Vibe-Trading - HKUDS Open Source AI Multi-Agent Financial Workspace - AiBoss

What is Vibe-Trading?

Vibe-Trading is an open-source AI-driven multi-agent financial workspace from the Data Science Lab (HKUDS) at the University of Hong Kong. It transforms natural language instructions into executable trading strategies, research insights, and portfolio analysis. The project covers global markets including A-shares, Hong Kong/US stocks, cryptocurrencies, futures, and forex. It incorporates six data sources and an automatic degradation mechanism, allowing most basic data to be used without an API key. The system features 74 financial professional skills, 29 pre-configured DAGs of expert agents, and seven backtesting engines (including cross-market portfolio backtesting). It supports one-click export to TradingView, Tongdaxin/Tonghuashun, and MetaTrader 5. The project offers four entry points: one-click PyPI installation, zero-configuration Docker deployment, MCP plugin integration, and one-click ClawHub installation. It is compatible with over 13 LLM providers such as OpenRouter, DeepSeek, Kimi, and Ollam, and is fully open-source under the MIT license.

Vibe-Trading's main functions

Executable strategies for natural language generationIt describes trading ideas in plain language, and AI automatically writes, tests, and exports trading code, covering strategy types such as technical indicators, multi-factor, and machine learning.
74 Financial Professional Skills DatabaseIt is divided into 8 major categories (data sources, strategies, analysis, asset classes, cryptocurrencies, fund flows, tools, and risk analysis), covering the entire chain of capabilities from data acquisition to quantitative research.
A team of 29 expert agents (Swarm): A multi-agent cluster with pre-set DAG arrangement, such as the investment committee's long/short debate, the global equity desk, the cryptocurrency trading desk, and the quantitative strategy desk, to simulate the collaborative process of real institutions.
7 backtesting enginesSupports backtesting of A-shares, Hong Kong/US stocks, cryptocurrencies, futures, foreign exchange, and cross-market portfolios, and integrates statistical tests such as Monte Carlo simulation, Bootstrap confidence intervals, and rolling forward validation.
One-click export of multi-platform strategiesThe generated strategies can be directly compiled into TradingView Pine Script v6, Tongdaxin/Tonghuashun formulas, and MetaTrader 5 MQL5 code, seamlessly integrating with live trading platforms.
Cross-session memory and skill evolutionAI-powered persistent memory of user preferences and historical insights, supporting FTS5 conversational search, and the ability to autonomously create and optimize reusable personal financial skills.
Zero-configuration coverage of 6 major data sourcesBuilt-in Tushare, yfinance, OKX, AKShare, CCXT, and Futu, automatically selects the best source and intelligently downgrades, and can be used in most markets without an API key.
Transaction logs and shadow account analysisUpload files exported from brokerage firms (such as Tonghuashun, Eastmoney, and Futu), automatically analyze trading profiles and behavioral deviations, and support shadow backtesting to generate an 8-chapter comparison report.
General document intelligent readingIt unifies the parsing of PDF, Word, Excel, PPT, image OCR, and 40+ text formats, and supports one-click upload and analysis of financial reports, research reports, and transaction records.
Real-time streaming interactive DashboardThe Web UI provides real-time state streams for the agent, visualization of backtesting results, interactive charts, and ECharts heatmaps. The front-end supports lazy loading optimization.
Seamless integration into the MCP ecosystemIt exposes 22 MCP tools (21 of which are keyless), which can be directly integrated into editors such as Claude Desktop, Cursor, and Windsurf, embedding financial capabilities into existing workflows.
Quantitative Analysis ToolkitBuilt-in professional modules include factor IC/IR analysis, Black-Scholes option pricing and Greek value calculation, technical pattern recognition, MVO/risk parity/BL portfolio optimization, etc.

How to use Vibe-Trading

Method 1: One-click installation of PyPI (Recommended)
- implement pip install vibe-trading-ai
- run vibe-trading init Interactive configuration of environment variables
- run vibe-trading Start the interactive TUI, or vibe-trading serve --port 8899 Start Web service
- run vibe-trading-mcp Start the MCP service to connect to the external agent
Method 2: Zero-configuration Docker deployment
- git clone After the warehouse, copy agent/.env.example arrive agent/.env And enter the LLM API Key
- implement docker compose up --build
- Open in browser http://localhost:8899 Ready to use
Method 3: Integrating the MCP plugin with the existing Agent
- Add to Claude Desktop configuration vibe-trading-mcp Order
- Alternatively, configure stdio/SSE transport in OpenClaw, Cursor, or Windsurf.
- 21/22 tools can be called without any API key.
Method 4: One-click installation via ClawHub
- implement npx clawhub@latest install vibe-trading --force
- Automatically downloads skill packs and MCP configurations to the Agent directory, eliminating the need for manual cloning.

