Vibe-Trading: HKU's Open-Source Research Agent for Markets, Not a Trading Bot

Polymarket Agents: Build AI Trading Bots for Prediction Markets • Orca: The ADE for Running Claude Code, Codex, and Cursor in Parallel

What Is Vibe-Trading? #

Vibe-Trading, built by HKUDS (the Hong Kong University Data Science lab behind several well-known open-source LLM projects), describes itself carefully: it’s a research workspace, not an autonomous trading bot you point at your savings. It turns natural-language finance questions into runnable analysis — market-data loaders, strategy generation, backtest engines, reports, exports, and persistent research memory tied together.

🔗 GitHub: https://github.com/HKUDS/Vibe-Trading 🌐 Homepage: https://vibetrading.wiki/

First pushed in April 2026 under the MIT license, Vibe-Trading reached 28,300+ GitHub stars by late July 2026, up from roughly 23,700 two weeks earlier.

Worth reading before anything else — the project’s own disclaimer: “Vibe-Trading is research and trading software. It is not investment advice, holds no funds, and runs no execution venue… This broker-trading capability is experimental and not verified by us against a real broker account — use it at your own risk. Past performance does not guarantee future results.”

Quick Example #

pip install vibe-trading-ai
vibe-trading init
vibe-trading run -p "Backtest a BTC-USDT 20/50 moving-average strategy for 2024 and summarize return and drawdown"

That single prompt drives data fetching, strategy code generation, a backtest run, and a summarized report — no separate scripts to write per market or per data provider.

Key Features #

Install #

Path A: pip (fastest) #

pip install vibe-trading-ai

Installed commands:

Path B: Docker (zero local setup) #

git clone https://github.com/HKUDS/Vibe-Trading.git
cd Vibe-Trading
cp agent/.env.example agent/.env
# edit agent/.env with your LLM provider key
docker compose up --build

Open http://localhost:8899. Persistent memory, backtest runs, and uploads survive updates via named Docker volumes (only removed by docker compose down -v).

Path C: MCP Plugin #

vibe-trading-mcp

Plugs Vibe-Trading’s tools directly into an existing MCP-compatible agent (Claude Desktop, Cursor, OpenClaw) rather than running it standalone.

LLM Providers #

Supports OpenRouter, OpenAI, Anthropic, DeepSeek, Gemini, Groq, Qwen, Zhipu, Kimi, MiniMax, NVIDIA NIM, and local Ollama (no API key required) — configured in .env.

Cross-Market Data, Free by Default #

All supported markets work without paid API keys, per the project, through an automatic fallback chain:

A Tushare token is optional for A-shares but not required.

The Shadow Account Feature #

One feature that stands out from typical “backtest a strategy” tools: Shadow Account ingests your own broker trade journal, runs behavior diagnostics on it, extracts trading rules, and compares your actual decisions against a rule-based baseline — producing exportable audit reports and generated strategy code. It’s framed around understanding your own past trading behavior, not generating new buy/sell signals.

Use Cases #

1. Turning a Research Question Into a Backtest #

Ask a plain-English question about a strategy idea and get runnable strategy code, metrics, and validation artifacts back, without hand-writing data-fetching code per market.

2. Reviewing Your Own Trading History #

Feed a broker export into Shadow Account to get a diagnostic on your actual trading patterns versus a rule-based comparison strategy.

3. Benchmarking Alpha Ideas Quickly #

Run the built-in 462-alpha zoo against your universe to get IC scores and alive/dead categorization before writing a custom factor from scratch.

4. Multi-Agent Investment Committee Simulation #

Use the separate investment/quant/crypto/risk agent teams to simulate a debate-style review of a position before committing to it.

Related Repositories #

Related Articles #

• Polymarket Agents: Build AI Trading Bots for Prediction Markets — a narrower, prediction-market-focused trading agent
• Orca: The ADE for Running Claude Code, Codex, and Cursor in Parallel — multi-agent orchestration, applied to coding rather than trading

Conclusion #

Vibe-Trading is an academically-backed (HKUDS), MIT-licensed research workspace that treats trading as a research problem first — natural-language-driven backtesting, multi-agent analyst teams, and its own trade-journal diagnostics via Shadow Account — with broker-authorized live trading positioned explicitly as an experimental, opt-in extra rather than the headline feature. Growing from roughly 23,700 to 28,300+ stars in two weeks (as of late July 2026) suggests real interest in the “research agent for markets” framing specifically.

Best for: Developers and quant-curious researchers who want to backtest and analyze trading ideas conversationally, and who take the project’s own “not investment advice, use at your own risk” disclaimer seriously before enabling any live broker connection.

GitHub: https://github.com/HKUDS/Vibe-Trading

Recommended Infrastructure for Self-Hosting #

If you want Vibe-Trading’s API server and web UI running persistently instead of on a laptop:

• DigitalOcean — $200 free credit for 60 days across 14+ global regions, a straightforward host for the Docker Compose setup.
• HTStack — Hong Kong VPS with low-latency access from mainland China — relevant if you’re pulling A-share data via mootdx/AKShare. This is the same IDC that hosts dibi8.com.

Affiliate links — they don’t cost you extra and they help keep dibi8.com running.

Last updated: 2026-07-29

References & Sources #

• Vibe-Trading
• Vibe-Trading homepage
• Polymarket Agents
• TradingAgents