What VibeTrading Does

The core engine operates as an autonomous loop spanning strategy formulation to iterative refinement:

1. Natural Language Strategy Generation: Users provide trading ideas via plain English prompts (e.g., detailing asset, indicator conditions, position sizing, leverage, take-profit targets, and stop-loss parameters). An integrated LLM translates this specification into framework-compatible Python code.
2. Code Validation and Static Analysis: The platform validates generated code using static analysis tools before execution. It audits syntax, checks for necessary imports, verifies proper stop-loss/take-profit risk controls, and flags excessive leverage parameters.
3. Realistic Backtesting Engine: The framework executes the generated strategy against historical candle (OHLCV) market data. The simulation accounts for realistic execution variables, including slippage in basis points, transaction fees, funding rates for perpetual contracts, margin handling, and liquidation events.
4. AI-Driven Performance Analysis: Post-simulation, an LLM evaluates the backtest outputs (returns, drawdowns, Sharpe ratio, win/loss streaks). It assigns an objective performance score from 1 to 10, identifies specific logical weaknesses or tail risks, and provides concrete suggestions for parameter adjustments.
5. Autonomous Strategy Evolution: Through its evolve() interface, the framework runs a closed-loop sequence: Generate -> Backtest -> Analyze -> Mutate Code -> Re-Backtest. This allows strategies to self-refine over multiple iterations against historical benchmarks without requiring manual human rewriting.

Products, Modules, and Services Offered

While VibeTrading is primarily distributed as an open-source library and developer ecosystem rather than a hosted retail SaaS product, it provides a comprehensive suite of software modules, developer tooling, and exchange interfaces.

1. The Core Python Framework (vibetrading)

Available via standard package managers (pip install vibetrading), the software is divided into modular components: * vibetrading.strategy: Handles prompt-based strategy generation, static code validation, and LLM-powered performance evaluations. It supports multi-provider model configurations including Anthropic Claude, OpenAI, Google Gemini, DeepSeek, and locally hosted LLMs. * vibetrading.backtest: The core simulation runtime (BacktestEngine) providing execution logic for market and limit orders, margin requirements, fee tracking, and portfolio rebalancing. * vibetrading.tools: Built-in data ingest modules capable of fetching and caching historical OHLCV data across dozens of exchanges via CCXT and dedicated market data endpoints. * vibetrading.templates: Pre-built, battle-tested algorithmic templates (such as RSI momentum, Bollinger Band mean reversion, grid trading, and dollar-cost averaging / DCA) that run deterministically without requiring an LLM call.

2. Strategy Analysis and Reporting

• Comprehensive Metrics Engine: Calculates traditional quantitative finance metrics, including Compound Annual Growth Rate (CAGR), Sharpe ratio, Sortino ratio, Calmar ratio, maximum drawdown, drawdown duration, profit factor, expectancy, and fee overhead.
• LLM Performance Diagnostic: Generates structured written summaries explaining why a strategy performed well or poorly across specific market regimes (e.g., bull trends vs. choppy consolidation).

3. Execution Interfaces and Live Trading Engine

• Unified Sandboxing (VibeSandboxBase): Ensures that strategy code developed for backtesting uses identical function signatures and logic when deployed into simulated (paper) or live environments, eliminating translation errors between test and production code.
• LiveRunner: A modular executor for periodic or continuous strategy execution, order routing, and exchange heartbeat monitoring.
• Exchange Adapters: Connectors tailored for centralized crypto platforms as well as high-performance decentralized derivative protocols (such as Hyperliquid, Paradex, and Lighter), alongside standard CCXT exchange connectivity (Binance, Bybit, OKX).

4. Developer Tools and CLI

• Command-Line Interface (CLI): Provides terminal commands allowing operators to generate code, run backtests, download market data, and inspect evaluation reports directly from standard shell environments.
• Type Safety and Testing: Fully typed code base (PEP 561 compliance) backed by an extensive automated test suite covering order execution mechanics, margin calculations, and API adapters.

Technical Workflow Example

A standard interaction with the VibeTrading framework follows this flow:

1. Strategy Prompt: The operator submits a request: "BTC momentum strategy with RSI 14 oversold entry, SMA crossover exit, 3x leverage, 8 percent take-profit, 4 percent stop-loss."
2. Generation and Backtesting: vibetrading.strategy.generate() builds the Python strategy class using the chosen model. vibetrading.backtest.run() executes the generated logic across downloaded 1-hour candle data.
3. Automated Evaluation: vibetrading.strategy.analyze() ingests the resulting equity curve, trades list, and drawdown metrics, returning a diagnostic scorecard.
4. Iteration: The user or an automated loop passes the scorecard back to the prompt pipeline to adjust indicators, stop thresholds, or entry filters until target risk-adjusted return hurdles are met.