 32 0.7 │
│ SOL $175 +3% │ │ 28 -20 18 0.1 │
├────────────────────┼────────────────────┼────────────────────┤
│ SIGNALS │ AI DECISIONS │ BOT STATS │
│ BTC LONG str=0.72 │ BTC BUY conf=85% │ Equity: $10,536 │
│ ETH SHORT str=0.65 │ ETH HOLD conf=40% │ PnL: +$536 │
│ │ │ Win Rate: 62.5% │
├────────────────────┴────────────────────┴────────────────────┤
│ ACTIVITY LOG │
│ 14:32:05 SIGNAL BTC LONG strength=0.72 │
│ 14:32:15 AI BTC -> BUY conf=85% $0.0012 │
│ 14:32:16 TRADE ORDER BUY BTC size=0.02 @ $69,833 │
└──────────────────────────────────────────────────────────────┘
q Quit m Settings p Pause l Log d Debug c Close All r Reset CB
- Hybrid AI + Local Analysis — Local technical indicators handle 80% of decisions for free; Claude AI handles the complex 20%
- Full Terminal UI — 6-panel live dashboard with prices, positions, indicators, signals, AI decisions, and stats
- Dynamic Coin Scanner — Automatically picks top coins by volatility, trend strength, momentum, and volume
- Risk Management — Position sizing, trailing stops, daily loss limits, max drawdown circuit breaker, profit targets
- Paper Trading — Same code path as live trading, backed by simulated execution with real market data
- Cost Optimized AI — Model tiering (Haiku/Sonnet), similarity-hash caching, summarized prompts (~$0.20-0.60/day)
- Settings Screen — Change mode, credentials, risk params, and AI config without restarting
- Backtesting — Validate strategies on historical data before risking capital
# Install uv (package manager)
curl -LsSf https://astral.sh/uv/install.sh | sh
# Setup
uv venv .venv --python 3.12
source .venv/bin/activate
uv pip install -e ".[dev,tui]"
# Launch (interactive menu)
hypertradeThat's it. The interactive menu gives you all options:
╔═══════════════════════════════════════════════════════════╗
║ ██╗ ██╗██╗ ██╗██████╗ ███████╗██████╗ ║
║ ███████║ ╚████╔╝ ██████╔╝█████╗ ██████╔╝ ║
║ ██║ ██║ ██║ ██║ ███████╗██║ ██║ ║
║ TRADE AI-Powered Crypto Futures on Hyperliquid ║
╚═══════════════════════════════════════════════════════════╝
Status: .env found | API key set | Wallet set
? What would you like to do?
❯ Demo Mode — Simulated data, no keys needed
Paper Trading — Real prices, simulated trades
Live Trading — Real orders on exchange
──────────────
Settings — Configure keys, risk, AI
Check Connection — Test exchange connectivity
Fetch Data — Download historical candles
Run Backtest — Test strategy on history
──────────────
Quit
Or use direct commands:
hypertrade demo # Skip menu, straight to demo
hypertrade check # Test connection
hypertrade fetch # Download historical data
hypertrade backtest # Run backtestWebSocket ──> DataFeed ──> IndicatorEngine ──> SignalGenerator
│
┌──────────┴──────────┐
Local Decision (80%) Claude AI (20%)
└──────────┬──────────┘
Strategy
│
RiskManager ──> OrderManager ──> Hyperliquid
│
PositionTracker / StopLossEngine / CircuitBreaker
Key principle: Pre-compute all indicators locally, send only a 200-token summary to Claude (not 15,000 tokens of raw candles). This cuts AI costs by 75x.
src/
├── core/ Config, event bus, orchestrator (main loop)
├── exchange/ Hyperliquid SDK wrapper, WebSocket data feed, paper exchange
├── analysis/ Technical indicators (pandas-ta), signal generator, coin scanner
├── ai/ Claude API client, prompt builder, response parser, cache
├── strategy/ Trading strategies (momentum, mean reversion)
├── risk/ Position sizing, stop-loss engine, circuit breaker
├── execution/ Order manager, position tracker
├── monitoring/ Structured logging, metrics, alerts
└── tui/ Terminal UI (Textual), settings screen, demo orchestrator
backtest/ Historical data loader, backtesting.py integration
config/ YAML configuration files
scripts/ Connection checker, historical data fetcher
| Mode | Testnet | What Happens |
|---|---|---|
paper |
true |
Simulated trades, testnet data. Start here. |
paper |
false |
Simulated trades, real mainnet prices |
live |
true |
Real orders on Hyperliquid testnet (free test money) |
live |
false |
Real money on mainnet (prompts for confirmation) |
| Key | Action |
|---|---|
q |
Quit |
m |
Open Settings (mode, credentials, risk, AI config) |
p |
Pause / Resume trading |
l |
Focus activity log |
d |
Toggle debug mode |
c |
Close all positions + auto-pause |
r |
Reset circuit breaker |
All settings configurable via:
- YAML file (
config/default.yaml) - Environment variables (
.env) - TUI Settings screen (press
m)
Key settings:
exchange:
mode: paper # paper | live
testnet: true # true | false
trading:
coins: [BTC, ETH] # Coins to trade (always kept)
auto_select_coins: false # Auto-pick top coins by opportunity
auto_select_top_n: 5 # How many to pick
default_leverage: 3
risk:
max_daily_loss_pct: 0.05 # Circuit breaker: 5% daily loss
max_drawdown_pct: 0.15 # Circuit breaker: 15% drawdown
profit_target_pct: 0.0 # 0 = disabled, 0.10 = stop at +10%
default_stop_loss_pct: 0.02 # 2% per-trade stop loss
default_take_profit_pct: 0.04 # 4% per-trade take profit
claude:
routine_interval_seconds: 900 # AI analysis every 15 min
daily_cost_cap_usd: 50.0 # Hard daily spending limit| Technique | Effect |
|---|---|
| Two-tier models (Haiku routine, Sonnet major) | 90% of calls at 10-20x lower cost |
| Pre-computed indicator summaries (~200 tokens) | 75x fewer input tokens vs raw candles |
| Similarity-hash response cache (30-min TTL) | ~30% fewer API calls |
| Local rule-based decisions for 80% of trades | 80% of decisions are free |
| Hard daily cost cap with local-only fallback | Prevents runaway costs |
Projected cost: ~$0.20-0.60/day
- Python 3.12+
- hyperliquid-python-sdk — Exchange API
- pandas-ta — 150+ technical indicators
- anthropic — Claude AI API
- textual — Terminal UI framework
- pydantic — Configuration validation
See docs/GUIDE.md for the full user guide covering:
- Installation & setup
- Wallet setup & testnet instructions
- All CLI commands with examples
- TUI dashboard & settings screen
- Dynamic coin scanner
- Profit target
- Configuration reference
- Backtesting
- Troubleshooting
This software is for educational and research purposes. Algorithmic trading carries significant financial risk. Past performance does not guarantee future results. Always start with paper trading and minimal capital. Never trade with money you cannot afford to lose.
MIT