Vibe-Trading is a Python-based personal trading agent framework that combines large language models with quantitative finance capabilities to enable algorithmic trading, backtesting, and portfolio management. Built with FastAPI on the backend and React 19 for the frontend, the project is distributed via PyPI as vibe-trading-ai and operates under an MIT license.
The core functionality centers on empowering AI agents with comprehensive trading capabilities through a single command. The system integrates multiple market data providers including Longbridge, Tushare, Alpaca, and Binance, with fallback mechanisms to ensure data availability across different market conditions and geographies. The backtesting engine has undergone significant correctness improvements, including causal rebalancing, order-independent execution, terminal close cost charging, and strict exposure cap enforcement. Recent updates have made the backtesting sandbox more robust through AST-hardening that blocks network access, subprocess execution, eval operations, and unsafe file operations.
The repository demonstrates active development with frequent releases addressing both feature additions and security concerns. A comprehensive external security audit in July 2026 identified ten findings, all of which have been closed on the main branch. Security hardening measures include Docker multi-stage builds with digest-pinned images, short-lived single-use authentication tickets, hardened Docker Compose configurations with read-only root filesystems and dropped capabilities, and hash-locked dependencies. The Web UI now writes configuration to a canonical location with proper permission error reporting and legacy configuration migration support.
Key features include a Strategy Development Manager that converts academic papers and broker research into registered factors and strategies with persistent artifact storage and automated performance decay monitoring. The system supports portfolio optimization across five different algorithms with look-ahead-bias corrections. Recent additions include TAP mode for Alpaca key isolation, a Frazzini-Pedersen betting-against-beta academic factor, and support for Binance USD-M perpetual futures with explicit routing and separation of execution and mark prices.
The project maintains multilingual documentation in English, Chinese, Japanese, Korean, and Arabic. Community engagement occurs through Discord, WeChat, and Feishu groups. The repository has addressed numerous issues related to provider reliability, configuration management, and platform-specific concerns such as Windows settings persistence and POSIX-specific system calls.
The MCP server implementation now supports Streamable HTTP transport and safely handles file write operations with path argument recovery. Provider integrations have expanded to include NVIDIA NIM as a first-class LLM option with versioned compatibility headers, and Requesty as an OpenAI-compatible gateway. The correlation analysis tools accept bare ticker inputs and traverse the full data loader fallback chain, while the local data loader now honors requested intervals through OHLCV resampling.
