Open-source AI quant projects: trading systems and tools by job
A selection guide. AI can help you research. It cannot promise a profit.
Short answer: several AI analysts on one stock, use TradingAgents or the China A-share edition. Python and machine-learning stock selection, use Microsoft Qlib. Reinforcement learning, use FinRL. A course, start with the companion code for Machine Learning for Trading. Treat every conclusion as research. Do not trade it.
How to choose: Want several AI analysts to study one stock together: start with TradingAgents, or TradingAgents-CN if you want a web UI aimed at China A-shares. You write Python and want machine-learning stock selection: Microsoft Qlib. Reinforcement learning: FinRL. A course with exercises: the companion code for Machine Learning for Trading. US filings for a model to read: edgartools. Every project here describes itself as research. Do not send an order from its conclusion. This page is not the how-to-choose tool order, not the site FAQ, not the AI stock-picking comparison, and not the trading-bot comparison.
Who it is for: Someone picking a first open-source AI quant project: a multi-agent research desk, a machine-learning platform, a data terminal, or a course repository. What you need is the job it does, the licence, and whether the repository is still maintained, before you clone it.
Who it is not for: Someone who wants a program to place live orders, or a promise that AI quant makes money. Order-sending bots are a different article. China broker terminals such as QMT and PTrade are a different article. This page is also not the how-to-choose tool order and not the site FAQ.
The same list in Chinese is open-source AI quant projects, grouped by job. Apps and chatbots that pick stocks, and how to tell a legal advisory product from a pitch, are how to choose AI stock-picking tools. That comparison is not this article. A program that sends the order is how to choose a trading bot. That comparison is not this article either.
This site has no commercial relationship with the projects below. Stars and maintenance status come from GitHub and refresh daily. The one-line description and the licence follow each project's own notice, checked on 15 Oct 2026. Licences can change. Read the repository before you rely on a row.
01 · Multi-agent research: several AI analysts split the work
Several language models play a fundamental analyst, a news analyst, a technical analyst, and a researcher, then argue toward a conclusion. These projects call themselves research tools. Use one as a research assistant. Do not let it send orders.
| Tool | What it is | Licence | GitHub stars | Maintenance |
|---|---|---|---|---|
| TradingAgents | The project this category grew from: a multi-agent LLM framework for financial trading. It describes itself as research | Apache-2.0 | 116.2k | Last commit 11 days ago |
| TradingAgents-CN | A Chinese research assistant for China A-shares, with natural-language screens, paper trading, and a course. There is a separate paid Pro edition | Mixed: core engine Apache-2.0; application layer is source-available for personal use, evaluation, and teaching only. Commercial use and redistribution are forbidden | 32.6k | Last commit 35 days ago |
| TradingAgents-astock | An A-share multi-agent research framework adapted from TradingAgents. It adds A-share datasets such as the most-active list and share-unlock schedules. Aimed at research and teaching | Apache-2.0 | 3.8k | Last commit 2 days ago |
| Vibe-Trading | A personal trading agent with research, backtest, and portfolio tools. Coverage includes China A-shares, Hong Kong, the US, and crypto. The project says this is not investment advice | MIT | 36.8k | Commits today |
Read the licence before you clone. TradingAgents-CN is mixed: the core engine is open source, and the application layer allows personal use, evaluation, and teaching only. Commercial use and redistribution are forbidden. To stand up one research agent on a small server, see the cloud quant desk. More projects in the same family are in the LLM agents category.
02 · Machine learning and reinforcement learning: open-source AI quant platforms
These projects do not chat their way to a trade. They train a forecast or a trading agent on history. You need Python and some machine learning.
| Tool | What it is | Licence | GitHub stars | Maintenance |
|---|---|---|---|---|
| Qlib | Microsoft's open-source AI quant platform, covering data processing, model training, and backtesting. The companion project RD-Agent uses a language model to search for factors | MIT | 49.7k | Last commit 5 days ago |
| FinRL | An early open-source reinforcement-learning framework for finance. This repository keeps the original teaching and research workflow. The newer architecture is FinRL-X | MIT | 16.8k | Last commit 2 days ago |
| PyBroker | A backtesting framework aimed at machine-learning strategies | Apache 2.0 with Commons Clause | 3.7k | Commits today |
Qlib and PyBroker are also in how to choose a backtesting framework, next to engines that are not built around a model. PyBroker's licence adds a Commons Clause: you cannot sell the software itself. Qlib's companion, RD-Agent, uses a language model to search for factors. Installing Qlib, loading China A-share data, and running a first model are in the Qlib walkthrough.
03 · AI finance terminals and data: feed the model first
The write-up is only as good as the data you put in. These two projects are about where that data comes from, and how to turn it into something a model can read.
| Tool | What it is | Licence | GitHub stars | Maintenance |
|---|---|---|---|---|
| Fincept Terminal | A desktop finance research terminal written in C++, with built-in Python analysis and AI features. There is a separate paid enterprise edition | AGPL-3.0 | 33.2k | Commits today |
| edgartools | A Python library that turns US SEC filings and financial statements into structured data, with an MCP server so a language model can read them | MIT | 2.8k | Last commit 2 days ago |
Fincept Terminal is AGPL-3.0. Using it yourself is fine. If you distribute changes, or you offer those changes as a service to other people, the changes have to be open source under the same terms. How to choose a market-data library is yfinance, AKShare, OpenBB, and EODHD.
