Vibe-Trading
An open-source research workspace where finance questions become runnable backtests — 90 skills, agent teams, and an audit trail.
The video tools earlier in this category showed the skills-plus-agent pattern applied to one domain. Vibe-Trading is the same idea aimed at markets: ask a question in plain language, and it pulls data, writes strategy code, runs a backtest, and hands back a report you can archive. It’s from HKU’s data science lab, MIT-licensed, and holds no funds of its own.
Try it
pip install vibe-trading-ai
vibe-trading init # interactive .env setup
vibe-trading run -p "Backtest a BTC-USDT 20/50 moving-average strategy for 2024, \
summarize return and drawdown, then export the report"
Three commands come with the package:
| Command | What it starts |
|---|---|
vibe-trading | The interactive CLI |
vibe-trading serve --port 8899 | FastAPI backend and web UI |
vibe-trading-mcp | An MCP server, so your existing agent gets the tools |
Python 3.11+, or docker compose up --build for the zero-setup path. It needs an LLM
key from one of about twenty providers — or Ollama locally, no key at
all. Market data works with no keys either: Yahoo, OKX, mootdx, and AKShare cover
equities, crypto, futures, and forex through an automatic fallback chain.
What’s actually in the box
- 90 skills across 9 categories — the finance equivalent of the specialist subagents earlier in this category. They’re editable Markdown, and yours override the bundled ones by filename.
- Agent teams — investment, quant, crypto, macro, and risk workers that research in parallel and persist their reports.
- An alpha zoo of 462 factors (Qlib, Kakushadze, GTJA-191, academic, PIT-safe fundamentals), benchable in one line:
vibe-trading alpha bench --zoo gtja191 --universe csi300 --period 2018-2025 --top 20
- Cross-market backtesting — A-shares, HK, US, Canada, UK, India, Korea, plus crypto, futures, and forex, with composite runs and run cards.
- Shadow Account — the most interesting feature and the one nobody else has. Upload a broker export; it profiles your actual behaviour (holding days, win rate, disposition effect, overtrading, anchoring), extracts the rules you actually trade by, backtests those rules, and shows where you broke them. It audits you, not a strategy.
- Exports that leave the tool — reports, TradingView Pine Script, MetaTrader 5, TDX.
Read this part before the fun part
- This is not investment advice, and the project says so itself. It’s research software. Past performance doesn’t guarantee anything, and an agent that writes a strategy has no idea whether the edge is real.
- Live trading is experimental. Broker execution runs only through a channel you authorise and within limits you set, and you can halt it — but the maintainers state plainly that it is not verified against a real broker account. Paper-trade it.
- A backtest an agent generated is a hypothesis, not a result. Look-ahead bias, survivorship, and overfitting to a period are exactly the failures a fluent report is best at hiding. The project puts real work into guarding these — PIT data, strict out-of-sample gates, a hash-chained audit ledger over every run — which is a reason to take it seriously, not a reason to skip the scrutiny.
- There is no Vibe-Trading token. The repository carries a standing warning that an X account, a Virtuals project, and a token contract using its name are impersonations. The project has never launched or endorsed a coin. Don’t connect a wallet to anything claiming otherwise.
Where it fits
| Want | Reach for |
|---|---|
| To research an idea and see it backtested end to end | Vibe-Trading |
| To understand your own trading habits | Its Shadow Account, on your broker export |
| To hand market tools to an agent you already run | vibe-trading-mcp |
| Someone to tell you what to buy | Nothing here. That’s not what this is |
The honest framing: it’s a research workspace that removes the plumbing between a question and a tested answer. What it can’t remove is the judgement about whether the answer means anything.
Next
The same pipeline with nobody asking permission before the order → AutoHedge
Last updated 07 Sep 2026, 00:00 UTC.