What Is Vibe-Trading: One Command to Give Your AI Agent Trading Capabilities
Vibe-Trading is a personal trading agent framework by HKUDS (HKU Data Science); one command gives an AI agent full trading capability (market data / backtest / order / risk). Python 3.11+ + FastAPI + React 19, MIT license, Trendshift-listed.
Core Highlights
- Shadow Account: simulated trading, no real money, validate strategies first.
- Multi-broker support: yfinance / CCXT (crypto) / akshare (A-shares) / MT5 / Tiger / Alpaca / OKX / Shoonya / Longbridge, global market coverage.
- Backtest + alpha bench: strategy backtest + strict alpha benchmark (same-universe random control + OOS gate),
--strictmode. - memory lifecycle: persistent memory + quality scoring + Ebbinghaus decay + archive GC (optional, off by default).
- MCP / API server: usable as an MCP server for AI agents, or standalone API.
- Correlation regime timeline: FUSED market-episode markers (descriptive risk, not a signal).
How to Use
pip install vibe-trading-ai
vibe-trading # startConfigure broker + LLM provider; use natural language to have the agent run strategy backtests or Shadow Account trades.
Who It's For
- Quant researchers: multi-broker + backtest + alpha bench, one-stop strategy validation.
- AI trading experimenters: MCP server for AI agents, run trades in natural language.
- Learning AI trading: Shadow Account zero-risk simulation.
vs. TradingAgents
TradingAgents is a multi-agent decision framework (analysts / researchers / trader / risk / portfolio manager debate); Vibe-Trading is a trading-execution + backtest framework (broker access + Shadow Account + alpha bench). Use TradingAgents for decisions, Vibe-Trading for execution/backtest-they chain.