Vibe-Trading: Local AI Agent for Quant Research

The Breakthrough

Vibe-Trading provides a local, agent-driven research environment that automates the entire quantitative workflow, from strategy formulation and backtesting to performance reporting, without requiring cloud-locked platforms or manual boilerplate coding.

What Actually Worked

• Alpha Zoo Integration: The tool ships with 452 verified quant factors sourced from academic papers and industry research, including Microsoft Qlib, Kakushadze 101, and GTJA 191, all validated against look-ahead bias and network-isolated to ensure backtest integrity.
• Agentic Swarms: Users can deploy preset multi-agent teams, such as an "Investment Committee" consisting of bull, bear, risk-reviewer, and portfolio-manager agents that debate and validate trading theses.
• Shadow Account Profiling: The agent ingests historical broker exports to reverse-engineer a user's actual trading habits, identifying specific behavioral biases and calculating the financial cost of those habits via backtesting.
• Flexible Deployment: Built on a FastAPI backend with a React frontend, the tool functions as a standalone terminal application, a web dashboard, or an MCP server compatible with IDEs like Cursor.

Before / After

• Workflow Efficiency: Reduces the time required to benchmark a library of factors against a stock universe from a multi-day manual coding effort to a single command-line prompt.

Context

Most open-source trading projects are either simulations that lack rigor or cloud-locked platforms that restrict user control. Vibe-Trading addresses this by providing a local-first workspace that handles the heavy lifting of data loading, factor validation, and backtesting. While it does not execute live trades, it serves as a personal quant intern for testing hypotheses before committing capital.

Content References

• tool: Microsoft Qlib, context: cited
• tool: TradingView, context: mentioned
• tool: MetaTrader 5, context: mentioned
• tool: Cursor, context: mentioned