Vibe-Trading Wiki (vibetrading.wiki) serves as the official documentation hub, research laboratory, and knowledge ecosystem for Vibe-Trading. Vibe-Trading is an open-source, self-hosted quantitative finance and autonomous research agent environment developed by the HKUDS laboratory at the University of Hong Kong in collaboration with an open-source contributor network. While it serves global financial markets including the United States, it functions primarily as a software framework and research portal rather than a commercial financial brokerage.
The platform is designed around conversational quantitative development, translating natural-language instructions into end-to-end investment research. Instead of requiring users to manually code API pipelines, pandas dataframes, statistical checks, and backtesting harnesses, Vibe-Trading handles the research pipeline from live and historical data fetching through strategy code synthesis, multi-asset backtesting, factor validation, and risk reporting. The framework is strictly built for research, strategy validation, and education rather than automated live order execution, maintaining an inspectable and reproducible paper trail for every step of the analytical loop.
Core Operational Model: The Research Loop
The platform operates an autonomous research workflow designed to eliminate typical hallucinations associated with large language models in financial analysis:
Routing: The central orchestrator analyzes the user query in natural language, determining the necessary domain skills, data feeds, quantitative tools, and sub-agent teams required to answer it.
Grounding: The agent pulls verified market context, SEC filings, financial news, macro indicators, or trade logs into an immutable data ledger so that every calculation is traceable to a specific timestamped source.
Execution and Testing: The system selects one of multiple market backtesting engines to run historical simulations, calculate risk metrics, or audit alpha signals against historical datasets.
Delivery: The framework exports runnable artifacts, including HTML summary cards, performance visualizations, diagnostics, and ready-to-run script files for third-party platforms.
Products, Features, and Capabilities
1. Agent Harness and CLI / Desktop Interface
Interactive Research Terminal: Distributed via standard Python package distribution (pip install vibe-trading-ai), allowing researchers to launch interactive workflows using simple commands.
Electron Desktop Shell & Web UI: Provides a graphical user interface consisting of a FastAPI backend and a React/Electron frontend for interactive charting, session histories, and visual backtest inspections.
Persistent Memory & Session Search: Saves user workspaces, backtest configurations, custom factor definitions, and research discussions across sessions with automated context compression.
Custom Skills & Tools Library: Ships with dozens of built-in deterministic calculation tools for quantitative pricing (Black-Scholes, implied volatility inversion), econometrics (GARCH modeling, stationarity, cointegration), and fixed-income mathematics.
2. Multi-Engine Backtesting Framework
The system contains specialized backtesting engines designed around the market microstructure, trading hours, and clearing rules of specific asset classes and regions:
* Global Equities Engine: Simulates standard execution for United States, Hong Kong, and international equities with realistic commissions, slippage models, and order execution constraints.
* Regional Equity Engines: Models specific international execution quirks, such as China A-shares (T+1 rules, price limits, ST filtering), India equities (circuit bands, STT, and stamp tax structures), Korea equities, and Vietnam equities.
* Cryptocurrency Engine: Models 24/7 continuous trading, spot execution, and perpetual futures funding settlements across regular intervals.
* Derivatives & Multi-Leg Options: Evaluates options portfolio payoffs, Greek profiles (Delta, Gamma, Vega, Theta), and multi-leg spreads.
* Composite Engine: Runs multi-asset portfolios simultaneously from a unified capital allocation base.
3. Alpha Factor Library ("Alpha Zoo")
Vibe-Trading packages over 450 standardized quantitative factor formulas drawn from institutional quantitative finance literature and research desks:
* Kakushadze 101 Alphas: Standard implementations of the cross-sectional equity alpha signals formulated by Zura Kakushadze.
* Guotai Junan (GTJA) 191 Alphas: Complete algorithmic adaptations of the short- and medium-horizon price-volume indicators widely referenced in Asian equity research.
* Academic Factors & Qlib Standards: Factor sets evaluating cross-sectional momentum, liquidity, short-horizon volatility, and reversion patterns.
* Automated Factor Benchmarking: Provides benchmark utilities to measure information coefficient (IC), decay horizon, and turnover directly from user prompts.
4. Agent Swarm Teams
Rather than relying on a single prompt response, the platform deploys collaborative multi-agent teams orchestrated to challenge and debate research assumptions:
* Investment Committee Preset: Deploys dedicated "Bullish" and "Bearish" agents to challenge an investment thesis, an independent Risk Manager agent to scrutinize drawdown risk, and a Portfolio Manager agent to deliver the synthesis.
* Quantitative Strategy Desk: Coordinates factor extraction, backtesting execution, and code safety auditing between distinct agents.
* Macro & Crypto Desks: Evaluates interest rate regimes, central bank guidance, commodity flows, or crypto funding rates.
5. Shadow Account Diagnostics
Broker Log Ingestion: Allows traders to upload CSV exports and transaction logs from brokerage accounts.
Behavioral Analysis: Reverse-engineers underlying user trading behavior, identifying common drawdown triggers, risk-reward asymmetries, holding-time inefficiencies, and fee drag.
Strategy Distillation: Synthesizes the trader's historical actions into an explicit, backtestable ruleset to evaluate what performance would have looked like under strict systematic rules.
6. Script Export and Third-Party Bridges
Pine Script & MQL5 Generator: Transpiles tested strategies into clean TradingView Pine Script (v6), MetaTrader 5 (MQL5), or cTrader C# code.
Pineify & Model Context Protocol (MCP) Integration: Interfaces with external AI clients and MCP endpoints to supply read-only market data and automated syntax verification before the trader deploys code to their charting platform.
Please note: This information is generated by AI. For legal use, verify with company sources.
