Overview
Vibe-Trading connects natural-language prompts to market-data loaders, strategy generation, backtest engines, and persistent research memory. Features 461+ alpha factors, 16 IM channel adapters, 10 broker connectors, and cross-market support for equities, crypto, futures, and forex.
Features
- ✓Natural-language trading research
- ✓Multi-agent investment committees
- ✓461+ pre-built alpha factors
- ✓Cross-market backtesting
- ✓16 IM channel adapters
- ✓10 broker connectors
- ✓Shadow Account analysis
- ✓Live trading with safety mandate
Installation
pip install vibe-trading-aiPros
- +Natural-language interface for quantitative finance
- +Massive alpha library for rapid prototyping
- +Cross-market support covering global equities and crypto
- +Safety-first live trading with mandate limits
Cons
- −Heavy dependency on multiple Python packages
- −Complex setup for live trading
- −Not suitable for production trading without validation
Alternatives
Documentation
Vibe-Trading
Overview
Vibe-Trading is an open-source AI research workspace built for finance questions. It connects natural-language prompts directly to market-data loaders, strategy generation engines, backtest pipelines, report generation, and persistent research memory. Maintained by HKUDS (Hong Kong University of Science and Technology), Vibe-Trading has rapidly become one of the most popular AI agent projects on GitHub, crossing 21,000 stars.
Unlike traditional backtesting frameworks that require hardcoding strategy logic, Vibe-Trading lets you describe a trading idea in plain English — the agent plans, grounds itself in live or historical market data, executes backtests, validates results, and delivers a full report.
The project supports cross-market research across A-share, Hong Kong, and US equities, as well as crypto, futures, and forex. While primarily designed for research and simulation, it can optionally execute live trades through authorized brokers within a user-defined mandate and kill switch.
Features
- Self-Improving Agent: Natural-language market research with memory-backed workflows that learn from prior sessions
- Multi-Agent Teams: Investment, quant, crypto, macro, and risk committees with streaming progress and persisted reports
- Cross-Market Backtesting: A-share/HK/US equities, crypto, futures, and forex with PIT-safe data and automatic fallback
- Alpha Zoo: 461+ pre-built quant alphas including Qlib (158), Kakushadze (101), GTJA (191), academic, and fundamental factors
- Shadow Account: Parse broker journals, diagnose trading behavior, extract rules, and compare against a rule-based shadow strategy
- 16 IM Channel Adapters: WebSocket, Telegram, Slack, Discord, Matrix, WhatsApp, Signal, QQ, WeChat, Feishu, DingTalk, Teams, email, and Mochat
- 10 Broker Connectors: IBKR, Robinhood, Tiger, Longbridge, Alpaca, OKX, Binance, Futu, Dhan, and Shoonya
- 19 Free Data Sources: Tencent, mootdx, eastmoney, baostock, akshare, tushare, yahoo, sina, stooq, yfinance, finnhub, alphavantage, tiingo, fmp, qveris, okx, ccxt, futu, and more
- Export Formats: TradingView Pine Script v6, TDX (通达信), MetaTrader 5 (MQL5), and PDF/HTML reports
Installation
# Install via pip
pip install vibe-trading-ai
# Or clone and install from source
git clone https://github.com/HKUDS/Vibe-Trading.git
cd Vibe-Trading
pip install -e .
Docker support is available via docker-compose.yml with multi-stage builds and named volumes for data persistence.
Quick Start
# Natural-language backtest
vibe-trading run -p "Backtest a BTC-USDT 20/50 moving-average strategy for 2024, summarize return and drawdown, then export the report"
# Benchmark a pre-built alpha zoo
vibe-trading alpha bench --zoo gtja191 --universe csi300 --period 2018-2025 --top 20
# Analyze your trading behavior (Shadow Account)
vibe-trading --upload trades_export.csv
vibe-trading run -p "Analyze my trading behavior, extract my shadow strategy, and compare it with my actual trades"
# Interactive research chat
vibe-trading chat
# Resume a past research session
vibe-trading resume <session-id>
Core Concepts
The research spine follows a consistent five-step evidence path:
- Plan — The agent selects relevant skills, tools, data sources, and swarm presets based on your prompt
- Ground — Market data is pulled through the loader registry with automatic fallback chains ordered by IP-ban risk
- Execute — Testable strategy code is generated, backtests are run, or journal data is analyzed
- Validate — Metrics, benchmark comparisons, Monte Carlo, Bootstrap, and Walk-Forward tests are applied
- Deliver — Reports, artifacts, tool traces, and exports are returned
Advanced Features
Live Trading with Commitment Mandate
Live trading operates within a user-defined mandate: symbol universe, order-size limits, exposure caps, leverage bounds, and a daily cap. A filesystem kill switch, preemptive flatten, mandate auto-expiry, and a full audit ledger provide safety guarantees. The project never trades outside user-defined limits.
Research Autopilot
Closes the loop between hypothesis generation, signal engineering, and backtesting. Metrics from backtests feed back into the hypothesis, enabling iterative improvement of trading strategies without manual intervention.
Strategy Dev Manager
Turns academic papers and broker research into registered factors and strategies with IC/Sharpe decay monitoring, making research reproducible and trackable over time.
Examples
Multi-Agent Investment Review
# Launch a multi-agent team for comprehensive equity analysis
vibe-trading run -p "Run the investment committee on NVIDIA: gather fundamentals, sentiment, and risk metrics, then deliver a buy/sell/hold recommendation"
Alpha Zoo Benchmarking
# Run top-20 alphas against CSI 300 universe
vibe-trading alpha bench --zoo gtja191 --universe csi300 --period 2018-2025 --top 20
Scheduled Background Research
# Set up a cron job to run research automatically
vibe-trading run -p "Daily BTC-USDT mean-reversion backtest and report" --schedule "0 8 * * *"
Pros
- ✅ Natural-language interface for complex quantitative finance workflows
- ✅ Massive alpha library (461+ pre-built factors) for rapid prototyping
- ✅ Cross-market support covering global equities and crypto
- ✅ Multi-agent team orchestration for collaborative research
- ✅ Safety-first live trading with mandate limits and kill switch
- ✅ Wide ecosystem of 16 IM channel adapters and 10 broker connectors
Cons
- ❌ Heavy dependency on multiple Python packages and data providers
- ❌ Requires significant setup for live trading (broker API keys, mandate configuration)
- ❌ Not suitable for real production trading without careful validation
- ❌ Chinese-market data sources may have limited availability outside mainland China
When to Use
- Quantitative research: When you need to explore trading strategies across multiple asset classes with minimal coding
- Backtesting: When you want to benchmark hundreds of alpha factors against a specific universe
- Trading behavior analysis: When you want to understand your own trading patterns through broker journal analysis
- Multi-agent financial analysis: When you need collaborative investment, quant, and risk committee reviews