04 · Learning material: understand the method before you run the code
Start here when you want a course, with exercises, before you clone a research repo and hope the README is enough.
| Tool | What it is | Licence | GitHub stars | Maintenance |
|---|---|---|---|---|
| machine-learning-for-trading | Companion code for the third edition of Machine Learning for Trading, from data and strategy research through monitoring after a strategy is live | MIT | 21.6k | Last commit 4 days ago |
| Introduction-to-Quantitative-Finance | A Chinese-language starter library for quant research: multi-factor tutorials, a tracker of AI and finance papers, and a reading list | MIT | 1.8k | Commits today |
05 · AI quant projects that gained the most stars in seven days
Sorted by the GitHub star increase over about seven days, and refreshed with the daily crawl. A fast rise means more people looked. It does not mean the project is useful, and it does not mean it makes money. Archived repositories, sniper and MEV bots, and projects below the heat cutoff are left off.
| Project | New stars, about 7 days | GitHub stars |
|---|---|---|
| TradingAgents | +4.1k | 116.2k |
| Vibe-Trading | +828 | 36.8k |
| Fincept Terminal | +392 | 33.2k |
| LangAlpha | +314 | 2.1k |
| Qlib | +204 | 49.7k |
| TradingAgents-CN | +202 | 32.6k |
| machine-learning-for-trading | +96 | 21.6k |
| FinRL | +77 | 16.8k |
06 · What AI quant can do, and where it stops
- It can read filings and news, line up a financial statement, list the bull case and the bear case, try factors and models in bulk, and turn a backtest into a repeatable run.
- It cannot promise a profit. A language model misreads, and it invents figures that were never in the data. A model that fits history often fails on the next period.
- A kind backtest is not a live result. Fees, slippage, and overfitting all make the account worse than the backtest. Run a paper account first. The worked version is paper trading after a backtest.
- Price the model calls. A multi-agent project calls the model many times for one analysis. On a paid API, run one name and read the bill before you loop a watchlist.
If you are not writing code, and you want to know how an in-app assistant differs from asking a general model, plus how to judge whether an AI stock-picking product is credible and legal, see how to choose AI stock-picking tools.
07 · AI trading bots are a different job
The projects above are research tools. If what you want is "the AI sends the order", for example OctoBot connecting a language model to a trade, or CloddsBot operating an account from a conversation, the risk is different. That comparison is how to choose a trading bot. The CloddsBot catalog entry is the Chinese tool page. It does not have an English tool page.
08 · How to choose an open-source quant trading system
Match the project to the job you actually have. Then read the licence row before you clone.
- Try AI research first. TradingAgents-CN has a web UI, so you are not writing the analyst loop. You still install the database and connect a language-model API. For rules closer to China A-shares, look at TradingAgents-astock.
- Python, and machine-learning stock selection. Qlib. The install and the first model are in the Qlib walkthrough.
- Reinforcement-learning research. FinRL. Stay on the original teaching workflow in this repository unless you have a reason to move to FinRL-X.
- US stocks, and filings a model can read. edgartools.
- You are not sure which tools a quant stack even needs. Start with open-source tools by job.
Three checks before you spend a weekend on a clone. The licence allows the use you have in mind: personal research, a course, or a product. The maintenance column is not "archived" or "no commits in over six months" unless you are reading code, not running it. One paid model run has a price you have already seen, so a loop cannot surprise you.
Every AI-related project in the catalog can be filtered by heat in the LLM agents category and on the tool directory. Each tool page shows the last update and whether maintenance has stopped.
Frequently asked questions
How to choose an open-source AI quant project?
Want several AI analysts to study one stock together: start with TradingAgents, or TradingAgents-CN if you want a web UI aimed at China A-shares. You write Python and want machine-learning stock selection: Microsoft Qlib. Reinforcement learning: FinRL. A course with exercises: the companion code for Machine Learning for Trading. US filings for a model to read: edgartools. Every project here describes itself as research. Do not send an order from its conclusion. This page is not the how-to-choose tool order, not the site FAQ, not the AI stock-picking comparison, and not the trading-bot comparison.
Who it is for?
Someone picking a first open-source AI quant project: a multi-agent research desk, a machine-learning platform, a data terminal, or a course repository. What you need is the job it does, the licence, and whether the repository is still maintained, before you clone it.
Who it is not for?
Someone who wants a program to place live orders, or a promise that AI quant makes money. Order-sending bots are a different article. China broker terminals such as QMT and PTrade are a different article. This page is also not the how-to-choose tool order and not the site FAQ.
What open-source AI quant projects are there?
Four jobs. Multi-agent research: TradingAgents, TradingAgents-CN, TradingAgents-astock, and Vibe-Trading. Machine-learning and reinforcement-learning platforms: Microsoft Qlib, FinRL, and PyBroker. AI terminals and data: Fincept Terminal and edgartools. Learning repositories: Machine Learning for Trading, and Introduction to Quantitative Finance. Stars and maintenance status come from the daily catalog. A fast rise in stars is attention, not a result.
Can AI quant make money?
No project on this page can promise that. They describe themselves as research tools. A language model can misread a filing and can invent a figure. A machine-learning model that looks strong on history often fails on the next period. Run a paper account after the backtest. Do not send an order from the write-up.
Can TradingAgents-CN be used commercially?
The licence is mixed. The core engine is Apache-2.0. The application layer is source-available for personal use, evaluation, and teaching only. Commercial use and redistribution are forbidden. Commercial use needs the author's permission. Read the project's own LICENSE before you ship anything on top of it.
Which AI quant project should a beginner start with?
To try AI research without writing a strategy: TradingAgents-CN. It has a web UI, so you are not writing the analyst loop, but you still install the database and connect a language-model API. Python and machine-learning stock selection: Qlib. A course: the companion code for Machine Learning for Trading. Whichever you pick, read the licence and the maintenance column first.
This is not investment advice. An AI write-up can be wrong, and a backtest is not a future return. Features and licences follow each project's own notice.