vibetrading 0.1.5

Agent-first trading framework: describe strategies in natural language, generate executable code, backtest and deploy across exchanges.

VibeTrading

Agent-first trading framework for cryptocurrency. Describe strategies in natural language, generate executable Python code, backtest on historical data, and iteratively improve with LLM-powered analysis.

Installation

pip install vibetrading

Quick Start

Setup

Set your LLM API key (at least one) and optional proxy:

export OPENAI_API_KEY=sk-...        # or ANTHROPIC_API_KEY, GEMINI_API_KEY, DEEPSEEK_API_KEY
export HTTPS_PROXY=http://127.0.0.1:7890   # optional

Or put them in a .env file — the package loads it automatically.

Generate a Strategy

import vibetrading.strategy

code = vibetrading.strategy.generate(
    "BTC momentum strategy: RSI(14) oversold entry, SMA crossover confirmation, "
    "3x leverage, 10% position size, 8% take-profit, 4% stop-loss",
    model="gpt-4o",
)

Backtest

import vibetrading.backtest
import vibetrading.tools

data = vibetrading.tools.download_data(["BTC"], exchange="binance", interval="1h")

results = vibetrading.backtest.run(code, interval="1h", data=data)

metrics = results["metrics"]
print(f"Return: {metrics['total_return']:.2%}")
print(f"Sharpe: {metrics['sharpe_ratio']:.2f}")
print(f"Max Drawdown: {metrics['max_drawdown']:.2%}")
print(f"Win Rate: {metrics['win_rate']:.2%}")

Analyze Results

Use an LLM to score performance and get actionable improvement suggestions:

report = vibetrading.strategy.analyze(results, strategy_code=code, model="gpt-4o")

print(f"Score: {report.score}/10")
print(report.summary)
for s in report.suggestions:
    print(f"  → {s}")

Evolve

One call to iteratively improve a strategy through generate → backtest → analyze feedback loops:

import vibetrading

result = vibetrading.evolve(
    "BTC momentum strategy with RSI and SMA crossover, 3x leverage",
    iterations=3,
    model="gpt-4o",
    interval="1h",
)

print(f"Best score: {result.best_score}/10")
print(f"Improved: {result.improved}")
print(result.best_code)

How It Works

Describe  ──▶  Generate  ──▶  Backtest  ──▶  Analyze  ──▶  Evolve
(prompt)       (LLM)          (engine)       (LLM)         (loop)
                  ▲                                           │
                  └──────────── feedback ──────────────────────┘

1. Describe — Write what you want in plain English.
2. Generate — An LLM produces framework-compatible strategy code with risk management.
3. Backtest — Run against historical data from any CCXT-supported exchange.
4. Analyze — An LLM evaluates backtest results: scores performance, finds weaknesses, suggests fixes.
5. Evolve — Repeat the loop. Each iteration feeds analysis back to the generator.

Features

• Any LLM — Works with OpenAI, Anthropic, Google, DeepSeek, or any OpenAI-compatible API via litellm.
• Built-in validator — Static analysis catches common errors before execution.
• LLM backtest analysis — Structured scoring (1-10), strengths/weaknesses, and actionable suggestions.
• Strategy evolution — vibetrading.evolve() runs the full generate → backtest → analyze → regenerate loop.
• CCXT data — Download and cache OHLCV data from Binance, Bybit, OKX, and 100+ exchanges.
• Realistic simulation — Limit/market orders, margin, leverage, funding rates, fees, and liquidation detection.

Modules

Module	Purpose
vibetrading	vibe decorator, evolve()
vibetrading.strategy	generate(), validate(), analyze(), prompt templates
vibetrading.backtest	BacktestEngine, run()
vibetrading.evolution	StrategyEvolver, evolve()
vibetrading.tools	download_data(), load_csv()

License

MIT