Vibe-Trading - AI Tool for Natural-Language Trading Strategy Generation, Backtesting, and Execution

An AI-based multi-agent platform that automatically generates trading strategy code from natural language input alone, tests it, and exports it.

• Supports 5 data sources, handling China A-shares, Hong Kong/US stocks, crypto, futures, and forex with automatic fallback and no configuration.
• Organizes 71 specialized skills into 7 categories, and supports self-evolving skills where agents create and improve their own workflows.
• Provides 29 preconfigured swarm team presets for DAG-based multi-agent orchestration such as investment committee debates, quant strategy desks, crypto trading desks, and global macro portfolio allocation.
• Features 7 backtesting engines plus a cross-market composite engine, with statistical validation via Monte Carlo, Bootstrap CI, and Walk-Forward, and 4 optimizers.
• Remembers user preferences and insights through cross-session memory and FTS5-based search of past conversations.
• Offers one-click export to TradingView (Pine Script), Tongdaxin/Dongfangcaifu (TDX), and MetaTrader 5.
• Provides 17 MCP tools that can connect to MCP-compatible clients like Claude Desktop, Cursor, and OpenClaw, with 16 tools working without an API key.
• Supports Docker one-command deployment, CLI/TUI, a FastAPI web UI, and installation via pip install vibe-trading-ai.
• Also includes quant analysis tools such as Factor IC/IR analysis, Black-Scholes pricing, Greeks calculation, technical pattern recognition, and portfolio optimization.

The project is for research, simulation, and backtesting only and does not support live trading. It is Python-based and MIT licensed.