autotrade

Autonomous trading strategy research using LLM-driven backtesting.

Setup

# Install dependencies
uv sync

# Configure environment
cp .env.example .env
# Edit .env with your settings (exchange, symbol, dates, etc.)

# Run a backtest
uv run python -m src.train

Autonomous Strategy Iteration

Use an LLM CLI to autonomously iterate on trading strategies. The LLM reads src/program.md and follows the experiment loop.

Claude Code

# Autonomous mode (skip all permission prompts)
claude -p "Read src/program.md and follow the experiment loop autonomously." \
  --dangerously-skip-permissions

# Safer: only allow specific tools
claude -p "Read src/program.md and follow the experiment loop." \
  --allowedTools "Read,Write,Edit,Bash,Glob,Grep"

# Limit iterations for testing
claude -p "Read src/program.md and run 5 strategy experiments." \
  --dangerously-skip-permissions \
  --max-turns 50

# Background execution
nohup claude -p "Read src/program.md and follow the experiment loop." \
  --dangerously-skip-permissions \
  > claude_session.log 2>&1 &

Codex (OpenAI)

# Autonomous mode (non-interactive)
codex exec "Read src/program.md and follow the experiment loop autonomously." \
  --full-auto

# Interactive with auto-approval
codex "Read src/program.md and follow the experiment loop." \
  --full-auto

# Fully autonomous (dangerous - no sandbox)
codex exec "Read src/program.md and follow the experiment loop." \
  --dangerously-bypass-approvals-and-sandbox

# Background execution
nohup codex exec "Read src/program.md and follow the experiment loop." \
  --full-auto \
  > codex_session.log 2>&1 &

Gemini CLI

# Autonomous mode
gemini -p "Read src/program.md and follow the experiment loop autonomously." \
  --sandbox none

# With tool restrictions
gemini -p "Read src/program.md and follow the experiment loop." \
  --tools "read_file,write_file,run_command"

# Background execution
nohup gemini -p "Read src/program.md and follow the experiment loop." \
  --sandbox none \
  > gemini_session.log 2>&1 &

Key Files

File	Description
src/program.md	LLM instructions for strategy iteration
src/train.py	Backtest runner (set STRATEGY_FILE, TIMEFRAME, MAX_LEVERAGE)
src/prepare.py	Data loading and settings (read from .env)
src/strategies/generated/	Generated strategy files
src/strategies/base.py	BaseStrategy base class
src/results.jsonl	Experiment log (JSON Lines)
.trash/strategies/	Discarded strategies

Configuration

See .env.example for all options:

TOTAL_CASH=100000          # Initial capital
MARKET_TYPE=spot           # spot or swap
EXCHANGE_ID=binance        # Exchange for data
SYMBOL=BTC/USDT            # Trading pair
START_DATE=2023-01-01      # Backtest start
END_DATE=                  # Backtest end (blank = today)
MAX_DRAWDOWN_LIMIT=-30     # Target max drawdown
MIN_PROFIT_FACTOR=1.0      # Target profit factor