Vibe-Trading's official website address

Github repository: https://github.com/HKUDS/Vibe-Trading

Key information and usage requirements of Vibe-Trading

Python version requirements: 3.11+.
PyPI package name:vibe-trading-ai(The command after installation is) vibe-trading,vibe-trading serve,vibe-trading-mcp).
LLM configuration: needs to be in agent/.env Medium configuration LANGCHAIN_PROVIDER And the corresponding API Key; supports running Ollam locally without a key.
Recommendation ModelFor complex scenes, we recommend Claude Opus/GPT-5.4/Gemini 3.1 Pro; for everyday use, we recommend cost-effective models such as DeepSeek-V3.2, Grok-4.20, and Kimi K2.5.
Data sourceTushare Token is optional (AKShare offers free downgrade); Yfinance (Hong Kong/US stocks), OKX (cryptocurrency), and CCXT (100+ exchanges) all offer key-free options.
Web UI PortFront-end development mode 5899(Agent to backend), production mode single service 8899.
File upload restrictionsWeb UI streaming upload, single file size limit of 1MB, to prevent memory overflow.
Open source licenseMIT License.
Latest versionv0.1.6 (Released on 2026-04-28).

Vibe-Trading's core advantages

Zero-threshold global market coverageIt integrates six major markets: A-shares, Hong Kong stocks, US stocks, cryptocurrencies, futures, and foreign exchange, with most basic data provided free of charge and requiring no key.
Intelligent agent cluster collaborationIt's not a single-agent Q&A session, but rather a collaborative process involving 29 pre-set teams arranged according to a Directed Acyclic Graph (DAG) to simulate real investment research, trading, and risk control processes.
The strategy can be implemented.The generated strategy is not demo code and can be directly compiled into live trading platforms such as TradingView, TongdaXin, and MT5.
Continuous Evolution Across SessionsIt has a persistent memory layer, so each conversation trains your personal trading assistant instead of repeatedly starting from scratch.
Statistical rigorBacktesting results include Monte Carlo simulation, Bootstrap confidence intervals, and rolling forward validation to avoid overfitting traps.
Seamless integration into the MCP ecosystem22 MCP tools (21 of which are keyless) can be directly integrated with editors such as Claude Desktop, Cursor, and Windsurf.
Open source and scalableMIT license, Python 3.11+, supports local Ollam running at zero cost, and can also be integrated with commercial APIs such as OpenRouter.

