档案库 · 金融科技 · 产品决策 · 2024–2026
Tauric 的 TradingAgents 开源了整个交易部门
Tauric Research 完全开源了 TradingAgents，一个模拟交易公司的多智能体系统——到3月获得39000颗星，到2026年5月达到71400颗。
Tauric Research
做的是什么生意
TradingAgents is an open-source framework that decomposes a trade decision into specialist LLM agents that mirror a real trading organization: a fundamentals analyst, sentiment analyst, news analyst and technical analyst gather evidence; bullish and bearish researchers debate it in structured rounds; a trader drafts a proposal; a risk team reviews exposure; and a portfolio manager approves, rejects or adjusts the order before it reaches a simulated exchange. Built on LangGraph, it supports a dozen-plus LLM providers including OpenAI, Anthropic, Google, DeepSeek, Qwen, GLM, Groq, Azure and Bedrock, keeps a decision log that reflects on past realized returns, and resumes interrupted runs from checkpoints. The companion Trading-R1 project applies the same team to a purpose-built financial LLM.
起因
Tauric Research is an AI research company focused on intelligent financial trading. Founder Yijia Xiao studied computer science at Tsinghua under Zhipu co-founder Tang Jie, then moved to UCLA for a PhD under co-authors Edward Sun, Di Luo and Wei Wang. The team posted TradingAgents to GitHub on 2024-12-28 with a research paper (arXiv 2412.20138) and no launch campaign; the repository read that after receiving numerous inquiries the team decided to fully open-source the framework.
经过
The quiet repo spent early 2025 mostly as a research artifact. In February 2026 v0.2.0 added multi-provider LLM support and the project began climbing; when the team announced the full open-source release in March 2026 it went viral, and 36kr reported 39,000+ stars and 7,200+ forks by month's end. By early May 2026 TradingAgents held #1 on GitHub's Python trending chart with 71,400+ stars and 13,800+ forks, gaining over 11,000 stars in one week, and TMTPost framed it as proof that vertical workflows were displacing general orchestration frameworks in developer attention. Releases through spring added decision memory, checkpoint resume, structured outputs and a grounded sentiment analyst; v0.4.0 in August 2026 added look-ahead fixes and a decision-log memory.
结果
As of 2026-09-05 TradingAgents is live and still growing — TechTarget's weekly list for early September again included it in the top 20 fastest-gaining repos — with v0.4.0 released in August. It remains a research framework with an explicit disclaimer that it is not investment advice and that returns vary with model, data and settings; no revenue or funding figures for Tauric Research have been published. Whether the viral open-source workflow converts into a commercial product, and whether its decisions would hold up under real capital, are still open.
背景
TradingAgents是Tauric Research（一家专注于智能交易的AI研究公司）的一个开源多智能体金融交易框架。它将交易公司编码为软件：基本面分析师、情绪分析师、新闻分析师和技术分析师并行收集证据；看涨和看跌研究员进行辩论；交易员提出交易；风险团队审查敞口；投资组合经理在模拟交易所执行前批准或拒绝。
该项目于2024年12月28日悄悄发布在GitHub上，附有创始人Yijia Xiao和UCLA合著者的arXiv论文，没有发布活动。团队在2026年3月宣布完全开源后迅速走红：36kr在3月底统计了39000多颗星和7200多个复刻，TMTPost在5月初统计了71400多颗星和13800多个复刻，当时仓库在GitHub的Python趋势榜上排名第一。
赌注是完整的垂直工作流优于通用智能体编排：用户提供股票代码、日期和一个LLM API密钥，框架处理分析、多空辩论、风险审查和决策日志。v0.2.0（2026年2月）增加的多提供商支持被认为加速了增长，2026年的版本增加了检查点恢复、结构化输出和反思过去回报的决策记忆。
截至2026年9月5日，TradingAgents仍然活跃且仅限研究，v0.4.0于8月发布，并在9月初进入GitHub每周前20。Tauric Research未公布收入或资金数据，框架明确声明不构成投资建议——社区成功能否成为业务仍是未知数。
这件事要成立，得有什么
- 当单个模型同时担任分析师、魔鬼代言人、交易员和风险管理师时，会产生黑盒；分离的智能体通过结构化辩论让每一步都有可审计的负责人。
- 垂直、经过验证的工作流回答了“我今天运行什么”而不是“我如何编排我的智能体”，这就是TradingAgents在Python趋势榜上取代通用框架的原因。
- 完全开源是一种刻意的信任策略：金融开发者不会将封闭系统接入他们的研究，但他们会给透明的、可检查和扩展的仓库加星和复刻。
- 仅限研究的免责声明设定了诚实的期望——该项目的价值是作为研究多智能体分析的支架，而不是声称它能印钞。
可借鉴之处
开发者支持的是完整可审计的工作流，而不是一个巧妙的模型——有价值的赌注是把公司的整个决策过程打包成可运行的开源。
后续进展
TradingAgents从安静的研究仓库变成2026年GitHub增长最快的项目之一：3月底39000多颗星，5月初71400多颗星，并在v0.4.0后的9月初进入每周前20。Tauric Research仍然很小——三个公开仓库——除了框架和Trading-R1研究线外，没有披露收入或资金。社区采用能否转化为付费用户，以及一个明确不打算作为投资建议的研究工具能否保持受众，仍是未解之谜。
资料来源
- 71.4K Star的AI交易团队：多智能体架构如何“炒”出一个华尔街
- 唐杰高徒打造龙虾投资军团，量化私募全线Agent，开源狂揽39k星
- What repos are trending on GitHub?
发现哪里写错了？告诉我们。
轮到你了
你刚读完一家。说说你在做什么，看看谁在赌同一件事。
免费账号 · 3 次免费提问 · 不用绑卡