How to choose AI quant tools: data, backtest, execution, then an agent
Decide the job first. Then one data source, one backtester, and paper trading.
Short answer: start with one market data library and one backtesting framework, get one simple strategy running, and paper trade before real money. Add a trading bot or an LLM agent only after that loop works. The English directory currently tracks 442 projects that have a written English summary; star counts are not a ranking of profitability.
The same order in Chinese is how to choose AI quant tools (Chinese). China A-share broker terminals are a separate article, linked in the section below.
Four kinds of tool
AI quant work uses four kinds of tools. Mixing them up is how people install a stack they cannot run.
- Market data — prices, fundamentals, or alternative data. Market data tools
- Backtesting — replay a strategy on history. Backtesting frameworks
- Execution — a bot or a broker connection that can send an order. Trading bots
- Research agents — a large language model that reads, summarizes, or drafts. LLM agents
If you are not sure which of those you need, use Pick. It asks what you want to do, how much code you write, which market you care about, and your budget, then returns three rule-based matches. It does not take payment to change the order.
An order that does not skip steps
01 · Name the job
Data, backtesting, sending orders, and LLM research are four different jobs. Pick the job before you pick a product. The tool list names a few open-source projects for each job.
02 · Pick a data library
Free libraries that read public pages are fine for learning. A paid API is for when those pages change or you need a wider history.
03 · Store a daily update
A one-off download is not a research stack. Keep daily bars in a local database and only fetch the days you are missing.
04 · Run one backtest
Use a maintained framework if you are starting a project you expect to keep. backtrader is still usable for learning, and the comparison explains what has replaced it for new work.
05 · Paper trade before live orders
A backtest that ignores fees, slippage, or a look-ahead bug will not survive a paper account. Run about two weeks on the same fill rules before any live order. Do not connect a broker or an exchange until that gap is something you can accept.
06 · Add an agent, or a portfolio library, only if you need one
An agent reads and drafts. It does not replace the backtest. Portfolio libraries are a separate step, after you already have returns to allocate.
The study order around these steps, including when a result is good enough to leave, is the beginner learning path. For portfolio weights after you have a strategy, see PyPortfolioOpt, Riskfolio-Lib, and skfolio. For an agent that talks to OKX, see OKX Agent Trade Kit: demo keys and read-only mode come before any live permission, and the key must not have withdrawal permission.
A machine that stays on
A crawler or a daily job that should run while your own computer is off needs a server. That walkthrough is a different article: the cloud quant desk. The same article in Chinese is the cloud quant desk (Chinese). It is not a language version of this page, and it is not the site FAQ. This page does not restate the machine sizes or the vendor discount.
China A-share data libraries
Tushare Pro and AKShare are the two Python libraries most often used for China A-share data. Points, the free tier, and why a free interface can fail are a different article: how to choose between Tushare and AKShare. It is not a language version of this page, and it is not the four-library comparison of yfinance, AKShare, OpenBB, and EODHD. This page does not restate those prices.
China A-share platforms
Broker terminals such as QMT and PTrade, and hosted sites such as JoinQuant, RiceQuant, and BigQuant, are a different article: where the code runs, and how it reaches your own account. That comparison is how to choose a quant trading platform. It is not a language version of this page. Account thresholds and fees are per the broker's own disclosures, and this page does not restate them.
- How to choose a quant trading platform
- The same platform comparison in Chinese
- The A-share learning path, including the filing note (Chinese)
- How to choose between QMT and PTrade
- The same QMT and PTrade guide in Chinese
- How to choose AI stock-picking tools
AI stock-picking tools
An assistant inside a market app, a general language model, AI factors on a quant platform, and an open-source research project you run yourself are a different article: how to choose AI stock-picking tools. The same article in Chinese is AI stock-picking tools (Chinese). It is not a language version of this page. It does not recommend any stock.
Trading bots
A bot that sends orders on an exchange is a different article from this order, and a different article from the China broker-terminal comparison. Freqtrade, OctoBot, Jesse, CloddsBot, and OKX's toolkit are compared in how to choose a trading bot. It is not a language version of this page. Paper trade before any real order.
Frequently asked questions
What should I install first?
One market data library and one backtesting framework. Get one simple strategy running, then paper trade before any real order. Add an LLM research agent only if you still need it. You do not need one tool from every category on day one.
Which free market data library should I start with?
For personal research on US stocks, yfinance is usually enough. AKShare is another free library, aimed at public Chinese market pages. OpenBB connects data providers you already have. EODHD APIs is a paid service for when free sources stop being enough. yfinance and AKShare read public websites and can break without notice.
Is backtrader still a reasonable first backtester?
It still installs and still runs, and it is fine for learning. Its repository has been quiet for a long time, so a new project is better on something still maintained: vectorbt for large parameter sweeps, zipline-reloaded for a classic event-driven engine, NautilusTrader when backtest and live trading must share code, or PyBroker for rule and model strategies.
When should I add an LLM trading agent?
After the data and backtest loop already works. TradingAgents is the original framework and the safest to build on, under Apache-2.0. TradingAgents-CN is a separate A-share project with a web interface; its application layer cannot be used commercially. Vibe-Trading is MIT-licensed, covers more markets, and is the quickest to install. All three are research tools.
Does a high star count mean the tool will make money?
No. Star trends and maintenance notes say whether a project is used and updated. They are not a forecast. A backtest is not a promise about future results, and a language model can misunderstand an instruction or invent a figure. Nothing on this site is investment advice.