Comparison of Vibe-Trading's similar products

Comparison Dimensions	Vibe-Trading (HKUDS)	Composer.trade	TradingAgents (TauricResearch)
Product Positioning	AI Multi-Agent Financial Workspace: Natural Language Processing → Executable Policies + Cross-Market Backtesting + Multi-Platform Export	AI-native no-code investment platform, using natural language to generate automated trading strategies (Symphony), vertically integrating the trading ecosystem.	A multi-agent LLM financial trading framework that simulates the collaborative decision-making process of a real trading firm (analyst → researcher → trader → risk control).
Open source license	MIT is completely open sourceIt can be deployed locally and further developed.	Closed-source commercial productsThe core strategy logic is not open.	Open source academic research frameworkBuilt on LangGraph, the code is scalable.
Intelligent agent architecture	29 preset DAGs choreographed by the Swarm team(Investment Committee, Global Equity Desk, Cryptocurrency Desk, etc.), supports streaming status monitoring.	Single-agent policy generation + community policy cloning, without the concept of multi-agent collaboration.	4-layer multi-agentAnalyst team (fundamentals/sentiment/news/technical analysis) → Research team (bullish/bearish debate) → Traders → Risk control/PM final decision-making
Natural Language Interaction	Supports AI automatically writing code, backtesting, and exporting to third-party platforms after you describe your thought process.	Supports converting natural language into executable, automated trading strategies within 60 seconds.	Supported. By inputting the ticker and date via CLI, the agent automatically completes the analysis and outputs a trading decision.
Market Coverage	A-shares, Hong Kong stocks, US stocks, cryptocurrencies, futures, foreign exchangeSix major markets, supporting cross-market portfolio backtesting	US stocks, ETFs, cryptocurrencies, optionsUS market only	byUS stocks(Data obtained via yfinance/Alpha Vantage)
Data source and configuration	6 major data sources(Tushare, yfinance, OKX, AKShare, CCXT, Futu),Automatic downgradeMost markets do not require an API Key	Owns its own data infrastructure; trade execution is integrated through Alpaca brokerage.	yfinance, Alpha Vantage, FinnHub; local Tauric TradingDB (under development)
Backtesting capability	7 backtesting engines(Includes a cross-market portfolio engine), integrating Monte Carlo, Bootstrap confidence intervals, rolling forward validation, and four optimizers.	Subsecond-level backtesting, historical performance validation, and AI overfit detection.	Supports backtesting, based on a self-built TradingDB dataset (not yet fully released).
Strategy Export and Live Trading	One-click export to multiple platformsTradingView Pine v6, Tongdaxin/Tonghuashun, MetaTrader 5 MQL5; does not directly connect to brokerage firms.	Automatic execution within the platformTrading and rebalancing can be performed directly through your own brokerage account; exporting to external platforms is not supported.	Connect onlySimulated ExchangeExecution is not supported; real brokerage API live trading is not supported.
MCP plugin support	22 MCP tools(21 keys without a key), compatible with editors such as Claude Desktop, Cursor, and Windsurf.	Officially provided composer-trade-mcpSupports Claude/Cursor for policy creation and backtesting.	No native MCP support
Cross-session memory	Persistent memory + FTS5 search + self-evolution skills(The agent can create and optimize reusable skills across sessions)	No cross-session memory; each policy creation is a separate task.	No cross-session memory; state is not retained after a single analysis process ends.
Quantitative professional tools	Factor IC/IR analysis, Black-Scholes Greeks value calculation, technical pattern identification, MVO/risk parity/BL portfolio optimization	Basic technical indicators (RSI/MA/EMA, etc.), AI-assisted parameter tuning	Technical indicators (MACD/RSI), social media sentiment scores, news macro analysis, and fundamental financial indicators.
Community and Ecology	ClawHub offers one-click installation, OpenSpace provides self-evolving and shared skills, and offers 74 scalable financial skills.	3000+ community strategy marketIt can clone and adapt other people's strategies.	The Tauric Research community is research-oriented and encourages academic citations.
Pricing Model	Completely free(Only requires a self-provided LLM API Key, supporting local zero-cost operation of Ollam)	Freemium: Free (backtesting only); $30/month (1 live strategy); $120/month Pro (unlimited strategies)	Free and open source(You need to provide your own LLM and data API Key)
Target audience	Individual quantitative traders, A-share/Hong Kong stock/cryptocurrency investors, AI agent developers, and finance students.	US retail investors, ordinary users with no programming experience who want to automate their trading.	Academic researchers, AI finance experimenters, and US stock traders hoping to simulate institutional decision-making processes.

Application scenarios of Vibe-Trading

Cryptocurrency monitoring 24/7This analysis of perpetual contract funding rates and basis, liquidation heatmaps, stablecoin fund flows, DeFi yields, and on-chain whale flows provides multi-dimensional data support for crypto asset trading.
Intelligent analysis of financial reports and research reportsUpload PDF financial statements, Word research reports, Excel financial models, or PPT roadshow materials. The system will automatically extract key risk points, performance exceeding/missing items, and changes in management guidance to generate a structured investment memorandum.
AI Agent Financial Capability Plug-inVibe-Trading integrates with editors such as Claude Desktop, Cursor, and Windsurf through 22 MCP tools (21 of which are keyless), enabling general AI assistants to acquire professional financial skills such as real-time market data query, backtesting execution, and strategy generation.
Localized privacy compliance deploymentIt enables end-to-end private operation through Docker or Ollam local large model, ensuring that transaction data and strategy logic do not leave the local environment, thus meeting the stringent requirements of financial institutions or high-net-worth individuals for data sovereignty and compliance.
A-share risk pre-screeningThe ashare-pre-st-filter feature automatically screens A-share ST/*ST risks and administrative penalty correlations, and combines Sina data sources to filter securities account-related noise, thus proactively avoiding delisting and regulatory risks.