TradingAgents vs TradingAgents-CN vs Vibe-Trading: open-source LLM trading agents compared
Published 11 Oct 2026 · repository data refreshes daily
Short answer. TradingAgents is the original framework and the safest choice for building on, thanks to its Apache-2.0 licence. TradingAgents-CN is a separate project for Chinese A-shares with a web interface and a licence that forbids commercial use of its application layer. Vibe-Trading is a newer, MIT-licensed agent that covers the most markets and is the quickest to install. All three are research tools. None is evidence that an LLM can trade profitably.
Side by side
| Tool | GitHub stars | Last activity | Licence | Markets | Runs as |
|---|---|---|---|---|---|
| TradingAgents | 110.5k | 3 Oct 2026 | Apache-2.0 | Mainly US-listed companies (SEC filings), plus crypto and macro data | Python package and CLI |
| TradingAgents-CN | 32.2k | 22 Sept 2026 | Apache-2.0 core, proprietary application layer | Chinese A-shares | Web app: FastAPI, Vue 3, MongoDB, Redis |
| Vibe-Trading | 35.1k | 10 Oct 2026 | MIT | Equities in several countries, crypto, futures, forex | PyPI install, Docker or MCP plugin |
Stars and last activity (the most recent push GitHub reports for the repository) come from our daily snapshots and can lag by a few days. Licence and scope were checked against each repository on 11 Oct 2026.
How the three relate
TradingAgents, from Tauric Research, came first. It models a trading firm as a team of LLM agents: analysts for fundamentals, news and sentiment feed researchers, a trader and a portfolio manager, who argue their way to a decision.
TradingAgents-CN describes itself as a Chinese enhanced edition and credits Tauric Research as the original project. On GitHub it is a standalone repository rather than a fork, and it has grown into its own product for the A-share market.
Vibe-Trading, from the HKUDS group, is a different code base with a different idea: one personal research agent you talk to in natural language, which can call on multi-agent teams when a task needs them.
TradingAgents: the reference framework
You run it as a Python package or from a command line. It pulls fundamentals from SEC EDGAR filings, macro series from FRED, company news, and sentiment from sources such as StockTwits and Reddit, and writes each decision to a report. Recent releases added backtesting over a grid of tickers and dates, with attention to point-in-time data so the agents only see what was published by each analysis date.
It works with many LLM providers, including any OpenAI-compatible endpoint and local models through Ollama. You bring your own API keys, and a full multi-agent run makes many model calls, so cost is something to watch. The README is explicit that the framework is for research and is not financial, investment or trading advice.
TradingAgents-CN: a web app for A-share research
This is the heaviest of the three to run. The current version is a web application built with FastAPI and Vue 3, and you install MongoDB and Redis yourself. Market and financial data come from Chinese public interfaces such as Tushare, AKShare and BaoStock, for which you register your own tokens.
In return you get features aimed at individual investors: debate-style single-stock research, natural-language stock screening, paper trading and a set of free courses. The documentation is mainly in Chinese.
Read the licence before building on it. The core engine is Apache-2.0, but the application layer is proprietary and source-available: personal, evaluation and educational use are allowed, redistribution and commercial use are not without a separate licence. There is also a paid Pro edition with batch and scheduled analysis.
Vibe-Trading: the widest reach, the fastest start
Vibe-Trading installs from PyPI with one command, or runs in Docker, or attaches to an existing agent as an MCP plugin. It needs an LLM API key from a supported provider, or a local model through Ollama.
Its market coverage is the broadest here: equities in mainland China, Hong Kong, the US and several other countries, plus crypto, futures and forex, with fallback between data sources. A feature called Shadow Account compares your broker trade journal with rule-based alternatives and exports an audit report. The project moves quickly, with changes landing almost daily, which is good for fixes and less good if you need a stable base.
One warning comes from the maintainers themselves: tokens and social accounts using the Vibe-Trading name are not official, and the project has never launched a token.
How to choose
- You want to study or extend the multi-agent idea. TradingAgents. It is the reference implementation and the licence lets you build on it.
- You research Chinese A-shares and read Chinese. TradingAgents-CN, as long as personal or educational use is all you need.
- You want an assistant running today across several markets. Vibe-Trading.
- You plan a commercial product. Start from TradingAgents or Vibe-Trading and read their licences in full.
What none of them proves
Star counts here measure interest, not returns. These projects can read filings, summarise news and argue both sides of a trade faster than a person can. Whether the resulting decisions beat a simple benchmark after costs is a separate question that each user has to test, on data the model has not seen. Treat the output as research input.
Nothing on this page is investment advice. See also this week's trending tools.