What It Is
A quant desk on your laptop
A hedge fund pays millions for a team that researches markets, writes trading strategies, and tests them against years of history before a single dollar moves. Vibe-Trading is that team, open-source and free, running on your own machine.
You type one command and tell it what you want in plain English, something like "build me a Bitcoin strategy and backtest it on the last year." That one line kicks off a team of AI agents. One pulls live market data. One writes the actual strategy. One backtests it against real history. And one checks the math so you're not fooling yourself. Over 21,000 people have already starred it on GitHub.
Here's the honest framing, because it matters: this is a research and backtesting workspace, not a money printer. It's built to help you research ideas, simulate them, and see how they would have played out. It holds no money. It only ever trades live if you connect a broker yourself and set the limits, and you can stop it instantly. Treat the backtests as homework, not guarantees.
How It Works
One sentence, a team of agents
You don't write code. You describe what you want and the agents split the work. This is what happens behind that one line:
Pulls the data
One agent fetches live market data for whatever you're looking at, stocks, crypto, futures, forex.
Writes the strategy
Another agent turns your plain-English idea into an actual, runnable trading strategy.
Backtests it against history
A third runs the strategy against real historical data and reports what would have happened, returns, drawdown, Sharpe ratio.
Checks the math
A fourth validates the results so a good-looking backtest isn't secretly cheating (look-ahead bias, bad data, and the usual traps). This is the part that keeps you honest.
The Setup
Three ways to get it running
Pick one, you don't need all three. Path 1 is the fastest and the one most people should use. Path 2 is if you want to run it through Claude. Path 3 is if you'd rather not install anything.
What you need first
Python 3.11+ for Paths 1 and 2 (most machines have it, or grab it from python.org). Docker for Path 3. And an LLM API key from any provider it supports (OpenAI, DeepSeek, Gemini, Kimi, OpenRouter, and more), or run it fully local and free with Ollama and no key at all. Good news on data: the market data itself needs zero keys, it falls back to free sources automatically.
One-line install
fastest · ~2 min
The quickest way in. One command installs it, a second sets up your keys, and you're running research from the terminal. This is the one command the video is talking about.
Run these
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"
Plug it into Claude (MCP)
turn Claude into a trader · ~3 min
This is the move if you want to run it through Claude instead of a separate app. It starts an MCP server that plugs the whole trading desk into Claude Desktop, Claude Code, Cursor, or any agent that speaks MCP. Then you just talk to Claude and it drives the desk.
Start the MCP server
pip install vibe-trading-ai vibe-trading init vibe-trading-mcp
Docker (zero local setup)
try it now · ~2 min
Don't want to install anything on your machine? Clone it, drop in your API key, and Docker builds the whole thing (backend plus a web dashboard) in one container. Open your browser and it's there.
Run these
git clone https://github.com/HKUDS/Vibe-Trading.git cd Vibe-Trading cp agent/.env.example agent/.env # open agent/.env, uncomment your LLM provider, paste your API key docker compose up --build # then open http://localhost:8899
The Best Prompts
What to actually ask it
Once it's running, you talk to it in plain English. In Path 1 you drop these after vibe-trading run -p or type them straight into the CLI. In Path 2, you just say them to Claude. Six worth stealing:
Backtest a strategy idea
Backtest a 20/50-day moving average crossover on AAPL for the past year. Show me the Sharpe ratio and max drawdown, then export the report.
Research a stock before you buy
Research NVDA: earnings trend, analyst consensus, option flow, and the key risks for next quarter. Give me the summary in plain English.
Test a specific rule
Test an RSI(14) mean-reversion strategy on BTC-USDT: buy below 30, sell above 70, over the last 6 months. Tell me if it actually would have made money and where it broke.
Analyze your own trades
Here is my broker export [upload your trades CSV]. Analyze my trading behavior, find my biases, extract my "shadow strategy," and compare it against what I actually did.
Run a full analyst team
Analyze the current Fed rate path, USD strength, and the impact on emerging-market equities and gold. Run it as a research team and give me one combined report.
Teach it your preferences
Remember this for all my future strategies: I prefer RSI-based setups, max 10% drawdown, and a hold period of 5 to 20 days. Build the next strategy to fit that.
Read This First
The honest cautions
This is not financial advice.
A backtest that looked great is a description of the past, not a promise about the future. You make every real decision. Do your own diligence and never risk money you can't lose.
Watch out for impersonators.
The team has never launched a token or memecoin. There are fake accounts and tokens pretending to be this project. Never connect a wallet, buy a token, or sign anything claiming to be Vibe-Trading. The only thing you should get from it is code.
Start in simulation. Stay there a while.
Research and backtest first. It only trades live if you deliberately connect a broker and set the limits yourself, and you can halt it instantly. There's no reason to rush into live trading to get the value here.
Credit where it's due
Vibe-Trading was built and open-sourced by HKUDS (the Data Intelligence Lab at the University of Hong Kong). It's MIT licensed and free. Repo: github.com/HKUDS/Vibe-Trading. If it's useful to you, go star it and support the people who made it.
Work with Me
Want me on your problem for an hour?
Book a 1:1 call and we build the fix live — the workflow you want automated, the tool you can’t crack, whatever’s stuck. Direct, hands-on, no pitch waiting at the end.